ns-3 Manual (html version)

For a pdf version of this manual, see http://www.nsnam.org/docs/manual.pdf.

Simulator version: ns-3.8

This is an ns-3 reference manual. Primary documentation for the ns-3 project is available in five forms:

• ns-3 Tutorial
• ns-3 Doxygen: Documentation of the public APIs of the simulator
• Reference Manual (this document)
• ns-3 Testing and Validation manual
• ns-3 wiki

This document is written in GNU Texinfo and is to be maintained in revision control on the ns-3 code server. Both PDF and HTML versions should be available on the server. Changes to the document should be discussed on the ns-developers@isi.edu mailing list.

This software is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.

This software is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.

1. Organization

This manual is organized into several parts with several chapters per part. This chapter describes the overall software organization and the corresponding organization of this manual.

ns-3 is a discrete-event network simulator in which the simulation core and models are implemented in C++. ns-3 is built as a library which may be statically or dynamically linked to a C++ main program that defines the simulation topology and starts the simulator. ns-3 also exports nearly all of its API to Python, allowing Python programs to import an "ns3" module in much the same way as in C++.

Figure 1.1: Software organization of ns-3

The source code for ns-3 is mostly organized in the src/ directory and can be described by the diagram in fig:organization. We will work our way from the bottom up; in general, modules only have dependencies on modules beneath them in the figure.

We first describe Part 1 of the manual. The simulation core is implemented in src/core, and the core is used to build the simulation engine src/simulator. Packets are fundamental objects in a network simulator and are implemented in src/packet. These three simulation modules by themselves are intended to comprise a generic simulation core that can be used by different kinds of networks, not just Internet-based networks. The above modules of ns-3 are independent of specific network and device models, which are covered in later parts of this manual.

In addition to the above ns-3 core, we describe also in Part 1 two other modules that supplement the core C++-based API. ns-3 programs may access all of the API directly or may make use of a so-called “helper API” that provides convenient wrappers or encapsulation of low-level API calls. The fact that ns-3 programs can be written to two APIs (or a combination thereof) is a fundamental aspect of the simulator and is also covered in Part 1. We also describe how Python is supported in ns-3 as the last chapter of Part 1.

The remainder of the manual is focused on documenting the models and supporting capabilities. Part 2 focuses on two fundamental objects in ns-3: the Node and NetDevice. Two special NetDevice types are designed to support network emulation use cases, and emulation is described in Part 3. Part 4 is devoted to Internet-related models, including the sockets API used by Internet applications. Part 5 covers applications, and Part 6 describes additional support for simulation, such as animators.

The project maintains a separate manual devoted to testing and validation of ns-3 code (see the ns-3 Testing and Validation manual).

Part 1: ns-3 core

2. Random variables

ns-3 contains a built-in pseudo-random number generator (PRNG). It is important for serious users of the simulator to understand the functionality, configuration, and usage of this PRNG, and to decide whether it is sufficient for his or her research use.

2.1 Quick Overview

ns-3 random numbers are provided via instances of class RandomVariable.

• by default, ns-3 simulations use a fixed seed; if there is any randomness in the simulation, each run of the program will yield identical results unless the seed and/or run number is changed.
  • in ns-3.3 and earlier, ns-3 simulations used a random seed by default; this marks a change in policy starting with ns-3.4
• to obtain randomness across multiple simulation runs, you must either set the seed differently or set the run number differently. To set a seed, call SeedManager::SetSeed (uint32_t) at the beginning of the program; to set a run number with the same seed, call SeedManager::SetRun (uint32_t) at the beginning of the program; See section Seeding and independent replications
• each RandomVariable used in ns-3 has a virtual random number generator associated with it; all random variables use either a fixed or random seed based on the use of the global seed (previous bullet);
• if you intend to perform multiple runs of the same scenario, with different random numbers, please be sure to read the section on how to perform independent replications: See section Seeding and independent replications.

Read further for more explanation about the random number facility for ns-3.

2.2 Background

Simulations use a lot of random numbers; one study found that most network simulations spend as much as 50% of the CPU generating random numbers. Simulation users need to be concerned with the quality of the (pseudo) random numbers and the independence between different streams of random numbers.

Users need to be concerned with a few issues, such as:

• the seeding of the random number generator and whether a simulation outcome is deterministic or not,
• how to acquire different streams of random numbers that are independent from one another, and
• how long it takes for streams to cycle

We will introduce a few terms here: a RNG provides a long sequence of (pseudo) random numbers. The length of this sequence is called the cycle length or period, after which the RNG will repeat itself. This sequence can be partitioned into disjoint streams. A stream of a RNG is a contiguous subset or block of the RNG sequence. For instance, if the RNG period is of length N, and two streams are provided from this RNG, then the first stream might use the first N/2 values and the second stream might produce the second N/2 values. An important property here is that the two streams are uncorrelated. Likewise, each stream can be partitioned disjointedly to a number of uncorrelated substreams. The underlying RNG hopefully produces a pseudo-random sequence of numbers with a very long cycle length, and partitions this into streams and substreams in an efficient manner.

ns-3 uses the same underlying random number generator as does ns-2: the MRG32k3a generator from Pierre L’Ecuyer. A detailed description can be found in http://www.iro.umontreal.ca/~lecuyer/myftp/papers/streams00.pdf. The MRG32k3a generator provides 1.8x10^19 independent streams of random numbers, each of which consists of 2.3x10^15 substreams. Each substream has a period (i.e., the number of random numbers before overlap) of 7.6x10^22. The period of the entire generator is 3.1x10^57.

Class ns3::RandomVariable is the public interface to this underlying random number generator. When users create new RandomVariables (such as UniformVariable, ExponentialVariable, etc.), they create an object that uses one of the distinct, independent streams of the random number generator. Therefore, each object of type RandomVariable has, conceptually, its own "virtual" RNG. Furthermore, each RandomVariable can be configured to use one of the set of substreams drawn from the main stream.

An alternate implementation would be to allow each RandomVariable to have its own (differently seeded) RNG. However, we cannot guarantee as strongly that the different sequences would be uncorrelated in such a case; hence, we prefer to use a single RNG and streams and substreams from it.

2.3 Seeding and independent replications

ns-3 simulations can be configured to produce deterministic or random results. If the ns-3 simulation is configured to use a fixed, deterministic seed with the same run number, it should give the same output each time it is run.

By default, ns-3 simulations use a fixed seed and run number. These values are stored in two ns3::GlobalValue instances: g_rngSeed and g_rngRun.

A typical use case is to run a simulation as a sequence of independent trials, so as to compute statistics on a large number of independent runs. The user can either change the global seed and rerun the simulation, or can advance the substream state of the RNG, which is referred to as incrementing the run number.

A class ns3::SeedManager () provides an API to control the seeding and run number behavior. This seeding and substream state setting must be called before any random variables are created; e.g.

  SeedManager::SetSeed (3);  // Changes seed from default of 1 to 3
  SeedManager::SetRun (7);  // Changes run number from default of 1 to 7
  // Now, create random variables
  UniformVariable x(0,10);
  ExponentialVariable y(2902);
  ...

Which is better, setting a new seed or advancing the substream state? There is no guarantee that the streams produced by two random seeds will not overlap. The only way to guarantee that two streams do not overlap is to use the substream capability provided by the RNG implementation. Therefore, use the substream capability to produce multiple independent runs of the same simulation. In other words, the more statistically rigorous way to configure multiple independent replications is to use a fixed seed and to advance the run number. This implementation allows for a maximum of 2.3x10^15 independent replications using the substreams.

For ease of use, it is not necessary to control the seed and run number from within the program; the user can set the NS_GLOBAL_VALUE environment variable as follows:

  NS_GLOBAL_VALUE="RngRun=3" ./waf --run program-name

Another way to control this is by passing a command-line argument; since this is an ns-3 GlobalValue instance, it is equivalently done such as follows:

  ./waf --command-template="%s --RngRun=3" --run program-name

or, if you are running programs directly outside of waf:

  ./build/optimized/scratch/program-name --RngRun=3

The above command-line variants make it easy to run lots of different runs from a shell script by just passing a different RngRun index.

2.4 class RandomVariable

All random variables should derive from class RandomVariable. This base class provides a few static methods for globally configuring the behavior of the random number generator. Derived classes provide API for drawing random variates from the particular distribution being supported.

Each RandomVariable created in the simulation is given a generator that is a new RNGStream from the underlying PRNG. Used in this manner, the L’Ecuyer implementation allows for a maximum of 1.8x10^19 random variables. Each random variable in a single replication can produce up to 7.6x10^22 random numbers before overlapping.

2.5 Base class public API

Below are excerpted a few public methods of class RandomVariable that access the next value in the substream.

  /**
   * \brief Returns a random double from the underlying distribution
   * \return A floating point random value
   */
  double GetValue (void) const;
  
  /**
   * \brief Returns a random integer integer from the underlying distribution
   * \return  Integer cast of ::GetValue()
   */
  uint32_t GetInteger (void) const;

We have already described the seeding configuration above. Different RandomVariable subclasses may have additional API.

2.6 Types of RandomVariables

The following types of random variables are provided, and are documented in the ns-3 Doxygen or by reading src/core/random-variable.h. Users can also create their own custom random variables by deriving from class RandomVariable.

• class UniformVariable
• class ConstantVariable
• class SequentialVariable
• class ExponentialVariable
• class ParetoVariable
• class WeibullVariable
• class NormalVariable
• class EmpiricalVariable
• class IntEmpiricalVariable
• class DeterministicVariable
• class LogNormalVariable
• class TriangularVariable
• class GammaVariable
• class ErlangVariable
• class ZipfVariable

2.7 Semantics of RandomVariable objects

RandomVariable objects have value semantics. This means that they can be passed by value to functions. The can also be passed by reference to const. RandomVariables do not derive from ns3::Object and we do not use smart pointers to manage them; they are either allocated on the stack or else users explicitly manage any heap-allocated RandomVariables.

RandomVariable objects can also be used in ns-3 attributes, which means that values can be set for them through the ns-3 attribute system. An example is in the propagation models for WifiNetDevice:

TypeId
RandomPropagationDelayModel::GetTypeId (void)
{ 
  static TypeId tid = TypeId ("ns3::RandomPropagationDelayModel")
    .SetParent<PropagationDelayModel> ()
    .AddConstructor<RandomPropagationDelayModel> ()
    .AddAttribute ("Variable",
                   "The random variable which generates random delays (s).",
                   RandomVariableValue (UniformVariable (0.0, 1.0)),
         MakeRandomVariableAccessor (&RandomPropagationDelayModel::m_variable), 
                   MakeRandomVariableChecker ())
    ;
  return tid;

Here, the ns-3 user can change the default random variable for this delay model (which is a UniformVariable ranging from 0 to 1) through the attribute system.

2.8 Using other PRNG

There is presently no support for substituting a different underlying random number generator (e.g., the GNU Scientific Library or the Akaroa package). Patches are welcome.

2.9 More advanced usage

To be completed

2.10 Publishing your results

When you publish simulation results, a key piece of configuration information that you should always state is how you used the the random number generator.

• what seeds you used,
• what RNG you used if not the default,
• how were independent runs performed,
• for large simulations, how did you check that you did not cycle.

It is incumbent on the researcher publishing results to include enough information to allow others to reproduce his or her results. It is also incumbent on the researcher to convince oneself that the random numbers used were statistically valid, and to state in the paper why such confidence is assumed.

2.11 Summary

Let’s review what things you should do when creating a simulation.

• Decide whether you are running with a fixed seed or random seed; a fixed seed is the default,
• Decide how you are going to manage independent replications, if applicable,
• Convince yourself that you are not drawing more random values than the cycle length, if you are running a very long simulation, and
• When you publish, follow the guidelines above about documenting your use of the random number generator.

3. Callbacks

Some new users to ns-3 are unfamiliar with an extensively used programming idiom used throughout the code: the “ns-3 callback”. This chapter provides some motivation on the callback, guidance on how to use it, and details on its implementation.

3.1 Motivation

Consider that you have two simulation models A and B, and you wish to have them pass information between them during the simulation. One way that you can do that is that you can make A and B each explicitly knowledgeable about the other, so that they can invoke methods on each other.

class A {
public:
  void ReceiveInput ( // parameters );
  ...
}

(in another source file:)

class B {
public:
  void DoSomething (void);
  ...

private:
  A* a_instance; // pointer to an A
}

void
B::DoSomething()
{
  // Tell a_instance that something happened
  a_instance->ReceiveInput ( // parameters);
  ...
}

This certainly works, but it has the drawback that it introduces a dependency on A and B to know about the other at compile time (this makes it harder to have independent compilation units in the simulator) and is not generalized; if in a later usage scenario, B needs to talk to a completely different C object, the source code for B needs to be changed to add a “c_instance” and so forth. It is easy to see that this is a brute force mechanism of communication that can lead to programming cruft in the models.

This is not to say that objects should not know about one another if there is a hard dependency between them, but that often the model can be made more flexible if its interactions are less constrained at compile time.

This is not an abstract problem for network simulation research, but rather it has been a source of problems in previous simulators, when researchers want to extend or modify the system to do different things (as they are apt to do in research). Consider, for example, a user who wants to add an IPsec security protocol sublayer between TCP and IP:

------------                   -----------
|   TCP    |                   |  TCP    |
------------                   -----------
     |           becomes ->        |
-----------                    -----------
|   IP    |                    | IPsec   |
-----------                    -----------
                                   |
                               -----------
                               |   IP    |
                               -----------

If the simulator has made assumptions, and hard coded into the code, that IP always talks to a transport protocol above, the user may be forced to hack the system to get the desired interconnections, This is clearly not an optimal way to design a generic simulator.

3.2 Background

The basic mechanism that allows one to address the problem above is known as a callback. The ultimate goal is to allow one piece of code to call a function (or method in C++) without any specific inter-module dependency.

This ultimately means you need some kind of indirection – you treat the address of the called function as a variable. This variable is called a pointer-to-function variable. The relationship between function and pointer-to-function pointer is really no different that that of object and pointer-to-object.

In C the canonical example of a pointer-to-function is a pointer-to-function-returning-integer (PFI). For a PFI taking one int parameter, this could be declared like,

  int (*pfi)(int arg) = 0;

What you get from this is a variable named simply “pfi” that is initialized to the value 0. If you want to initialize this pointer to something meaningful, you have to have a function with a matching signature. In this case,

  int MyFunction (int arg) {}

If you have this target, you can initialize the variable to point to your function like,

  pfi = MyFunction;

You can then call MyFunction indirectly using the more suggestive form of the call,

  int result = (*pfi) (1234);

This is suggestive since it looks like you are dereferencing the function pointer just like you would dereference any pointer. Typically, however, people take advantage of the fact that the compiler knows what is going on and will just use a shorter form,

  int result = pfi (1234);

Notice that the function pointer obeys value semantics, so you can pass it around like any other value. Typically, when you use an asynchronous interface you will pass some entity like this to a function which will perform an action and “call back” to let you know it completed. It calls back by following the indirection and executing the provided function.

In C++ you have the added complexity of objects. The analogy with the PFI above means you have a pointer to a member function returning an int (PMI) instead of the pointer to function returning an int (PFI).

The declaration of the variable providing the indirection looks only slightly different,

  int (MyClass::*pmi) (int arg) = 0;

This declares a variable named “pmi” just as the previous example declared a variable named “pfi.” Since the will be to call a method of an instance of a particular class, one must declare that method in a class.

  class MyClass {
  public:
    int MyMethod (int arg);
  };

Given this class declaration, one would then initialize that variable like this,

  pmi = &MyClass::MyMethod;

This assigns the address of the code implementing the method to the variable, completing the indirection. In order to call a method, the code needs a “this” pointer. This, in turn, means there must be an object of MyClass to refer to. A simplistic example of this is just calling a method indirectly (think virtual function).

  int (MyClass::*pmi) (int arg) = 0;  // Declare a PMI
  pmi = &MyClass::MyMethod;           // Point at the implementation code

  MyClass myClass;                    // Need an instance of the class
  (myClass.*pmi) (1234);              // Call the method with an object ptr

Just like in the C example, you can use this in an asynchronous call to another module which will “call back” using a method and an object pointer. The straightforward extension one might consider is to pass a pointer to the object and the PMI variable. The module would just do,

  (*objectPtr.*pmi) (1234);

to execute the callback on the desired object.

One might ask at this time, “what’s the point”? The called module will have to understand the concrete type of the calling object in order to properly make the callback. Why not just accept this, pass the correctly typed object pointer and do object->Method(1234) in the code instead of the callback? This is precisely the problem described above. What is needed is a way to decouple the calling function from the called class completely. This requirement led to the development of the Functor.

A functor is the outgrowth of something invented in the 1960s called a closure. It is basically just a packaged-up function call, possibly with some state.

A functor has two parts, a specific part and a generic part, related through inheritance. The calling code (the code that executes the callback) will execute a generic overloaded operator () of a generic functor to cause the callback to be called. The called code (the code that wants to be called back) will have to provide a specialized implementation of the operator () that performs the class-specific work that caused the close-coupling problem above.

With the specific functor and its overloaded operator () created, the called code then gives the specialized code to the module that will execute the callback (the calling code).

The calling code will take a generic functor as a parameter, so an implicit cast is done in the function call to convert the specific functor to a generic functor. This means that the calling module just needs to understand the generic functor type. It is decoupled from the calling code completely.

The information one needs to make a specific functor is the object pointer and the pointer-to-method address.

The essence of what needs to happen is that the system declares a generic part of the functor,

  template <typename T>
  class Functor
  {
  public:
      virtual void operator() (T arg) = 0;
   };

The caller defines a specific part of the functor that really is just there to implement the specific operator() method,

  template <typename T, typename ARG>
  class SpecificFunctor : public Functor
   {
   public:
      SpecificFunctor(T* p, int (T::*_pmi)(ARG arg))
      {
        m_p = p;
        m_pmi = pmi;
      }

      virtual int operator() (ARG arg)
      {
        (*m_p.*m_pmi)(arg);
      }
  private:
      void (T::*m_pmi)(ARG arg);
      T* m_p;
   };

N.B. The previous code is not real ns-3 code. It is simplistic example code used only to illustrate the concepts involved and to help you understand the system more. Do not expect to find this code anywhere in the ns-3 tree

Notice that there are two variables defined in the class above. The m_p variable is the object pointer and m_pmi is the variable containing the address of the function to execute.

Notice that when operator() is called, it in turn calls the method provided with the object pointer using the C++ PMI syntax.

To use this, one could then declare some model code that takes a generic functor as a parameter

  void LibraryFunction (Functor functor);

The code that will talk to the model would build a specific functor and pass it to LibraryFunction,

  MyClass myClass;
  SpecificFunctor<MyClass, int> functor (&myclass, MyClass::MyMethod);

When LibraryFunction is done, it executes the callback using the operator() on the generic functor it was passed, and in this particular case, provides the integer argument:

  void 
  LibraryFunction (Functor functor)
  {
    // Execute the library function
    functor(1234);
  }

Notice that LibraryFunction is completely decoupled from the specific type of the client. The connection is made through the Functor polymorphism.

The Callback API in ns-3 implements object-oriented callbacks using the functor mechanism. This callback API, being based on C++ templates, is type-safe; that is, it performs static type checks to enforce proper signature compatibility between callers and callees. It is therefore more type-safe to use than traditional function pointers, but the syntax may look imposing at first. This section is designed to walk you through the Callback system so that you can be comfortable using it in ns-3.

3.3 Using the Callback API

The Callback API is fairly minimal, providing only two services:

• callback type declaration: a way to declare a type of callback with a given signature, and,
• callback instantiation: a way to instantiate a template-generated forwarding callback which can forward any calls to another C++ class member method or C++ function.

This is best observed via walking through an example, based on samples/main-callback.cc.

3.3.1 Using the Callback API with static functions

Consider a function:

  static double
  CbOne (double a, double b)
  {
    std::cout << "invoke cbOne a=" << a << ", b=" << b << std::endl;
    return a;
  }

Consider also the following main program snippet:

  int main (int argc, char *argv[])
  {
    // return type: double
    // first arg type: double
    // second arg type: double
    Callback<double, double, double> one;
  }

This is an example of a C-style callback – one which does not include or need a this pointer. The function template Callback is essentially the declaration of the variable containing the pointer-to-function. In the example above, we explicitly showed a pointer to a function that returned an integer and took a single integer as a parameter, The Callback template function is a generic version of that – it is used to declare the type of a callback.

Note1: Readers unfamiliar with C++ templates may consult this reference.

The Callback template requires one mandatory argument (the return type of the function to be assigned to this callback) and up to five optional arguments, which each specify the type of the arguments (if your particular callback function has more than five arguments, then this can be handled by extending the callback implementation).

So in the above example, we have a declared a callback named "one" that will eventually hold a function pointer. The signature of the function that it will hold must return double and must support two double arguments. If one tries to pass a function whose signature does not match the declared callback, the compilation will fail.

Now, we need to tie together this callback instance and the actual target function (CbOne). Notice above that CbOne has the same function signature types as the callback– this is important. We can pass in any such properly-typed function to this callback. Let’s look at this more closely:

static double CbOne (double a, double b) {}
          ^           ^          ^
          |        ---|    ------|
          |        |       | 
Callback<double, double, double> one;

You can only bind a function to a callback if they have the matching signature. The first template argument is the return type, and the additional template arguments are the types of the arguments of the function signature.

Now, let’s bind our callback "one" to the function that matches its signature:

  // build callback instance which points to cbOne function
  one = MakeCallback (&CbOne);

This call to MakeCallback is, in essence, creating one of the specialized functors mentioned above. The variable declared using the Callback template function is going to be playing the part of the generic functor. The assignment one = MakeCallback (&CbOne) is the cast that converts the specialized functor known to the callee to a generic functor known to the caller.

Then, later in the program, if the callback is needed, it can be used as follows:

  NS_ASSERT (!one.IsNull ());

  // invoke cbOne function through callback instance
  double retOne;
  retOne = one (10.0, 20.0);

The check for IsNull() ensures that the callback is not null – that there is a function to call behind this callback. Then, one() executes the generic operator() which is really overloaded with a specific implementation of operator() and returns the same result as if CbOne() had been called directly.

3.3.2 Using the Callback API with member functions

Generally, you will not be calling static functions but instead public member functions of an object. In this case, an extra argument is needed to the MakeCallback function, to tell the system on which object the function should be invoked. Consider this example, also from main-callback.cc:

  class MyCb {
  public:
    int CbTwo (double a) {
        std::cout << "invoke cbTwo a=" << a << std::endl;
        return -5;
    }
  };

  int main ()
  {
    ...
    // return type: int
    // first arg type: double
    Callback<int, double> two;
    MyCb cb;
    // build callback instance which points to MyCb::cbTwo
    two = MakeCallback (&MyCb::CbTwo, &cb);
    ...
  }

Here, we pass an additional object pointer to the MakeCallback<> function. Recall from the background section above that Operator() will use the pointer to member syntax when it executes on an object:

      virtual int operator() (ARG arg)
      {
        (*m_p.*m_pmi)(arg);
      }

And so we needed to provide the two variables (m_p and m_pmi) when we made the specific functor. The line,

    two = MakeCallback (&MyCb::CbTwo, &cb);

does precisely that. In this case,

When two () is invoked,

  int result = two (1.0);

I will result in a call the CbTwo member function (method) on the object pointed to by &cb.

3.3.3 Building Null Callbacks

It is possible for callbacks to be null; hence it may be wise to check before using them. There is a special construct for a null callback, which is preferable to simply passing "0" as an argument; it is the MakeNullCallback<> construct:

  two = MakeNullCallback<int, double> ();
  NS_ASSERT (two.IsNull ());

Invoking a null callback is just like invoking a null function pointer: it will crash at runtime.

3.4 Bound Callbacks

A very useful extension to the functor concept is that of a Bound Callback. Previously it was mentioned that closures were originally function calls packaged up for later execution. Notice that in all of the Callback descriptions above, there is no way to package up any parameters for use later – when the Callback is called via operator(). All of the parameters are provided by the calling function.

What if it is desired to allow the client function (the one that provides the callback) to provide some of the parameters? Alexandrescu calls the process of allowing a client to specify one of the parameters binding. One of the parameters of operator() has been bound (fixed) by the client.

Some of our pcap tracing code provides a nice example of this. There is a function that needs to be called whenever a packet is received. This function calls an object that actually writes the packet to disk in the pcap file format. The signature of one of these functions will be,

  static void SniffEvent (Ptr<PcapWriter> writer, Ptr<const Packet> packet);

The static keyword means this is a static function which does not need a this pointer, so it will be using C-style callbacks. We don’t want the calling code to have to know about anything but the Packet. What we want there is just a call that looks like,

  m_promiscSnifferTrace (m_currentPkt);

What we want to do is to bind the Ptr<PcapWriter> writer to the specific callback implementation when it is created and arrange for the operator() of the Callback to provide that parameter for free.

We provide the MakeBoundCallback template function for that purpose. It takes the same parameters as the MakeCallback template function but also takes the parameters to be bound. In the case of the example above,

    MakeBoundCallback (&CsmaHelper::SniffEvent, pcap));

will create a specific callback implementation that knows to add in the extra bound arguments. Conceptually, it extends the specific functor described above with one or more bound arguments

  template <typename T, typename ARG, typename BOUND_ARG>
  class SpecificFunctor : public Functor
   {
   public:
      SpecificFunctor(T* p, int (T::*_pmi)(ARG arg), BOUND_ARG boundArg)
      {
        m_p = p;
        m_pmi = pmi;
        m_boundArg = boundArg;
      }

      virtual int operator() (ARG arg)
      {
        (*m_p.*m_pmi)(m_boundArg, arg);
      }
  private:
      void (T::*m_pmi)(ARG arg);
      T* m_p;
      BOUND_ARG m_boundArg;
   };

You can see that when the specific functor is created, the bound argument is saved in the functor / callback object itself. When the operator() is invoked with the single parameter, as in

  m_promiscSnifferTrace (m_currentPkt);

the implementation of operator() adds the bound parameter into the actual function call:

  (*m_p.*m_pmi)(m_boundArg, arg);

3.5 Traced Callbacks

3.6 Callback locations in ns-3

3.7 Callback locations in ns-3

Where are callbacks frequently used in ns-3? Here are some of the more visible ones to typical users:

• Socket API
• Layer-2/Layer-3 API
• Tracing subsystem
• API between IP and routing subsystems

3.8 Implementation details

The code snippets above are simplistic and only designed to illustrate the mechanism itself. The actual Callback code is quite complicated and very template-intense and a deep understanding of the code is not required. If interested, expert users may find the following useful:

The code was originally written based on the techniques described in http://www.codeproject.com/cpp/TTLFunction.asp. It was subsequently rewritten to follow the architecture outlined in Modern C++ Design: Generic Programming and Design Patterns Applied– Alexandrescu, chapter 5, "Generalized Functors".

This code uses:

• default template parameters to saves users from having to specify empty parameters when the number of parameters is smaller than the maximum supported number
• the pimpl idiom: the Callback class is passed around by value and delegates the crux of the work to its pimpl pointer.
• two pimpl implementations which derive from CallbackImpl FunctorCallbackImpl can be used with any functor-type while MemPtrCallbackImpl can be used with pointers to member functions.
• a reference list implementation to implement the Callback’s value semantics.

This code most notably departs from the Alexandrescu implementation in that it does not use type lists to specify and pass around the types of the callback arguments. Of course, it also does not use copy-destruction semantics and relies on a reference list rather than autoPtr to hold the pointer.

4. Object model

ns-3 is fundamentally a C++ object system. Objects can be declared and instantiated as usual, per C++ rules. ns-3 also adds some features to traditional C++ objects, as described below, to provide greater functionality and features. This manual chapter is intended to introduce the reader to the ns-3 object model.

This section describes the C++ class design for ns-3 objects. In brief, several design patterns in use include classic object-oriented design (polymorphic interfaces and implementations), separation of interface and implementation, the non-virtual public interface design pattern, an object aggregation facility, and reference counting for memory management. Those familiar with component models such as COM or Bonobo will recognize elements of the design in the ns-3 object aggregation model, although the ns-3 design is not strictly in accordance with either.

4.1 Object-oriented behavior

C++ objects, in general, provide common object-oriented capabilities (abstraction, encapsulation, inheritance, and polymorphism) that are part of classic object-oriented design. ns-3 objects make use of these properties; for instance:

class Address
{
public:
  Address ();
  Address (uint8_t type, const uint8_t *buffer, uint8_t len);
  Address (const Address & address);
  Address &operator = (const Address &address);
  ...
private:
  uint8_t m_type;
  uint8_t m_len;
  ...
};

4.2 Object base classes

There are three special base classes used in ns-3. Classes that inherit from these base classes can instantiate objects with special properties. These base classes are:

• class Object
• class ObjectBase
• class SimpleRefCount

It is not required that ns-3 objects inherit from these class, but those that do get special properties. Classes deriving from class Object get the following properties.

• the ns-3 type and attribute system (see Attributes)
• an object aggregation system
• a smart-pointer reference counting system (class Ptr)

Classes that derive from class ObjectBase get the first two properties above, but do not get smart pointers. Classes that derive from class SimpleRefCount get only the smart-pointer reference counting system.

In practice, class Object is the variant of the three above that the ns-3 developer will most commonly encounter.

4.3 Memory management and class Ptr

Memory management in a C++ program is a complex process, and is often done incorrectly or inconsistently. We have settled on a reference counting design described as follows.

All objects using reference counting maintain an internal reference count to determine when an object can safely delete itself. Each time that a pointer is obtained to an interface, the object’s reference count is incremented by calling Ref(). It is the obligation of the user of the pointer to explicitly Unref() the pointer when done. When the reference count falls to zero, the object is deleted.

• When the client code obtains a pointer from the object itself through object creation, or via GetObject, it does not have to increment the reference count.
• When client code obtains a pointer from another source (e.g., copying a pointer) it must call Ref() to increment the reference count.
• All users of the object pointer must call Unref() to release the reference.

The burden for calling Unref() is somewhat relieved by the use of the reference counting smart pointer class described below.

Users using a low-level API who wish to explicitly allocate non-reference-counted objects on the heap, using operator new, are responsible for deleting such objects.

4.3.1 Reference counting smart pointer (Ptr)

Calling Ref() and Unref() all the time would be cumbersome, so ns-3 provides a smart pointer class Ptr similar to Boost::intrusive_ptr. This smart-pointer class assumes that the underlying type provides a pair of Ref and Unref methods that are expected to increment and decrement the internal refcount of the object instance.

This implementation allows you to manipulate the smart pointer as if it was a normal pointer: you can compare it with zero, compare it against other pointers, assign zero to it, etc.

It is possible to extract the raw pointer from this smart pointer with the GetPointer and PeekPointer methods.

If you want to store a newed object into a smart pointer, we recommend you to use the CreateObject template functions to create the object and store it in a smart pointer to avoid memory leaks. These functions are really small convenience functions and their goal is just to save you a small bit of typing.

4.3.2 CreateObject and Create

Objects in C++ may be statically, dynamically, or automatically created. This holds true for ns-3 also, but some objects in the system have some additional frameworks available. Specifically, reference counted objects are usually allocated using a templated Create or CreateObject method, as follows.

For objects deriving from class Object:

 Ptr<WifiNetDevice> device = CreateObject<WifiNetDevice> ();

Please do not create such objects using operator new; create them using CreateObject() instead.

For objects deriving from class SimpleRefCount, or other objects that support usage of the smart pointer class, a templated helper function is available and recommended to be used:

  Ptr<B> b = Create<B> ();

This is simply a wrapper around operator new that correctly handles the reference counting system.

In summary, use Create<B> if B is not an object but just uses reference counting (e.g. class Packet), and use CreateObject<B> if B derives from ns3::Object.

4.3.3 Aggregation

The ns-3 object aggregation system is motivated in strong part by a recognition that a common use case for ns-2 has been the use of inheritance and polymorphism to extend protocol models. For instance, specialized versions of TCP such as RenoTcpAgent derive from (and override functions from) class TcpAgent.

However, two problems that have arisen in the ns-2 model are downcasts and “weak base class.” Downcasting refers to the procedure of using a base class pointer to an object and querying it at run time to find out type information, used to explicitly cast the pointer to a subclass pointer so that the subclass API can be used. Weak base class refers to the problems that arise when a class cannot be effectively reused (derived from) because it lacks necessary functionality, leading the developer to have to modify the base class and causing proliferation of base class API calls, some of which may not be semantically correct for all subclasses.

ns-3 is using a version of the query interface design pattern to avoid these problems. This design is based on elements of the Component Object Model and GNOME Bonobo although full binary-level compatibility of replaceable components is not supported and we have tried to simplify the syntax and impact on model developers.

4.3.3.1 Aggregation example

class Node is a good example of the use of aggregation in ns-3. Note that there are not derived classes of Nodes in ns-3 such as class InternetNode. Instead, components (protocols) are aggregated to a node. Let’s look at how some Ipv4 protocols are added to a node.

static void
AddIpv4Stack(Ptr<Node> node)
{
  Ptr<Ipv4L3Protocol> ipv4 = CreateObject<Ipv4L3Protocol> ();
  ipv4->SetNode (node);
  node->AggregateObject (ipv4);
  Ptr<Ipv4Impl> ipv4Impl = CreateObject<Ipv4Impl> ();
  ipv4Impl->SetIpv4 (ipv4);
  node->AggregateObject (ipv4Impl);
}

Note that the Ipv4 protocols are created using CreateObject(). Then, they are aggregated to the node. In this manner, the Node base class does not need to be edited to allow users with a base class Node pointer to access the Ipv4 interface; users may ask the node for a pointer to its Ipv4 interface at runtime. How the user asks the node is described in the next subsection.

Note that it is a programming error to aggregate more than one object of the same type to an ns3::Object. So, for instance, aggregation is not an option for storing all of the active sockets of a node.

4.3.3.2 GetObject example

GetObject is a type-safe way to achieve a safe downcasting and to allow interfaces to be found on an object.

Consider a node pointer m_node that points to a Node object that has an implementation of IPv4 previously aggregated to it. The client code wishes to configure a default route. To do so, it must access an object within the node that has an interface to the IP forwarding configuration. It performs the following:

  Ptr<Ipv4> ipv4 = m_node->GetObject<Ipv4> ();

If the node in fact does not have an Ipv4 object aggregated to it, then the method will return null. Therefore, it is good practice to check the return value from such a function call. If successful, the user can now use the Ptr to the Ipv4 object that was previously aggregated to the node.

Another example of how one might use aggregation is to add optional models to objects. For instance, an existing Node object may have an “Energy Model” object aggregated to it at run time (without modifying and recompiling the node class). An existing model (such as a wireless net device) can then later "GetObject" for the energy model and act appropriately if the interface has been either built in to the underlying Node object or aggregated to it at run time. However, other nodes need not know anything about energy models.

We hope that this mode of programming will require much less need for developers to modify the base classes.

4.4 Object factories

A common use case is to create lots of similarly configured objects. One can repeatedly call CreateObject but there is also a factory design pattern in use in the ns-3 system. It is heavily used in the "helper" API.

Class ObjectFactory can be used to instantiate objects and to configure the attributes on those objects

  void SetTypeId (TypeId tid);
  void Set (std::string name, const AttributeValue &value);
  Ptr<T> Create (void) const;

The first method allows one to use the ns-3 TypeId system to specify the type of objects created. The second allows one to set attributes on the objects to be created, and the third allows one to create the objects themselves.

For example:

  ObjectFactory factory;
  // Make this factory create objects of type FriisPropagationLossModel
  factory.SetTypeId ("ns3::FriisPropagationLossModel")
  // Make this factory object change a default value of an attribute, for
  // subsequently created objects
  factory.Set ("SystemLoss", DoubleValue (2.0));
  // Create one such object
  Ptr<Object> object = m_factory.Create (); 
  factory.Set ("SystemLoss", DoubleValue (3.0));
  // Create another object
  Ptr<Object> object = m_factory.Create (); 

4.5 Downcasting

A question that has arisen several times is, "If I have a base class pointer (Ptr) to an object and I want the derived class pointer, should I downcast (via C++ dynamic cast) to get the derived pointer, or should I use the object aggregation system to GetObject<> () to find a Ptr to the interface to the subclass API?"

The answer to this is that in many situations, both techniques will work. ns-3 provides a templated function for making the syntax of Object dynamic casting much more user friendly:

template <typename T1, typename T2>
Ptr<T1>
DynamicCast (Ptr<T2> const&p)
{
  return Ptr<T1> (dynamic_cast<T1 *> (PeekPointer (p)));
}

DynamicCast works when the programmer has a base type pointer and is testing against a subclass pointer. GetObject works when looking for different objects aggregated, but also works with subclasses, in the same way as DynamicCast. If unsure, the programmer should use GetObject, as it works in all cases. If the programmer knows the class hierarchy of the object under consideration, it is more direct to just use DynamicCast.

5. Attributes

In ns-3 simulations, there are two main aspects to configuration:

• the simulation topology and how objects are connected
• the values used by the models instantiated in the topology

This chapter focuses on the second item above: how the many values in use in ns-3 are organized, documented, and modifiable by ns-3 users. The ns-3 attribute system is also the underpinning of how traces and statistics are gathered in the simulator.

Before delving into details of the attribute value system, it will help to review some basic properties of class ns3::Object.

5.1 Object Overview

ns-3 is fundamentally a C++ object-based system. By this we mean that new C++ classes (types) can be declared, defined, and subclassed as usual.

Many ns-3 objects inherit from the ns3::Object base class. These objects have some additional properties that we exploit for organizing the system and improving the memory management of our objects:

• a "metadata" system that links the class name to a lot of meta-information about the object, including the base class of the subclass, the set of accessible constructors in the subclass, and the set of "attributes" of the subclass
• a reference counting smart pointer implementation, for memory management.

ns-3 objects that use the attribute system derive from either ns3::Object or ns3::ObjectBase. Most ns-3 objects we will discuss derive from ns3::Object, but a few that are outside the smart pointer memory management framework derive from ns3::ObjectBase.

Let’s review a couple of properties of these objects.

5.1.1 Smart pointers

As introduced in the ns-3 tutorial, ns-3 objects are memory managed by a reference counting smart pointer implementation, class ns3::Ptr.

Smart pointers are used extensively in the ns-3 APIs, to avoid passing references to heap-allocated objects that may cause memory leaks. For most basic usage (syntax), treat a smart pointer like a regular pointer:

  Ptr<WifiNetDevice> nd = ...;
  nd->CallSomeFunction ();
  // etc.

5.1.2 CreateObject

As we discussed above in Memory management and class Ptr, at the lowest-level API, objects of type ns3::Object are not instantiated using operator new as usual but instead by a templated function called CreateObject().

A typical way to create such an object is as follows:

  Ptr<WifiNetDevice> nd = CreateObject<WifiNetDevice> ();

You can think of this as being functionally equivalent to:

  WifiNetDevice* nd = new WifiNetDevice ();

Objects that derive from ns3::Object must be allocated on the heap using CreateObject(). Those deriving from ns3::ObjectBase, such as ns-3 helper functions and packet headers and trailers, can be allocated on the stack.

In some scripts, you may not see a lot of CreateObject() calls in the code; this is because there are some helper objects in effect that are doing the CreateObject()s for you.

5.1.3 TypeId

ns-3 classes that derive from class ns3::Object can include a metadata class called TypeId that records meta-information about the class, for use in the object aggregation and component manager systems:

• a unique string identifying the class
• the base class of the subclass, within the metadata system
• the set of accessible constructors in the subclass

5.1.4 Object Summary

Putting all of these concepts together, let’s look at a specific example: class ns3::Node.

The public header file node.h has a declaration that includes a static GetTypeId function call:

class Node : public Object
{
public:
  static TypeId GetTypeId (void);
  ...

This is defined in the node.cc file as follows:

TypeId 
Node::GetTypeId (void)
{
  static TypeId tid = TypeId ("ns3::Node")
    .SetParent<Object> ()
    .AddConstructor<Node> ()
    .AddAttribute ("DeviceList", "The list of devices associated to this Node.",
                   ObjectVectorValue (),
                   MakeObjectVectorAccessor (&Node::m_devices),
                   MakeObjectVectorChecker<NetDevice> ())
    .AddAttribute ("ApplicationList", "The list of applications associated to this Node.",
                   ObjectVectorValue (),
                   MakeObjectVectorAccessor (&Node::m_applications),
                   MakeObjectVectorChecker<Application> ())
    .AddAttribute ("Id", "The id (unique integer) of this Node.",
                   TypeId::ATTR_GET, // allow only getting it.
                   UintegerValue (0),
                   MakeUintegerAccessor (&Node::m_id),
                   MakeUintegerChecker<uint32_t> ())
    ;
  return tid;

Consider the TypeId of an ns-3 Object class as an extended form of run time type information (RTTI). The C++ language includes a simple kind of RTTI in order to support dynamic_cast and typeid operators.

The “.SetParent<Object> ()” call in the declaration above is used in conjunction with our object aggregation mechanisms to allow safe up- and down-casting in inheritance trees during GetObject.

The “.AddConstructor<Node> ()” call is used in conjunction with our abstract object factory mechanisms to allow us to construct C++ objects without forcing a user to know the concrete class of the object she is building.

The three calls to “.AddAttribute” associate a given string with a strongly typed value in the class. Notice that you must provide a help string which may be displayed, for example, via command line processors. Each Attribute is associated with mechanisms for accessing the underlying member variable in the object (for example, MakeUintegerAccessor tells the generic Attribute code how to get to the node ID above). There are also “Checker” methods which are used to validate values.

When users want to create Nodes, they will usually call some form of CreateObject,

  Ptr<Node> n = CreateObject<Node> ();

or more abstractly, using an object factory, you can create a Node object without even knowing the concrete C++ type

  ObjectFactory factory;
  const std::string typeId = "ns3::Node'';
  factory.SetTypeId (typeId);
  Ptr<Object> node = factory.Create <Object> ();

Both of these methods result in fully initialized attributes being available in the resulting Object instances.

We next discuss how attributes (values associated with member variables or functions of the class) are plumbed into the above TypeId.

5.2 Attribute Overview

The goal of the attribute system is to organize the access of internal member objects of a simulation. This goal arises because, typically in simulation, users will cut and paste/modify existing simulation scripts, or will use higher-level simulation constructs, but often will be interested in studying or tracing particular internal variables. For instance, use cases such as:

• "I want to trace the packets on the wireless interface only on the first access point"
• "I want to trace the value of the TCP congestion window (every time it changes) on a particular TCP socket"
• "I want a dump of all values that were used in my simulation."

Similarly, users may want fine-grained access to internal variables in the simulation, or may want to broadly change the initial value used for a particular parameter in all subsequently created objects. Finally, users may wish to know what variables are settable and retrievable in a simulation configuration. This is not just for direct simulation interaction on the command line; consider also a (future) graphical user interface that would like to be able to provide a feature whereby a user might right-click on an node on the canvas and see a hierarchical, organized list of parameters that are settable on the node and its constituent member objects, and help text and default values for each parameter.

5.2.1 Functional overview

We provide a way for users to access values deep in the system, without having to plumb accessors (pointers) through the system and walk pointer chains to get to them. Consider a class DropTailQueue that has a member variable that is an unsigned integer m_maxPackets; this member variable controls the depth of the queue.

If we look at the declaration of DropTailQueue, we see the following:

class DropTailQueue : public Queue {
public:
  static TypeId GetTypeId (void);
  ...

private:
  std::queue<Ptr<Packet> > m_packets;
  uint32_t m_maxPackets;
};

Let’s consider things that a user may want to do with the value of m_maxPackets:

• Set a default value for the system, such that whenever a new DropTailQueue is created, this member is initialized to that default.
• Set or get the value on an already instantiated queue.

The above things typically require providing Set() and Get() functions, and some type of global default value.

In the ns-3 attribute system, these value definitions and accessor functions are moved into the TypeId class; e.g.:

NS_OBJECT_ENSURE_REGISTERED (DropTailQueue);

TypeId DropTailQueue::GetTypeId (void) 
{
  static TypeId tid = TypeId ("ns3::DropTailQueue")
    .SetParent<Queue> ()
    .AddConstructor<DropTailQueue> ()
    .AddAttribute ("MaxPackets", 
                   "The maximum number of packets accepted by this DropTailQueue.",
                   UintegerValue (100),
                   MakeUintegerAccessor (&DropTailQueue::m_maxPackets),
                   MakeUintegerChecker<uint32_t> ())
    ;
  
  return tid;

The AddAttribute() method is performing a number of things with this value:

• Binding the variable m_maxPackets to a string "MaxPackets"
• Providing a default value (100 packets)
• Providing some help text defining the value
• Providing a "checker" (not used in this example) that can be used to set bounds on the allowable range of values

The key point is that now the value of this variable and its default value are accessible in the attribute namespace, which is based on strings such as "MaxPackets" and TypeId strings. In the next section, we will provide an example script that shows how users may manipulate these values.

Note that initialization of the attribute relies on the macro NS_OBJECT_ENSURE_REGISTERED (DropTailQueue) being called; if you leave this out of your new class implementation, your attributes will not be initialized correctly.

While we have described how to create attributes, we still haven’t described how to access and manage these values. For instance, there is no globals.h header file where these are stored; attributes are stored with their classes. Questions that naturally arise are how do users easily learn about all of the attributes of their models, and how does a user access these attributes, or document their values as part of the record of their simulation?

5.2.2 Default values and command-line arguments

Let’s look at how a user script might access these values. This is based on the script found at samples/main-attribute-value.cc, with some details stripped out.

//
// This is a basic example of how to use the attribute system to
// set and get a value in the underlying system; namely, an unsigned
// integer of the maximum number of packets in a queue
//

int 
main (int argc, char *argv[])
{

  // By default, the MaxPackets attribute has a value of 100 packets
  // (this default can be observed in the function DropTailQueue::GetTypeId)
  // 
  // Here, we set it to 80 packets.  We could use one of two value types:
  // a string-based value or a Uinteger value
  Config::SetDefault ("ns3::DropTailQueue::MaxPackets", StringValue ("80"));
  // The below function call is redundant
  Config::SetDefault ("ns3::DropTailQueue::MaxPackets", UintegerValue (80));

  // Allow the user to override any of the defaults and the above
  // SetDefaults() at run-time, via command-line arguments
  CommandLine cmd;
  cmd.Parse (argc, argv);

The main thing to notice in the above are the two calls to Config::SetDefault. This is how we set the default value for all subsequently instantiated DropTailQueues. We illustrate that two types of Value classes, a StringValue and a UintegerValue class, can be used to assign the value to the attribute named by "ns3::DropTailQueue::MaxPackets".

Now, we will create a few objects using the low-level API; here, our newly created queues will not have a m_maxPackets initialized to 100 packets but to 80 packets, because of what we did above with default values.

  Ptr<Node> n0 = CreateObject<Node> ();

  Ptr<PointToPointNetDevice> net0 = CreateObject<PointToPointNetDevice> ();
  n0->AddDevice (net0);

  Ptr<Queue> q = CreateObject<DropTailQueue> ();
  net0->AddQueue(q);

At this point, we have created a single node (Node 0) and a single PointToPointNetDevice (NetDevice 0) and added a DropTailQueue to it.

Now, we can manipulate the MaxPackets value of the already instantiated DropTailQueue. Here are various ways to do that.

5.2.3 Pointer-based access

We assume that a smart pointer (Ptr) to a relevant network device is in hand; in the current example, it is the net0 pointer.

One way to change the value is to access a pointer to the underlying queue and modify its attribute.

First, we observe that we can get a pointer to the (base class) queue via the PointToPointNetDevice attributes, where it is called TxQueue

  PointerValue tmp;
  net0->GetAttribute ("TxQueue", tmp);
  Ptr<Object> txQueue = tmp.GetObject ();

Using the GetObject function, we can perform a safe downcast to a DropTailQueue, where MaxPackets is a member

  Ptr<DropTailQueue> dtq = txQueue->GetObject <DropTailQueue> ();
  NS_ASSERT (dtq != 0);

Next, we can get the value of an attribute on this queue. We have introduced wrapper "Value" classes for the underlying data types, similar to Java wrappers around these types, since the attribute system stores values and not disparate types. Here, the attribute value is assigned to a UintegerValue, and the Get() method on this value produces the (unwrapped) uint32_t.

  UintegerValue limit;
  dtq->GetAttribute ("MaxPackets", limit);
  NS_LOG_INFO ("1.  dtq limit: " << limit.Get () << " packets");

Note that the above downcast is not really needed; we could have done the same using the Ptr<Queue> even though the attribute is a member of the subclass

  txQueue->GetAttribute ("MaxPackets", limit);
  NS_LOG_INFO ("2.  txQueue limit: " << limit.Get () << " packets");

Now, let’s set it to another value (60 packets)

  txQueue->SetAttribute("MaxPackets", UintegerValue (60));
  txQueue->GetAttribute ("MaxPackets", limit);
  NS_LOG_INFO ("3.  txQueue limit changed: " << limit.Get () << " packets");

5.2.4 Namespace-based access

An alternative way to get at the attribute is to use the configuration namespace. Here, this attribute resides on a known path in this namespace; this approach is useful if one doesn’t have access to the underlying pointers and would like to configure a specific attribute with a single statement.

  Config::Set ("/NodeList/0/DeviceList/0/TxQueue/MaxPackets", UintegerValue (25));
  txQueue->GetAttribute ("MaxPackets", limit); 
  NS_LOG_INFO ("4.  txQueue limit changed through namespace: " << 
    limit.Get () << " packets");

We could have also used wildcards to set this value for all nodes and all net devices (which in this simple example has the same effect as the previous Set())

  Config::Set ("/NodeList/*/DeviceList/*/TxQueue/MaxPackets", UintegerValue (15));
  txQueue->GetAttribute ("MaxPackets", limit); 
  NS_LOG_INFO ("5.  txQueue limit changed through wildcarded namespace: " << 
    limit.Get () << " packets");

5.2.5 Object Name Service-based access

Another way to get at the attribute is to use the object name service facility. Here, this attribute is found using a name string. This approach is useful if one doesn’t have access to the underlying pointers and it is difficult to determine the required concrete configuration namespaced path.

  Names::Add ("server", serverNode);
  Names::Add ("server/eth0", serverDevice);

  ...

  Config::Set ("/Names/server/eth0/TxQueue/MaxPackets", UintegerValue (25));

See section Object names for a fuller treatment of the ns-3 configuration namespace.

5.2.6 Setting through constructors helper classes

Arbitrary combinations of attributes can be set and fetched from the helper and low-level APIs; either from the constructors themselves:

Ptr<Object> p = CreateObject<MyNewObject> ("n1", v1, "n2", v2, ...);

or from the higher-level helper APIs, such as:

  mobility.SetPositionAllocator ("GridPositionAllocator",
                                 "MinX", DoubleValue (-100.0),
                                 "MinY", DoubleValue (-100.0),
                                 "DeltaX", DoubleValue (5.0),
                                 "DeltaY", DoubleValue (20.0),
                                 "GridWidth", UintegerValue (20),
                                 "LayoutType", StringValue ("RowFirst"));

5.2.7 Implementation details

5.2.7.1 Value classes

Readers will note the new FooValue classes which are subclasses of the AttributeValue base class. These can be thought of as an intermediate class that can be used to convert from raw types to the Values that are used by the attribute system. Recall that this database is holding objects of many types with a single generic type. Conversions to this type can either be done using an intermediate class (IntegerValue, DoubleValue for "floating point") or via strings. Direct implicit conversion of types to Value is not really practical. So in the above, users have a choice of using strings or values:

p->Set ("cwnd", StringValue ("100")); // string-based setter
p->Set ("cwnd", IntegerValue (100)); // integer-based setter

The system provides some macros that help users declare and define new AttributeValue subclasses for new types that they want to introduce into the attribute system:

• ATTRIBUTE_HELPER_HEADER
• ATTRIBUTE_HELPER_CPP

5.2.7.2 Initialization order

Attributes in the system must not depend on the state of any other Attribute in this system. This is because an ordering of Attribute initialization is not specified, nor enforced, by the system. A specific example of this can be seen in automated configuration programs such as ns3::ConfigStore. Although a given model may arrange it so that Attributes are initialized in a particular order, another automatic configurator may decide independently to change Attributes in, for example, alphabetic order.

Because of this non-specific ordering, no Attribute in the system may have any dependence on any other Attribute. As a corollary, Attribute setters must never fail due to the state of another Attribute. No Attribute setter may change (set) any other Attribute value as a result of changing its value.

This is a very strong restriction and there are cases where Attributes must set consistently to allow correct operation. To this end we do allow for consistency checking when the attribute is used (cf. NS_ASSERT_MSG or NS_ABORT_MSG).

In general, the attribute code to assign values to the underlying class member variables is executed after an object is constructed. But what if you need the values assigned before the constructor body executes, because you need them in the logic of the constructor? There is a way to do this, used for example in the class ns3::ConfigStore: call ObjectBase::ConstructSelf () as follows:

ConfigStore::ConfigStore ()
{
  ObjectBase::ConstructSelf (AttributeList ());
  // continue on with constructor.
}

5.3 Extending attributes

The ns-3 system will place a number of internal values under the attribute system, but undoubtedly users will want to extend this to pick up ones we have missed, or to add their own classes to this.

5.3.1 Adding an existing internal variable to the metadata system

Consider this variable in class TcpSocket:

 uint32_t m_cWnd;   // Congestion window

Suppose that someone working with TCP wanted to get or set the value of that variable using the metadata system. If it were not already provided by ns-3, the user could declare the following addition in the runtime metadata system (to the TypeId declaration for TcpSocket):

    .AddAttribute ("Congestion window", 
                   "Tcp congestion window (bytes)",
                   UintegerValue (1),
                   MakeUintegerAccessor (&TcpSocket::m_cWnd),
                   MakeUintegerChecker<uint16_t> ())

Now, the user with a pointer to the TcpSocket can perform operations such as setting and getting the value, without having to add these functions explicitly. Furthermore, access controls can be applied, such as allowing the parameter to be read and not written, or bounds checking on the permissible values can be applied.

5.3.2 Adding a new TypeId

Here, we discuss the impact on a user who wants to add a new class to ns-3; what additional things must be done to hook it into this system.

We’ve already introduced what a TypeId definition looks like:

TypeId
RandomWalk2dMobilityModel::GetTypeId (void)
{
  static TypeId tid = TypeId ("ns3::RandomWalk2dMobilityModel")
    .SetParent<MobilityModel> ()
    .SetGroupName ("Mobility")
    .AddConstructor<RandomWalk2dMobilityModel> ()
    .AddAttribute ("Bounds",
                   "Bounds of the area to cruise.",
                   RectangleValue (Rectangle (0.0, 0.0, 100.0, 100.0)),
                   MakeRectangleAccessor (&RandomWalk2dMobilityModel::m_bounds),
                   MakeRectangleChecker ())
    .AddAttribute ("Time",
                   "Change current direction and speed after moving for this delay.",
                   TimeValue (Seconds (1.0)),
                   MakeTimeAccessor (&RandomWalk2dMobilityModel::m_modeTime),
                   MakeTimeChecker ())
    // etc (more parameters).
    ;
  return tid;

The declaration for this in the class declaration is one-line public member method:

 public:
  static TypeId GetTypeId (void);

Typical mistakes here involve:

• Not calling the SetParent method or calling it with the wrong type
• Not calling the AddConstructor method of calling it with the wrong type
• Introducing a typographical error in the name of the TypeId in its constructor
• Not using the fully-qualified c++ typename of the enclosing c++ class as the name of the TypeId

None of these mistakes can be detected by the ns-3 codebase so, users are advised to check carefully multiple times that they got these right.

5.4 Adding new class type to the attribute system

From the perspective of the user who writes a new class in the system and wants to hook it in to the attribute system, there is mainly the matter of writing the conversions to/from strings and attribute values. Most of this can be copy/pasted with macro-ized code. For instance, consider class declaration for Rectangle in the src/mobility/ directory:

5.4.1 Header file

/**
 * \brief a 2d rectangle
 */
class Rectangle
{
  ...

  double xMin;
  double xMax;
  double yMin;
  double yMax;
;

One macro call and two operators, must be added below the class declaration in order to turn a Rectangle into a value usable by the Attribute system:

std::ostream &operator << (std::ostream &os, const Rectangle &rectangle);
std::istream &operator >> (std::istream &is, Rectangle &rectangle);

ATTRIBUTE_HELPER_HEADER (Rectangle);

5.4.2 Implementation file

In the class definition (.cc file), the code looks like this:

ATTRIBUTE_HELPER_CPP (Rectangle);

std::ostream &
operator << (std::ostream &os, const Rectangle &rectangle)
{
  os << rectangle.xMin << "|" << rectangle.xMax << "|" << rectangle.yMin << "|"
     << rectangle.yMax;
  return os;

std::istream &
operator >> (std::istream &is, Rectangle &rectangle)
 {
  char c1, c2, c3;
  is >> rectangle.xMin >> c1 >> rectangle.xMax >> c2 >> rectangle.yMin >> c3 
     >> rectangle.yMax;
  if (c1 != '|' ||
      c2 != '|' ||
      c3 != '|')
    {
      is.setstate (std::ios_base::failbit);
    
  return is;

These stream operators simply convert from a string representation of the Rectangle ("xMin|xMax|yMin|yMax") to the underlying Rectangle, and the modeler must specify these operators and the string syntactical representation of an instance of the new class.

5.5 ConfigStore

Feedback requested: This is an experimental feature of ns-3. It is found in src/contrib and not in the main tree. If you like this feature and would like to provide feedback on it, please email us.

Values for ns-3 attributes can be stored in an ASCII or XML text file and loaded into a future simulation. This feature is known as the ns-3 ConfigStore. The ConfigStore code is in src/contrib/. It is not yet main-tree code, because we are seeking some user feedback and experience with this.

We can explore this system by using an example. Copy the csma-bridge.cc file to the scratch directory:

  cp examples/csma-bridge.cc scratch/
  ./waf

Let’s edit it to add the ConfigStore feature. First, add an include statement to include the contrib module, and then add these lines:

#include "contrib-module.h"
...
int main (...)
{
  // setup topology

  // Invoke just before entering Simulator::Run ()
  ConfigStore config;
  config.ConfigureDefaults ();
  config.ConfigureAttributes ();

  Simulator::Run ();

There are three attributes that govern the behavior of the ConfigStore: "Mode", "Filename", and "FileFormat". The Mode (default "None") configures whether ns-3 should load configuration from a previously saved file (specify "Mode=Load") or save it to a file (specify "Mode=Save"). The Filename (default "") is where the ConfigStore should store its output data. The FileFormat (default "RawText") governs whether the ConfigStore format is Xml or RawText format.

So, using the above modified program, try executing the following waf command and

./waf --command-template="%s --ns3::ConfigStore::Filename=csma-bridge-config.xml
--ns3::ConfigStore::Mode=Save --ns3::ConfigStore::FileFormat=Xml" --run scratch/csma-bridge

After running, you can open the csma-bridge-config.xml file and it will display the configuration that was applied to your simulation; e.g.

<?xml version="1.0" encoding="UTF-8"?>
<ns3>
 <default name="ns3::V4Ping::Remote" value="102.102.102.102"/>
 <default name="ns3::MsduStandardAggregator::MaxAmsduSize" value="7935"/>
 <default name="ns3::EdcaTxopN::MinCw" value="31"/>
 <default name="ns3::EdcaTxopN::MaxCw" value="1023"/>
 <default name="ns3::EdcaTxopN::Aifsn" value="3"/>
 <default name="ns3::QstaWifiMac::ProbeRequestTimeout" value="50000000ns"/>
 <default name="ns3::QstaWifiMac::AssocRequestTimeout" value="500000000ns"/>
 <default name="ns3::QstaWifiMac::MaxMissedBeacons" value="10"/>
 <default name="ns3::QstaWifiMac::ActiveProbing" value="false"/>
...

This file can be archived with your simulation script and output data.

While it is possible to generate a sample config file and lightly edit it to change a couple of values, there are cases where this process will not work because the same value on the same object can appear multiple times in the same automatically-generated configuration file under different configuration paths.

As such, the best way to use this class is to use it to generate an initial configuration file, extract from that configuration file only the strictly necessary elements, and move these minimal elements to a new configuration file which can then safely be edited and loaded in a subsequent simulation run.

When the ConfigStore object is instantiated, its attributes Filename, Mode, and FileFormat must be set, either via command-line or via program statements.

As a more complicated example, let’s assume that we want to read in a configuration of defaults from an input file named "input-defaults.xml", and write out the resulting attributes to a separate file called "output-attributes.xml". (Note– to get this input xml file to begin with, it is sometimes helpful to run the program to generate an output xml file first, then hand-edit that file and re-input it for the next simulation run).

#include "contrib-module.h"
...
int main (...)
{

  Config::SetDefault ("ns3::ConfigStore::Filename", StringValue ("input-defaults.xml"));
  Config::SetDefault ("ns3::ConfigStore::Mode", StringValue ("Load"));
  Config::SetDefault ("ns3::ConfigStore::FileFormat", StringValue ("Xml"));
  ConfigStore inputConfig;
  inputConfig.ConfigureDefaults ();
  
  //
  // Allow the user to override any of the defaults and the above Bind() at
  // run-time, via command-line arguments
  //
  CommandLine cmd;
  cmd.Parse (argc, argv);

  // setup topology
  ...

  // Invoke just before entering Simulator::Run ()
  Config::SetDefault ("ns3::ConfigStore::Filename", StringValue ("output-attributes.xml"));
  Config::SetDefault ("ns3::ConfigStore::Mode", StringValue ("Save"));
  ConfigStore outputConfig;
  outputConfig.ConfigureAttributes ();
  Simulator::Run ();

5.5.1 GTK-based ConfigStore

There is a GTK-based front end for the ConfigStore. This allows users to use a GUI to access and change variables. Screenshots of this feature are available in the ns-3 Overview presentation.

To use this feature, one must install libgtk and libgtk-dev; an example Ubuntu installation command is:

sudo apt-get install libgtk2.0-0 libgtk2.0-dev

To check whether it is configured or not, check the output of the ./waf configure step:

---- Summary of optional NS-3 features:
Threading Primitives          : enabled
Real Time Simulator           : enabled
GtkConfigStore                : not enabled (library 'gtk+-2.0 >= 2.12' not found)

In the above example, it was not enabled, so it cannot be used until a suitable version is installed and ./waf configure; ./waf is rerun.

Usage is almost the same as the non-GTK-based version, but there are no ConfigStore attributes involved:

  // Invoke just before entering Simulator::Run ()
  GtkConfigStore config;
  config.ConfigureDefaults ();
  config.ConfigureAttributes ();

Now, when you run the script, a GUI should pop up, allowing you to open menus of attributes on different nodes/objects, and then launch the simulation execution when you are done.

5.5.2 Future work

There are a couple of possible improvements:

• save a unique version number with date and time at start of file
• save rng initial seed somewhere.
• make each RandomVariable serialize its own initial seed and re-read it later
• add the default values

6. Object names

7. Logging

8. Tracing

The tracing subsystem is one of the most important mechanisms to understand in ns-3. In most cases, ns-3 users will have a brilliant idea for some new and improved networking feature. In order to verify that this idea works, the researcher will make changes to an existing system and then run experiments to see how the new feature behaves by gathering statistics that capture the behavior of the feature.

In other words, the whole point of running a simulation is to generate output for further study. In ns-3, the subsystem that enables a researcher to do this is the tracing subsystem.

8.1 Motivation

There are many ways to get information out of a program. The most straightforward way is to just directly print the information to the standard output, as in,

  #include <iostream>
  ...
  int main ()
  {
    ...
    std::cout << ``The value of x is `` << x << std::endl;
    ...
  } 

This is workable in small environments, but as your simulations get more and more complicated, you end up with more and more prints and the task of parsing and performing computations on the output begins to get harder and harder.

Another thing to consider is that every time a new tidbit is needed, the software core must be edited and another print introduced. There is no standardized way to control all of this output, so the amount of output tends to grow without bounds. Eventually, the bandwidth required for simply outputting this information begins to limit the running time of the simulation. The output files grow to enormous sizes and parsing them becomes a problem.

ns-3 provides a simple mechanism for logging and providing some control over output via Log Components, but the level of control is not very fine grained at all. The logging module is a relatively blunt instrument.

It is desirable to have a facility that allows one to reach into the core system and only get the information required without having to change and recompile the core system. Even better would be a system that notified the user when an item of interest changed or an interesting event happened.

The ns-3 tracing system is designed to work along those lines and is well-integrated with the Attribute and Config substems allowing for relatively simple use scenarios.

8.2 Overview

The tracing subsystem relies heavily on the ns-3 Callback and Attribute mechanisms. You should read and understand the corresponding sections of the manual before attempting to understand the tracing system.

The ns-3 tracing system is built on the concepts of independent tracing sources and tracing sinks; along with a uniform mechanism for connecting sources to sinks.

Trace sources are entities that can signal events that happen in a simulation and provide access to interesting underlying data. For example, a trace source could indicate when a packet is received by a net device and provide access to the packet contents for interested trace sinks. A trace source might also indicate when an interesting state change happens in a model. For example, the congestion window of a TCP model is a prime candidate for a trace source.

Trace sources are not useful by themselves; they must be connected to other pieces of code that actually do something useful with the information provided by the source. The entities that consume trace information are called trace sinks. Trace sources are generators of events and trace sinks are consumers.

This explicit division allows for large numbers of trace sources to be scattered around the system in places which model authors believe might be useful. Unless a user connects a trace sink to one of these sources, nothing is output. This arrangement allows relatively unsophisticated users to attach new types of sinks to existing tracing sources, without requiring editing and recompiling the core or models of the simulator.

There can be zero or more consumers of trace events generated by a trace source. One can think of a trace source as a kind of point-to-multipoint information link.

The “transport protocol” for this conceptual point-to-multipoint link is an ns-3 Callback.

Recall from the Callback Section that callback facility is a way to allow two modules in the system to communicate via function calls while at the same time decoupling the calling function from the called class completely. This is the same requirement as outlined above for the tracing system.

Basically, a trace source is a callback to which multiple functions may be registered. When a trace sink expresses interest in receiving trace events, it adds a callback to a list of callbacks held by the trace source. When an interesting event happens, the trace source invokes its operator() providing zero or more parameters. This tells the source to go through its list of callbacks invoking each one in turn. In this way, the parameter(s) are communicated to the trace sinks, which are just functions.

8.2.1 The Simplest Example

It will be useful to go walk a quick example just to reinforce what we’ve said.

  #include ``ns3/object.h''
  #include ``ns3/uinteger.h''
  #include ``ns3/traced-value.h''
  #include ``ns3/trace-source-accessor.h''
  
  #include <iostream>
  
  using namespace ns3;

The first thing to do is include the required files. As mentioned above, the trace system makes heavy use of the Object and Attribute systems. The first two includes bring in the declarations for those systems. The file, traced-value.h brings in the required declarations for tracing data that obeys value semantics.

In general, value semantics just means that you can pass the object around, not an address. In order to use value semantics at all you have to have an object with an associated copy constructor and assignment operator available. We extend the requirements to talk about the set of operators that are pre-defined for plain-old-data (POD) types. Operator=, operator++, operator–, operator+, operator==, etc.

What this all means is that you will be able to trace changes to an object made using those operators.

  class MyObject : public Object
  {
  public:
    static TypeId GetTypeId (void)
    {
      static TypeId tid = TypeId ("MyObject")
        .SetParent (Object::GetTypeId ())
        .AddConstructor<MyObject> ()
        .AddTraceSource ("MyInteger",
                         "An integer value to trace.",
                         MakeTraceSourceAccessor (&MyObject::m_myInt))
        ;
      return tid;
    
  
    MyObject () {
    TracedValue<uint32_t> m_myInt;
  ;

Since the tracing system is integrated with Attributes, and Attributes work with Objects, there must be an ns-3 Object for the trace source to live in. The two important lines of code are the .AddTraceSource and the TracedValue declaration.

The .AddTraceSource provides the “hooks” used for connecting the trace source to the outside world. The TracedValue declaration provides the infrastructure that overloads the operators mentioned above and drives the callback process.

  void
  IntTrace (Int oldValue, Int newValue)
  {
    std::cout << ``Traced `` << oldValue << `` to `` << newValue << std::endl;
  

This is the definition of the trace sink. It corresponds directly to a callback function. This function will be called whenever one of the operators of the TracedValue is executed.

  int
  main (int argc, char *argv[])
  {
    Ptr<MyObject> myObject = CreateObject<MyObject> ();
  
    myObject->TraceConnectWithoutContext ("MyInteger", MakeCallback(&IntTrace));

    myObject->m_myInt = 1234;
  

In this snippet, the first thing that needs to be done is to create the object in which the trace source lives.

The next step, the TraceConnectWithoutContext, forms the connection between the trace source and the trace sink. Notice the MakeCallback template function. Recall from the Callback section that this creates the specialized functor responsible for providing the overloaded operator() used to “fire” the callback. The overloaded operators (++, –, etc.) will use this operator() to actually invoke the callback. The TraceConnectWithoutContext, takes a string parameter that provides the name of the Attribute assigned to the trace source. Let’s ignore the bit about context for now since it is not important yet.

Finally, the line,

   myObject->m_myInt = 1234;

should be interpreted as an invocation of operator= on the member variable m_myInt with the integer 1234 passed as a parameter. It turns out that this operator is defined (by TracedValue) to execute a callback that returns void and takes two integer values as parameters – an old value and a new value for the integer in question. That is exactly the function signature for the callback function we provided – IntTrace.

To summarize, a trace source is, in essence, a variable that holds a list of callbacks. A trace sink is a function used as the target of a callback. The Attribute and object type information systems are used to provide a way to connect trace sources to trace sinks. The act of “hitting” a trace source is executing an operator on the trace source which fires callbacks. This results in the trace sink callbacks registering interest in the source being called with the parameters provided by the source.

8.2.2 Using the Config Subsystem to Connect to Trace Sources

The TraceConnectWithoutContext call shown above in the simple example is actually very rarely used in the system. More typically, the Config subsystem is used to allow selecting a trace source in the system using what is called a config path.

For example, one might find something that looks like the following in the system (taken from examples/tcp-large-transfer.cc)

  void CwndTracer (uint32_t oldval, uint32_t newval) {

  ...

  Config::ConnectWithoutContext (
    "/NodeList/0/$ns3::TcpL4Protocol/SocketList/0/CongestionWindow", 
    MakeCallback (&CwndTracer));

This should look very familiar. It is the same thing as the previous example, except that a static member function of class Config is being called instead of a method on Object; and instead of an Attribute name, a path is being provided.

The first thing to do is to read the path backward. The last segment of the path must be an Attribute of an Object. In fact, if you had a pointer to the Object that has the “CongestionWindow” Attribute handy (call it theObject), you could write this just like the previous example:

  void CwndTracer (uint32_t oldval, uint32_t newval) {

  ...

  theObject->TraceConnectWithoutContext ("CongestionWindow", MakeCallback (&CwndTracer));

It turns out that the code for Config::ConnectWithoutContext does exactly that. This function takes a path that represents a chain of Object pointers and follows them until it gets to the end of the path and interprets the last segment as an Attribute on the last object. Let’s walk through what happens.

The leading “/” character in the path refers to a so-called namespace. One of the predefined namespaces in the config system is “NodeList” which is a list of all of the nodes in the simulation. Items in the list are referred to by indices into the list, so “/NodeList/0” refers to the zeroth node in the list of nodes created by the simulation. This node is actually a Ptr<Node> and so is a subclass of an ns3::Object.

As described in the Object Model section, ns-3 supports an object aggregation model. The next path segment begins with the “$” character which indicates a GetObject call should be made looking for the type that follows. When a node is initialized by an InternetStackHelper a number of interfaces are aggregated to the node. One of these is the TCP level four protocol. The runtime type of this protocol object is “ns3::TcpL4Protocol”. When the GetObject is executed, it returns a pointer to the object of this type.

The TcpL4Protocol class defines an Attribute called “SocketList” which is a list of sockets. Each socket is actually an ns3::Object with its own Attributes. The items in the list of sockets are referred to by index just as in the NodeList, so “SocketList/0” refers to the zeroth socket in the list of sockets on the zeroth node in the NodeList – the first node constructed in the simulation.

This socket, the type of which turns out to be an ns3::TcpSocketImpl defines an attribute called “CongestionWindow” which is a TracedValue<uint32_t>. The Config::ConnectWithoutContext now does a,

  object->TraceConnectWithoutContext ("CongestionWindow", MakeCallback (&CwndTracer));

using the object pointer from “SocketList/0” which makes the connection between the trace source defined in the socket to the callback – CwndTracer.

Now, whenever a change is made to the TracedValue<uint32_t> representing the congestion window in the TCP socket, the registered callback will be executed and the function CwndTracer will be called printing out the old and new values of the TCP congestion window.

8.3 Using the Tracing API

There are three levels of interaction with the tracing system:

• Beginning user can easily control which objects are participating in tracing;
• Intermediate users can extend the tracing system to modify the output format generated or use existing trace sources in different ways, without modifying the core of the simulator;
• Advanced users can modify the simulator core to add new tracing sources and sinks.

8.4 Using Trace Helpers

The ns-3 trace helpers provide a rich environment for configuring and selecting different trace events and writing them to files. In previous sections, primarily “Building Topologies,” we have seen several varieties of the trace helper methods designed for use inside other (device) helpers.

Perhaps you will recall seeing some of these variations:

  pointToPoint.EnablePcapAll ("second");
  pointToPoint.EnablePcap ("second", p2pNodes.Get (0)->GetId (), 0);
  csma.EnablePcap ("third", csmaDevices.Get (0), true);
  pointToPoint.EnableAsciiAll (ascii.CreateFileStream ("myfirst.tr"));

What may not be obvious, though, is that there is a consistent model for all of the trace-related methods found in the system. We will now take a little time and take a look at the “big picture”.

There are currently two primary use cases of the tracing helpers in ns-3: Device helpers and protocol helpers. Device helpers look at the problem of specifying which traces should be enabled through a node, device pair. For example, you may want to specify that pcap tracing should be enabled on a particular device on a specific node. This follows from the ns-3 device conceptual model, and also the conceptual models of the various device helpers. Following naturally from this, the files created follow a <prefix>-<node>-<device> naming convention.

Protocol helpers look at the problem of specifying which traces should be enabled through a protocol and interface pair. This follows from the ns-3 protocol stack conceptual model, and also the conceptual models of internet stack helpers. Naturally, the trace files should follow a <prefix>-<protocol>-<interface> naming convention.

The trace helpers therefore fall naturally into a two-dimensional taxonomy. There are subtleties that prevent all four classes from behaving identically, but we do strive to make them all work as similarly as possible; and whenever possible there are analogs for all methods in all classes.

                   | pcap | ascii |
  -----------------+------+-------|
  Device Helper    |      |       |
  -----------------+------+-------|
  Protocol Helper  |      |       |
  -----------------+------+-------|

We use an approach called a mixin to add tracing functionality to our helper classes. A mixin is a class that provides functionality to that is inherited by a subclass. Inheriting from a mixin is not considered a form of specialization but is really a way to collect functionality.

Let’s take a quick look at all four of these cases and their respective mixins.

8.4.1 Pcap Tracing Device Helpers

The goal of these helpers is to make it easy to add a consistent pcap trace facility to an ns-3 device. We want all of the various flavors of pcap tracing to work the same across all devices, so the methods of these helpers are inherited by device helpers. Take a look at src/helper/trace-helper.h if you want to follow the discussion while looking at real code.

The class PcapHelperForDevice is a mixin provides the high level functionality for using pcap tracing in an ns-3 device. Every device must implement a single virtual method inherited from this class.

  virtual void EnablePcapInternal (std::string prefix, Ptr<NetDevice> nd, bool promiscuous) = 0;

The signature of this method reflects the device-centric view of the situation at this level. All of the public methods inherited from class 2PcapUserHelperForDevice reduce to calling this single device-dependent implementation method. For example, the lowest level pcap method,

  void EnablePcap (std::string prefix, Ptr<NetDevice> nd, bool promiscuous = false, bool explicitFilename = false);

will call the device implementation of EnablePcapInternal directly. All other public pcap tracing methods build on this implementation to provide additional user-level functionality. What this means to the user is that all device helpers in the system will have all of the pcap trace methods available; and these methods will all work in the same way across devices if the device implements EnablePcapInternal correctly.

8.4.1.1 Pcap Tracing Device Helper Methods

  void EnablePcap (std::string prefix, Ptr<NetDevice> nd, bool promiscuous = false, bool explicitFilename = false);
  void EnablePcap (std::string prefix, std::string ndName, bool promiscuous = false, bool explicitFilename = false);
  void EnablePcap (std::string prefix, NetDeviceContainer d, bool promiscuous = false);
  void EnablePcap (std::string prefix, NodeContainer n, bool promiscuous = false);
  void EnablePcap (std::string prefix, uint32_t nodeid, uint32_t deviceid, bool promiscuous = false);
  void EnablePcapAll (std::string prefix, bool promiscuous = false);

In each of the methods shown above, there is a default parameter called promiscuous that defaults to false. This parameter indicates that the trace should not be gathered in promiscuous mode. If you do want your traces to include all traffic seen by the device (and if the device supports a promiscuous mode) simply add a true parameter to any of the calls above. For example,

  Ptr<NetDevice> nd;
  ...
  helper.EnablePcap ("prefix", nd, true);

will enable promiscuous mode captures on the NetDevice specified by nd.

The first two methods also include a default parameter called explicitFilename that will be discussed below.

You are encouraged to peruse the Doxygen for class PcapHelperForDevice to find the details of these methods; but to summarize ...

You can enable pcap tracing on a particular node/net-device pair by providing a Ptr<NetDevice> to an EnablePcap method. The Ptr<Node> is implicit since the net device must belong to exactly one Node. For example,

  Ptr<NetDevice> nd;
  ...
  helper.EnablePcap ("prefix", nd);

You can enable pcap tracing on a particular node/net-device pair by providing a std::string representing an object name service string to an EnablePcap method. The Ptr<NetDevice> is looked up from the name string. Again, the @code<Node> is implicit since the named net device must belong to exactly one Node. For example,

  Names::Add ("server" ...);
  Names::Add ("server/eth0" ...);
  ...
  helper.EnablePcap ("prefix", "server/ath0");

You can enable pcap tracing on a collection of node/net-device pairs by providing a NetDeviceContainer. For each NetDevice in the container the type is checked. For each device of the proper type (the same type as is managed by the device helper), tracing is enabled. Again, the @code<Node> is implicit since the found net device must belong to exactly one Node. For example,

  NetDeviceContainer d = ...;
  ...
  helper.EnablePcap ("prefix", d);

You can enable pcap tracing on a collection of node/net-device pairs by providing a NodeContainer. For each Node in the NodeContainer its attached NetDevices are iterated. For each NetDevice attached to each node in the container, the type of that device is checked. For each device of the proper type (the same type as is managed by the device helper), tracing is enabled.

  NodeContainer n;
  ...
  helper.EnablePcap ("prefix", n);

You can enable pcap tracing on the basis of node ID and device ID as well as with explicit Ptr. Each Node in the system has an integer node ID and each device connected to a node has an integer device ID.

  helper.EnablePcap ("prefix", 21, 1);

Finally, you can enable pcap tracing for all devices in the system, with the same type as that managed by the device helper.

  helper.EnablePcapAll ("prefix");

8.4.1.2 Pcap Tracing Device Helper Filename Selection

Implicit in the method descriptions above is the construction of a complete filename by the implementation method. By convention, pcap traces in the ns-3 system are of the form “<prefix>-<node id>-<device id>.pcap”

As previously mentioned, every node in the system will have a system-assigned node id; and every device will have an interface index (also called a device id) relative to its node. By default, then, a pcap trace file created as a result of enabling tracing on the first device of node 21 using the prefix “prefix” would be “prefix-21-1.pcap”.

You can always use the ns-3 object name service to make this more clear. For example, if you use the object name service to assign the name “server” to node 21, the resulting pcap trace file name will automatically become, “prefix-server-1.pcap” and if you also assign the name “eth0” to the device, your pcap file name will automatically pick this up and be called “prefix-server-eth0.pcap”.

Finally, two of the methods shown above,

  void EnablePcap (std::string prefix, Ptr<NetDevice> nd, bool promiscuous = false, bool explicitFilename = false);
  void EnablePcap (std::string prefix, std::string ndName, bool promiscuous = false, bool explicitFilename = false);

have a default parameter called explicitFilename. When set to true, this parameter disables the automatic filename completion mechanism and allows you to create an explicit filename. This option is only available in the methods which enable pcap tracing on a single device.

For example, in order to arrange for a device helper to create a single promiscuous pcap capture file of a specific name (“my-pcap-file.pcap”) on a given device, one could:

  Ptr<NetDevice> nd;
  ...
  helper.EnablePcap ("my-pcap-file.pcap", nd, true, true);

The first true parameter enables promiscuous mode traces and the second tells the helper to interpret the prefix parameter as a complete filename.

8.4.2 Ascii Tracing Device Helpers

The behavior of the ascii trace helper mixin is substantially similar to the pcap version. Take a look at src/helper/trace-helper.h if you want to follow the discussion while looking at real code.

The class AsciiTraceHelperForDevice adds the high level functionality for using ascii tracing to a device helper class. As in the pcap case, every device must implement a single virtual method inherited from the ascii trace mixin.

  virtual void EnableAsciiInternal (Ptr<OutputStreamWrapper> stream, std::string prefix, Ptr<NetDevice> nd) = 0;

The signature of this method reflects the device-centric view of the situation at this level; and also the fact that the helper may be writing to a shared output stream. All of the public ascii-trace-related methods inherited from class AsciiTraceHelperForDevice reduce to calling this single device- dependent implementation method. For example, the lowest level ascii trace methods,

  void EnableAscii (std::string prefix, Ptr<NetDevice> nd);
  void EnableAscii (Ptr<OutputStreamWrapper> stream, Ptr<NetDevice> nd);
@verbatim

will call the device implementation of @code{EnableAsciiInternal} directly,
providing either a valid prefix or stream.  All other public ascii tracing 
methods will build on these low-level functions to provide additional user-level
functionality.  What this means to the user is that all device helpers in the 
system will have all of the ascii trace methods available; and these methods
will all work in the same way across devices if the devices implement 
@code{EnablAsciiInternal} correctly.

@subsubsection Ascii Tracing Device Helper Methods

@verbatim
  void EnableAscii (std::string prefix, Ptr<NetDevice> nd);
  void EnableAscii (Ptr<OutputStreamWrapper> stream, Ptr<NetDevice> nd);

  void EnableAscii (std::string prefix, std::string ndName);
  void EnableAscii (Ptr<OutputStreamWrapper> stream, std::string ndName);

  void EnableAscii (std::string prefix, NetDeviceContainer d);
  void EnableAscii (Ptr<OutputStreamWrapper> stream, NetDeviceContainer d);

  void EnableAscii (std::string prefix, NodeContainer n);
  void EnableAscii (Ptr<OutputStreamWrapper> stream, NodeContainer n);

  void EnableAscii (std::string prefix, uint32_t nodeid, uint32_t deviceid);
  void EnableAscii (Ptr<OutputStreamWrapper> stream, uint32_t nodeid, uint32_t deviceid);

  void EnableAsciiAll (std::string prefix);
  void EnableAsciiAll (Ptr<OutputStreamWrapper> stream);

You are encouraged to peruse the Doxygen for class TraceHelperForDevice to find the details of these methods; but to summarize ...

There are twice as many methods available for ascii tracing as there were for pcap tracing. This is because, in addition to the pcap-style model where traces from each unique node/device pair are written to a unique file, we support a model in which trace information for many node/device pairs is written to a common file. This means that the <prefix>-<node>-<device> file name generation mechanism is replaced by a mechanism to refer to a common file; and the number of API methods is doubled to allow all combinations.

Just as in pcap tracing, you can enable ascii tracing on a particular node/net-device pair by providing a Ptr<NetDevice> to an EnableAscii method. The Ptr<Node> is implicit since the net device must belong to exactly one Node. For example,

  Ptr<NetDevice> nd;
  ...
  helper.EnableAscii ("prefix", nd);

In this case, no trace contexts are written to the ascii trace file since they would be redundant. The system will pick the file name to be created using the same rules as described in the pcap section, except that the file will have the suffix “.tr” instead of “.pcap”.

If you want to enable ascii tracing on more than one net device and have all traces sent to a single file, you can do that as well by using an object to refer to a single file:

  Ptr<NetDevice> nd1;
  Ptr<NetDevice> nd2;
  ...
  Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("trace-file-name.tr");
  ...
  helper.EnableAscii (stream, nd1);
  helper.EnableAscii (stream, nd2);
@verbatim

In this case, trace contexts are written to the ascii trace file since they
are required to disambiguate traces from the two devices.  Note that since the
user is completely specifying the file name, the string should include the ``,tr''
for consistency.

You can enable ascii tracing on a particular node/net-device pair by providing a
@code{std::string} representing an object name service string to an 
@code{EnablePcap} method.  The @code{Ptr<NetDevice>} is looked up from the name
string.  Again, the @code<Node> is implicit since the named net device must 
belong to exactly one @code{Node}.  For example, 

@verbatim
  Names::Add ("client" ...);
  Names::Add ("client/eth0" ...);
  Names::Add ("server" ...);
  Names::Add ("server/eth0" ...);
  ...
  helper.EnableAscii ("prefix", "client/eth0");
  helper.EnableAscii ("prefix", "server/eth0");

This would result in two files named “prefix-client-eth0.tr” and “prefix-server-eth0.tr” with traces for each device in the respective trace file. Since all of the EnableAscii functions are overloaded to take a stream wrapper, you can use that form as well:

  Names::Add ("client" ...);
  Names::Add ("client/eth0" ...);
  Names::Add ("server" ...);
  Names::Add ("server/eth0" ...);
  ...
  Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("trace-file-name.tr");
  ...
  helper.EnableAscii (stream, "client/eth0");
  helper.EnableAscii (stream, "server/eth0");

This would result in a single trace file called “trace-file-name.tr” that contains all of the trace events for both devices. The events would be disambiguated by trace context strings.

You can enable ascii tracing on a collection of node/net-device pairs by providing a NetDeviceContainer. For each NetDevice in the container the type is checked. For each device of the proper type (the same type as is managed by the device helper), tracing is enabled. Again, the @code<Node> is implicit since the found net device must belong to exactly one Node. For example,

  NetDeviceContainer d = ...;
  ...
  helper.EnableAscii ("prefix", d);

This would result in a number of ascii trace files being created, each of which follows the <prefix>-<node id>-<device id>.tr convention. Combining all of the traces into a single file is accomplished similarly to the examples above:

  NetDeviceContainer d = ...;
  ...
  Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("trace-file-name.tr");
  ...
  helper.EnableAscii (stream, d);

You can enable ascii tracing on a collection of node/net-device pairs by providing a NodeContainer. For each Node in the NodeContainer its attached NetDevices are iterated. For each NetDevice attached to each node in the container, the type of that device is checked. For each device of the proper type (the same type as is managed by the device helper), tracing is enabled.

  NodeContainer n;
  ...
  helper.EnableAscii ("prefix", n);

This would result in a number of ascii trace files being created, each of which follows the <prefix>-<node id>-<device id>.tr convention. Combining all of the traces into a single file is accomplished similarly to the examples above:

You can enable pcap tracing on the basis of node ID and device ID as well as with explicit Ptr. Each Node in the system has an integer node ID and each device connected to a node has an integer device ID.

  helper.EnableAscii ("prefix", 21, 1);

Of course, the traces can be combined into a single file as shown above.

Finally, you can enable pcap tracing for all devices in the system, with the same type as that managed by the device helper.

  helper.EnableAsciiAll ("prefix");

This would result in a number of ascii trace files being created, one for every device in the system of the type managed by the helper. All of these files will follow the <prefix>-<node id>-<device id>.tr convention. Combining all of the traces into a single file is accomplished similarly to the examples above.

8.4.2.1 Ascii Tracing Device Helper Filename Selection

Implicit in the prefix-style method descriptions above is the construction of the complete filenames by the implementation method. By convention, ascii traces in the ns-3 system are of the form “<prefix>-<node id>-<device id>.tr”

As previously mentioned, every node in the system will have a system-assigned node id; and every device will have an interface index (also called a device id) relative to its node. By default, then, an ascii trace file created as a result of enabling tracing on the first device of node 21, using the prefix “prefix”, would be “prefix-21-1.tr”.

You can always use the ns-3 object name service to make this more clear. For example, if you use the object name service to assign the name “server” to node 21, the resulting ascii trace file name will automatically become, “prefix-server-1.tr” and if you also assign the name “eth0” to the device, your ascii trace file name will automatically pick this up and be called “prefix-server-eth0.tr”.

8.4.3 Pcap Tracing Protocol Helpers

The goal of these mixins is to make it easy to add a consistent pcap trace facility to protocols. We want all of the various flavors of pcap tracing to work the same across all protocols, so the methods of these helpers are inherited by stack helpers. Take a look at src/helper/trace-helper.h if you want to follow the discussion while looking at real code.

In this section we will be illustrating the methods as applied to the protocol Ipv4. To specify traces in similar protocols, just substitute the appropriate type. For example, use a Ptr<Ipv6> instead of a Ptr<Ipv4> and call EnablePcapIpv6 instead of EnablePcapIpv4.

The class PcapHelperForIpv4 provides the high level functionality for using pcap tracing in the Ipv4 protocol. Each protocol helper enabling these methods must implement a single virtual method inherited from this class. There will be a separate implementation for Ipv6, for example, but the only difference will be in the method names and signatures. Different method names are required to disambiguate class Ipv4 from Ipv6 which are both derived from class Object, and methods that share the same signature.

  virtual void EnablePcapIpv4Internal (std::string prefix, Ptr<Ipv4> ipv4, uint32_t interface) = 0;

The signature of this method reflects the protocol and interface-centric view of the situation at this level. All of the public methods inherited from class PcapHelperForIpv4 reduce to calling this single device-dependent implementation method. For example, the lowest level pcap method,

  void EnablePcapIpv4 (std::string prefix, Ptr<Ipv4> ipv4, uint32_t interface);
@verbatim

will call the device implementation of @code{EnablePcapIpv4Internal} directly.
All other public pcap tracing methods build on this implementation to provide 
additional user-level functionality.  What this means to the user is that all 
protocol helpers in the system will have all of the pcap trace methods 
available; and these methods will all work in the same way across 
protocols if the helper implements @code{EnablePcapIpv4Internal} correctly.

@subsubsection Pcap Tracing Protocol Helper Methods

These methods are designed to be in one-to-one correspondence with the @code{Node}-
and @code{NetDevice}- centric versions of the device versions.  Instead of
@code{Node} and @code{NetDevice} pair constraints, we use protocol and interface
constraints.

Note that just like in the device version, there are six methods:

@verbatim
  void EnablePcapIpv4 (std::string prefix, Ptr<Ipv4> ipv4, uint32_t interface);
  void EnablePcapIpv4 (std::string prefix, std::string ipv4Name, uint32_t interface);
  void EnablePcapIpv4 (std::string prefix, Ipv4InterfaceContainer c);
  void EnablePcapIpv4 (std::string prefix, NodeContainer n);
  void EnablePcapIpv4 (std::string prefix, uint32_t nodeid, uint32_t interface);
  void EnablePcapIpv4All (std::string prefix);

You are encouraged to peruse the Doxygen for class PcapHelperForIpv4 to find the details of these methods; but to summarize ...

You can enable pcap tracing on a particular protocol/interface pair by providing a Ptr<Ipv4> and interface to an EnablePcap method. For example,

  Ptr<Ipv4> ipv4 = node->GetObject<Ipv4> ();
  ...
  helper.EnablePcapIpv4 ("prefix", ipv4, 0);

You can enable pcap tracing on a particular node/net-device pair by providing a std::string representing an object name service string to an EnablePcap method. The Ptr<Ipv4> is looked up from the name string. For example,

  Names::Add ("serverIPv4" ...);
  ...
  helper.EnablePcapIpv4 ("prefix", "serverIpv4", 1);

You can enable pcap tracing on a collection of protocol/interface pairs by providing an Ipv4InterfaceContainer. For each Ipv4 / interface pair in the container the protocol type is checked. For each protocol of the proper type (the same type as is managed by the device helper), tracing is enabled for the corresponding interface. For example,

  NodeContainer nodes;
  ...
  NetDeviceContainer devices = deviceHelper.Install (nodes);
  ... 
  Ipv4AddressHelper ipv4;
  ipv4.SetBase ("10.1.1.0", "255.255.255.0");
  Ipv4InterfaceContainer interfaces = ipv4.Assign (devices);
  ...
  helper.EnablePcapIpv4 ("prefix", interfaces);

You can enable pcap tracing on a collection of protocol/interface pairs by providing a NodeContainer. For each Node in the NodeContainer the appropriate protocol is found. For each protocol, its interfaces are enumerated and tracing is enabled on the resulting pairs. For example,

  NodeContainer n;
  ...
  helper.EnablePcapIpv4 ("prefix", n);

You can enable pcap tracing on the basis of node ID and interface as well. In this case, the node-id is translated to a Ptr<Node> and the appropriate protocol is looked up in the node. The resulting protocol and interface are used to specify the resulting trace source.

  helper.EnablePcapIpv4 ("prefix", 21, 1);

Finally, you can enable pcap tracing for all interfaces in the system, with associated protocol being the same type as that managed by the device helper.

  helper.EnablePcapIpv4All ("prefix");

8.4.3.1 Pcap Tracing Protocol Helper Filename Selection

Implicit in all of the method descriptions above is the construction of the complete filenames by the implementation method. By convention, pcap traces taken for devices in the ns-3 system are of the form “<prefix>-<node id>-<device id>.pcap”. In the case of protocol traces, there is a one-to-one correspondence between protocols and Nodes. This is because protocol Objects are aggregated to Node Objects. Since there is no global protocol id in the system, we use the corresponding node id in file naming. Therefore there is a possibility for file name collisions in automatically chosen trace file names. For this reason, the file name convention is changed for protocol traces.

As previously mentioned, every node in the system will have a system-assigned node id. Since there is a one-to-one correspondence between protocol instances and node instances we use the node id. Each interface has an interface id relative to its protocol. We use the convention "<prefix>-n<node id>-i<interface id>.pcap" for trace file naming in protocol helpers.

Therefore, by default, a pcap trace file created as a result of enabling tracing on interface 1 of the Ipv4 protocol of node 21 using the prefix “prefix” would be “prefix-n21-i1.pcap”.

You can always use the ns-3 object name service to make this more clear. For example, if you use the object name service to assign the name “serverIpv4” to the Ptr<Ipv4> on node 21, the resulting pcap trace file name will automatically become, “prefix-nserverIpv4-i1.pcap”.

8.4.4 Ascii Tracing Protocol Helpers

The behavior of the ascii trace helpers is substantially similar to the pcap case. Take a look at src/helper/trace-helper.h if you want to follow the discussion while looking at real code.

In this section we will be illustrating the methods as applied to the protocol Ipv4. To specify traces in similar protocols, just substitute the appropriate type. For example, use a Ptr<Ipv6> instead of a Ptr<Ipv4> and call EnableAsciiIpv6 instead of EnableAsciiIpv4.

The class AsciiTraceHelperForIpv4 adds the high level functionality for using ascii tracing to a protocol helper. Each protocol that enables these methods must implement a single virtual method inherited from this class.

  virtual void EnableAsciiIpv4Internal (Ptr<OutputStreamWrapper> stream, std::string prefix, 
                                        Ptr<Ipv4> ipv4, uint32_t interface) = 0;

The signature of this method reflects the protocol- and interface-centric view of the situation at this level; and also the fact that the helper may be writing to a shared output stream. All of the public methods inherited from class PcapAndAsciiTraceHelperForIpv4 reduce to calling this single device- dependent implementation method. For example, the lowest level ascii trace methods,

  void EnableAsciiIpv4 (std::string prefix, Ptr<Ipv4> ipv4, uint32_t interface);
  void EnableAsciiIpv4 (Ptr<OutputStreamWrapper> stream, Ptr<Ipv4> ipv4, uint32_t interface);
@verbatim

will call the device implementation of @code{EnableAsciiIpv4Internal} directly,
providing either the prefix or the stream.  All other public ascii tracing 
methods will build on these low-level functions to provide additional user-level
functionality.  What this means to the user is that all device helpers in the 
system will have all of the ascii trace methods available; and these methods
will all work in the same way across protocols if the protocols implement 
@code{EnablAsciiIpv4Internal} correctly.

@subsubsection Ascii Tracing Device Helper Methods

@verbatim
  void EnableAsciiIpv4 (std::string prefix, Ptr<Ipv4> ipv4, uint32_t interface);
  void EnableAsciiIpv4 (Ptr<OutputStreamWrapper> stream, Ptr<Ipv4> ipv4, uint32_t interface);

  void EnableAsciiIpv4 (std::string prefix, std::string ipv4Name, uint32_t interface);
  void EnableAsciiIpv4 (Ptr<OutputStreamWrapper> stream, std::string ipv4Name, uint32_t interface);

  void EnableAsciiIpv4 (std::string prefix, Ipv4InterfaceContainer c);
  void EnableAsciiIpv4 (Ptr<OutputStreamWrapper> stream, Ipv4InterfaceContainer c);

  void EnableAsciiIpv4 (std::string prefix, NodeContainer n);
  void EnableAsciiIpv4 (Ptr<OutputStreamWrapper> stream, NodeContainer n);

  void EnableAsciiIpv4 (std::string prefix, uint32_t nodeid, uint32_t deviceid);
  void EnableAsciiIpv4 (Ptr<OutputStreamWrapper> stream, uint32_t nodeid, uint32_t interface);

  void EnableAsciiIpv4All (std::string prefix);
  void EnableAsciiIpv4All (Ptr<OutputStreamWrapper> stream);

You are encouraged to peruse the Doxygen for class PcapAndAsciiHelperForIpv4 to find the details of these methods; but to summarize ...

There are twice as many methods available for ascii tracing as there were for pcap tracing. This is because, in addition to the pcap-style model where traces from each unique protocol/interface pair are written to a unique file, we support a model in which trace information for many protocol/interface pairs is written to a common file. This means that the <prefix>-n<node id>-<interface> file name generation mechanism is replaced by a mechanism to refer to a common file; and the number of API methods is doubled to allow all combinations.

Just as in pcap tracing, you can enable ascii tracing on a particular protocol/interface pair by providing a Ptr<Ipv4> and an interface to an EnableAscii method. For example,

  Ptr<Ipv4> ipv4;
  ...
  helper.EnableAsciiIpv4 ("prefix", ipv4, 1);

In this case, no trace contexts are written to the ascii trace file since they would be redundant. The system will pick the file name to be created using the same rules as described in the pcap section, except that the file will have the suffix “.tr” instead of “.pcap”.

If you want to enable ascii tracing on more than one interface and have all traces sent to a single file, you can do that as well by using an object to refer to a single file. We have already something similar to this in the “cwnd” example above:

  Ptr<Ipv4> protocol1 = node1->GetObject<Ipv4> ();
  Ptr<Ipv4> protocol2 = node2->GetObject<Ipv4> ();
  ...
  Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("trace-file-name.tr");
  ...
  helper.EnableAsciiIpv4 (stream, protocol1, 1);
  helper.EnableAsciiIpv4 (stream, protocol2, 1);
@verbatim

In this case, trace contexts are written to the ascii trace file since they
are required to disambiguate traces from the two interfaces.  Note that since 
the user is completely specifying the file name, the string should include the
``,tr'' for consistency.

You can enable ascii tracing on a particular protocol by providing a 
@code{std::string} representing an object name service string to an 
@code{EnablePcap} method.  The @code{Ptr<Ipv4>} is looked up from the name
string.  The @code<Node> in the resulting filenames is implicit since there
is a one-to-one correspondence between protocol instances and nodes,
For example, 

@verbatim
  Names::Add ("node1Ipv4" ...);
  Names::Add ("node2Ipv4" ...);
  ...
  helper.EnableAsciiIpv4 ("prefix", "node1Ipv4", 1);
  helper.EnableAsciiIpv4 ("prefix", "node2Ipv4", 1);

This would result in two files named “prefix-nnode1Ipv4-i1.tr” and “prefix-nnode2Ipv4-i1.tr” with traces for each interface in the respective trace file. Since all of the EnableAscii functions are overloaded to take a stream wrapper, you can use that form as well:

  Names::Add ("node1Ipv4" ...);
  Names::Add ("node2Ipv4" ...);
  ...
  Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("trace-file-name.tr");
  ...
  helper.EnableAsciiIpv4 (stream, "node1Ipv4", 1);
  helper.EnableAsciiIpv4 (stream, "node2Ipv4", 1);

This would result in a single trace file called “trace-file-name.tr” that contains all of the trace events for both interfaces. The events would be disambiguated by trace context strings.

You can enable ascii tracing on a collection of protocol/interface pairs by providing an Ipv4InterfaceContainer. For each protocol of the proper type (the same type as is managed by the device helper), tracing is enabled for the corresponding interface. Again, the @code<Node> is implicit since there is a one-to-one correspondence between each protocol and its node. For example,

  NodeContainer nodes;
  ...
  NetDeviceContainer devices = deviceHelper.Install (nodes);
  ... 
  Ipv4AddressHelper ipv4;
  ipv4.SetBase ("10.1.1.0", "255.255.255.0");
  Ipv4InterfaceContainer interfaces = ipv4.Assign (devices);
  ...
  ...
  helper.EnableAsciiIpv4 ("prefix", interfaces);

This would result in a number of ascii trace files being created, each of which follows the <prefix>-n<node id>-i<interface>.tr convention. Combining all of the traces into a single file is accomplished similarly to the examples above:

  NodeContainer nodes;
  ...
  NetDeviceContainer devices = deviceHelper.Install (nodes);
  ... 
  Ipv4AddressHelper ipv4;
  ipv4.SetBase ("10.1.1.0", "255.255.255.0");
  Ipv4InterfaceContainer interfaces = ipv4.Assign (devices);
  ...
  Ptr<OutputStreamWrapper> stream = asciiTraceHelper.CreateFileStream ("trace-file-name.tr");
  ...
  helper.EnableAsciiIpv4 (stream, interfaces);

You can enable ascii tracing on a collection of protocol/interface pairs by providing a NodeContainer. For each Node in the NodeContainer the appropriate protocol is found. For each protocol, its interfaces are enumerated and tracing is enabled on the resulting pairs. For example,

  NodeContainer n;
  ...
  helper.EnableAsciiIpv4 ("prefix", n);

This would result in a number of ascii trace files being created, each of which follows the <prefix>-<node id>-<device id>.tr convention. Combining all of the traces into a single file is accomplished similarly to the examples above:

You can enable pcap tracing on the basis of node ID and device ID as well. In this case, the node-id is translated to a Ptr<Node> and the appropriate protocol is looked up in the node. The resulting protocol and interface are used to specify the resulting trace source.

  helper.EnableAsciiIpv4 ("prefix", 21, 1);

Of course, the traces can be combined into a single file as shown above.

Finally, you can enable ascii tracing for all interfaces in the system, with associated protocol being the same type as that managed by the device helper.

  helper.EnableAsciiIpv4All ("prefix");

This would result in a number of ascii trace files being created, one for every interface in the system related to a protocol of the type managed by the helper. All of these files will follow the <prefix>-n<node id>-i<interface.tr convention. Combining all of the traces into a single file is accomplished similarly to the examples above.

8.4.4.1 Ascii Tracing Device Helper Filename Selection

Implicit in the prefix-style method descriptions above is the construction of the complete filenames by the implementation method. By convention, ascii traces in the ns-3 system are of the form “<prefix>-<node id>-<device id>.tr”

As previously mentioned, every node in the system will have a system-assigned node id. Since there is a one-to-one correspondence between protocols and nodes we use to node-id to identify the protocol identity. Every interface on a given protocol will have an interface index (also called simply an interface) relative to its protocol. By default, then, an ascii trace file created as a result of enabling tracing on the first device of node 21, using the prefix “prefix”, would be “prefix-n21-i1.tr”. Use the prefix to disambiguate multiple protocols per node.

You can always use the ns-3 object name service to make this more clear. For example, if you use the object name service to assign the name “serverIpv4” to the protocol on node 21, and also specify interface one, the resulting ascii trace file name will automatically become, “prefix-nserverIpv4-1.tr”.

8.5 Implementation details

9. Real-Time Scheduler

ns-3 has been designed for integration into testbed and virtual machine environments. To integrate with real network stacks and emit/consume packets, a real-time scheduler is needed to try to lock the simulation clock with the hardware clock. We describe here a component of this: the RealTime scheduler.

The purpose of the realtime scheduler is to cause the progression of the simulation clock to occur synchronously with respect to some external time base. Without the presence of an external time base (wall clock), simulation time jumps instantly from one simulated time to the next.

9.1 Behavior

When using a non-realtime scheduler (the default in ns-3), the simulator advances the simulation time to the next scheduled event. During event execution, simulation time is frozen. With the realtime scheduler, the behavior is similar from the perspective of simulation models (i.e., simulation time is frozen during event execution), but between events, the simulator will attempt to keep the simulation clock aligned with the machine clock.

When an event is finished executing, and the scheduler moves to the next event, the scheduler compares the next event execution time with the machine clock. If the next event is scheduled for a future time, the simulator sleeps until that realtime is reached and then executes the next event.

It may happen that, due to the processing inherent in the execution of simulation events, that the simulator cannot keep up with realtime. In such a case, it is up to the user configuration what to do. There are two ns-3 attributes that govern the behavior. The first is ns3::RealTimeSimulatorImpl::SynchronizationMode. The two entries possible for this attribute are BestEffort (the default) or HardLimit. In "BestEffort" mode, the simulator will just try to catch up to realtime by executing events until it reaches a point where the next event is in the (realtime) future, or else the simulation ends. In BestEffort mode, then, it is possible for the simulation to consume more time than the wall clock time. The other option "HardLimit" will cause the simulation to abort if the tolerance threshold is exceeded. This attribute is ns3::RealTimeSimulatorImpl::HardLimit and the default is 0.1 seconds.

A different mode of operation is one in which simulated time is not frozen during an event execution. This mode of realtime simulation was implemented but removed from the ns-3 tree because of questions of whether it would be useful. If users are interested in a realtime simulator for which simulation time does not freeze during event execution (i.e., every call to Simulator::Now() returns the current wall clock time, not the time at which the event started executing), please contact the ns-developers mailing list.

9.2 Usage

The usage of the realtime simulator is straightforward, from a scripting perspective. Users just need to set the attribute SimulatorImplementationType to the Realtime simulator, such as follows:

  GlobalValue::Bind ("SimulatorImplementationType",
    StringValue ("ns3::RealtimeSimulatorImpl"));

There is a script in examples/realtime-udp-echo.cc that has an example of how to configure the realtime behavior. Try:

./waf --run realtime-udp-echo

Whether the simulator will work in a best effort or hard limit policy fashion is governed by the attributes explained in the previous section.

9.3 Implementation

The implementation is contained in the following files:

• src/simulator/realtime-simulator-impl.{cc,h}
• src/simulator/wall-clock-synchronizer.{cc,h}

In order to create a realtime scheduler, to a first approximation you just want to cause simulation time jumps to consume real time. We propose doing this using a combination of sleep- and busy- waits. Sleep-waits cause the calling process (thread) to yield the processor for some amount of time. Even though this specified amount of time can be passed to nanosecond resolution, it is actually converted to an OS-specific granularity. In Linux, the granularity is called a Jiffy. Typically this resolution is insufficient for our needs (on the order of a ten milliseconds), so we round down and sleep for some smaller number of Jiffies. The process is then awakened after the specified number of Jiffies has passed. At this time, we have some residual time to wait. This time is generally smaller than the minimum sleep time, so we busy-wait for the remainder of the time. This means that the thread just sits in a for loop consuming cycles until the desired time arrives. After the combination of sleep- and busy-waits, the elapsed realtime (wall) clock should agree with the simulation time of the next event and the simulation proceeds.

10. Distributed Simulation with MPI

Parallel and distributed discrete event simulation allows the execution of a single simulation program on multiple processors. By splitting up the simulation into logical processes, LPs, each LP can be executed by a different processor. This simulation methodology enables very large-scale simulations by leveraging increased processing power and memory availability. In order to ensure proper execution of a distributed simulation, message passing between LPs is required. To support distributed simulation in ns-3, the standard Message Passing Interface (MPI) is used, along with a new distributed simulator class. Currently, dividing a simulation for distributed purposes in ns-3 can only occur across point-to-point links.

10.1 Current Implementation Details

During the course of a distributed simulation, many packets must cross simulator boundaries. In other words, a packet that originated on one LP is destined for a different LP, and in order to make this transition, a message containing the packet contents must be sent to the remote LP. Upon receiving this message, the remote LP can rebuild the packet and proceed as normal. The process of sending an receiving messages between LPs is handled easily by the new MPI interface in ns-3.

Along with simple message passing between LPs, a distributed simulator is used on each LP to determine which events to process. It is important to process events in time-stamped order to ensure proper simulation execution. If a LP receives a message containing an event from the past, clearly this is an issue, since this event could change other events which have already been executed. To address this problem, a conservative synchronization algorithm with lookahead is used in ns-3. For more information on different synchronization approaches and parallel and distributed simulation in general, please refer to "Parallel and Distributed Simulation Systems" by Richard Fujimoto.

10.1.1 Remote point-to-point links

As described in the introduction, dividing a simulation for distributed purposes in ns-3 currently can only occur across point-to-point links; therefore, the idea of remote point-to-point links is very important for distributed simulation in ns-3. When a point-to-point link is installed, connecting two nodes, the point-to-point helper checks the system id, or rank, of both nodes. The rank should be assigned during node creation for distributed simulation and is intended to signify on which LP a node belongs. If the two nodes are on the same rank, a regular point-to-point link is created. If, however, the two nodes are on different ranks, then these nodes are intended for different LPs, and a remote point-to-point link is used. If a packet is to be sent across a remote point-to-point link, MPI is used to send the message to the remote LP.

10.1.2 Distributing the topology

Currently, the full topology is created on each rank, regardless of the individual node system ids. Only the applications are specific to a rank. For example, consider node 1 on LP 1 and node 2 on LP 2, with a traffic generator on node 1. Both node 1 and node 2 will be created on both LP1 and LP2; however, the traffic generator will only be installed on LP1. While this is not optimal for memory efficiency, it does simplify routing, since all current routing implementations in ns-3 will work with distributed simulation.

10.2 Running Distributed Simulations

10.2.1 Prerequisites

Ensure that MPI is installed, as well as mpic++. In Ubuntu repositories, these are openmpi-bin, openmpi-common, openmpi-doc, libopenmpi-dev. In Fedora, these are openmpi and openmpi-devel.

Note: There is a conflict on some Fedora systems between libotf and openmpi. A possible "quick-fix" is to yum remove libotf before installing openmpi. This will remove conflict, but it will also remove emacs. Alternatively, these steps could be followed to resolve the conflict:

1) Rename the tiny otfdump which emacs says it needs:

     mv /usr/bin/otfdump /usr/bin/otfdump.emacs-version

2) Manually resolve openmpi dependencies:

     sudo yum install libgfortran libtorque numactl

3) Download rpm packages: 

     openmpi-1.3.1-1.fc11.i586.rpm
     openmpi-devel-1.3.1-1.fc11.i586.rpm
     openmpi-libs-1.3.1-1.fc11.i586.rpm
     openmpi-vt-1.3.1-1.fc11.i586.rpm

     from

     http://mirrors.kernel.org/fedora/releases/11/Everything/i386/os/Packages/

4) Force the packages in:

     sudo rpm -ivh --force openmpi-1.3.1-1.fc11.i586.rpm
     openmpi-libs-1.3.1-1.fc11.i586.rpm openmpi-devel-1.3.1-1.fc11.i586.rpm
     openmpi-vt-1.3.1-1.fc11.i586.rpm

Also, it may be necessary to add the openmpi bin directory to PATH in order to execute mpic++ and mpirun from the command line. Alternatively, the full path to these executables can be used. Finally, if openmpi complains about the inability to open shared libraries, such as libmpi_cxx.so.0, it may be necessary to add the openmpi lib directory to LD_LIBRARY_PATH.

10.2.2 Building and Running Examples

If you already built ns-3 without MPI enabled, you must re-build:

./waf distclean

Configure ns-3 with the –enable-mpi option:

./waf -d debug configure --enable-mpi

Ensure that MPI is enabled by checking the optional features shown from the output of configure.

Next, build ns-3:

./waf

After building ns-3 with mpi enabled, the example programs are now ready to run with mpirun. Here are a few examples (from the root ns-3 directory):

mpirun -np 2 ./waf --run simple-distributed
mpirun -np 4 -machinefile mpihosts ./waf --run 'nms-udp-nix --LAN=2 --CN=4 --nix=1 --tracing=0'

The np switch is the number of logical processors to use. The machinefile switch is which machines to use. In order to use machinefile, the target file must exist (in this case mpihosts). This can simply contain something like:

localhost
localhost
localhost
...

Or if you have a cluster of machines, you can name them.

** NOTE: Some users have experienced issues using mpirun and waf together. An alternative way to run distributed examples is shown below:

./waf shell
cd build/debug
mpirun -np 2 examples/mpi/simple-distributed

10.2.3 Creating custom topologies

The example programs in examples/mpi give a good idea of how to create different topologies for distributed simulation. The main points are assigning system ids to individual nodes, creating point-to-point links where the simulation should be divided, and installing applications only on the LP associated with the target node.

Assigning system ids to nodes is simple and can be handled two different ways. First, a NodeContainer can be used to create the nodes and assign system ids:

NodeContainer nodes;
nodes.Create (5, 1); // Creates 5 nodes with system id 1.

Alternatively, nodes can be created individually, assigned system ids, and added to a NodeContainer. This is useful if a NodeContainer holds nodes with different system ids:

NodeContainer nodes;
Ptr<Node> node1 = CreateObject<Node> (0); // Create node1 with system id 0
Ptr<Node> node2 = CreateObject<Node> (1); // Create node2 with system id 1
nodes.Add (node1);
nodes.Add (node2);

Next, where the simulation is divided is determined by the placement of point-to-point links. If a point-to-point link is created between two nodes with different system ids, a remote point-to-point link is created, as described in Current Implementation Details.

Finally, installing applications only on the LP associated with the target node is very important. For example, if a traffic generator is to be placed on node 0, which is on LP0, only LP0 should install this application. This is easily accomplished by first checking the simulator system id, and ensuring that it matches the system id of the target node before installing the application.

10.3 Tracing During Distributed Simulations

Depending on the system id (rank) of the simulator, the information traced will be different, since traffic originating on one simulator is not seen by another simulator until it reaches nodes specific to that simulator. The easiest way to keep track of different traces is to just name the trace files or pcaps differently, based on the system id of the simulator. For example, something like this should work well, assuming all of these local variables were previously defined:

if (MpiInterface::GetSystemId () == 0)
  {
    pointToPoint.EnablePcapAll ("distributed-rank0");
    phy.EnablePcap ("distributed-rank0", apDevices.Get (0));
    csma.EnablePcap ("distributed-rank0", csmaDevices.Get (0), true);
  }
else if (MpiInterface::GetSystemId () == 1)
  {
    pointToPoint.EnablePcapAll ("distributed-rank1");
    phy.EnablePcap ("distributed-rank1", apDevices.Get (0));
    csma.EnablePcap ("distributed-rank1", csmaDevices.Get (0), true);
  }

11. Packets

The design of the Packet framework of ns was heavily guided by a few important use-cases:

• avoid changing the core of the simulator to introduce new types of packet headers or trailers
• maximize the ease of integration with real-world code and systems
• make it easy to support fragmentation, defragmentation, and, concatenation which are important, especially in wireless systems.
• make memory management of this object efficient
• allow actual application data or dummy application bytes for emulated applications

Each network packet contains a byte buffer, a set of byte tags, a set of packet tags, and metadata.

The byte buffer stores the serialized content of the headers and trailers added to a packet. The serialized representation of these headers is expected to match that of real network packets bit for bit (although nothing forces you to do this) which means that the content of a packet buffer is expected to be that of a real packet.

Fragmentation and defragmentation are quite natural to implement within this context: since we have a buffer of real bytes, we can split it in multiple fragments and re-assemble these fragments. We expect that this choice will make it really easy to wrap our Packet data structure within Linux-style skb or BSD-style mbuf to integrate real-world kernel code in the simulator. We also expect that performing a real-time plug of the simulator to a real-world network will be easy.

One problem that this design choice raises is that it is difficult to pretty-print the packet headers without context. The packet metadata describes the type of the headers and trailers which were serialized in the byte buffer. The maintenance of metadata is optional and disabled by default. To enable it, you must call Packet::EnableMetadata() and this will allow you to get non-empty output from Packet::Print and Packet::Print.

Also, developers often want to store data in packet objects that is not found in the real packets (such as timestamps or flow-ids). The Packet class deals with this requirement by storing a set of tags (class Tag). We have found two classes of use cases for these tags, which leads to two different types of tags. So-called ’byte’ tags are used to tag a subset of the bytes in the packet byte buffer while ’packet’ tags are used to tag the packet itself. The main difference between these two kinds of tags is what happens when packets are copied, fragmented, and reassembled: ’byte’ tags follow bytes while ’packet’ tags follow packets. Another important difference between these two kinds of tags is that byte tags cannot be removed and are expected to be written once, and read many times, while packet tags are expected to be written once, read many times, and removed exactly once. An example of a ’byte’ tag is a FlowIdTag which contains a flow id and is set by the application generating traffic. An example of a ’packet’ tag is a cross-layer QoS class id set by an application and processed by a lower-level MAC layer.

Memory management of Packet objects is entirely automatic and extremely efficient: memory for the application-level payload can be modeled by a virtual buffer of zero-filled bytes for which memory is never allocated unless explicitly requested by the user or unless the packet is fragmented or serialized out to a real network device. Furthermore, copying, adding, and, removing headers or trailers to a packet has been optimized to be virtually free through a technique known as Copy On Write.

Packets (messages) are fundamental objects in the simulator and their design is important from a performance and resource management perspective. There are various ways to design the simulation packet, and tradeoffs among the different approaches. In particular, there is a tension between ease-of-use, performance, and safe interface design.

11.1 Packet design overview

Unlike ns-2, in which Packet objects contain a buffer of C++ structures corresponding to protocol headers, each network packet in ns-3 contains a byte Buffer, a list of byte Tags, a list of packet Tags, and a PacketMetadata object:

• The byte buffer stores the serialized content of the chunks added to a packet. The serialized representation of these chunks is expected to match that of real network packets bit for bit (although nothing forces you to do this) which means that the content of a packet buffer is expected to be that of a real packet. Packets can also be created with an arbitrary zero-filled payload for which no real memory is allocated.
• Each list of tags stores an arbitrarily large set of arbitrary user-provided data structures in the packet. Each Tag is uniquely identified by its type; only one instance of each type of data structure is allowed in a list of tags. These tags typically contain per-packet cross-layer information or flow identifiers (i.e., things that you wouldn’t find in the bits on the wire).

Figure 11.1: Implementation overview of Packet class.

Figure fig:packets is a high-level overview of the Packet implementation; more detail on the byte Buffer implementation is provided later in Figure fig:buffer. In \nsthree, the Packet byte buffer is analogous to a Linux skbuff or BSD mbuf; it is a serialized representation of the actual data in the packet. The tag lists are containers for extra items useful for simulation convenience; if a Packet is converted to an emulated packet and put over an actual network, the tags are stripped off and the byte buffer is copied directly into a real packet.

Packets are reference counted objects. They are handled with smart pointer (Ptr) objects like many of the objects in the ns-3 system. One small difference you will see is that class Packet does not inherit from class Object or class RefCountBase, and implements the Ref() and Unref() methods directly. This was designed to avoid the overhead of a vtable in class Packet.

The Packet class is designed to be copied cheaply; the overall design is based on Copy on Write (COW). When there are multiple references to a packet object, and there is an operation on one of them, only so-called "dirty" operations will trigger a deep copy of the packet:

• ns3::Packet::AddHeader()
• ns3::Packet::AddTrailer()
• both versions of ns3::Packet::AddAtEnd()
• Packet::RemovePacketTag()

The fundamental classes for adding to and removing from the byte buffer are class Header and class Trailer. Headers are more common but the below discussion also largely applies to protocols using trailers. Every protocol header that needs to be inserted and removed from a Packet instance should derive from the abstract Header base class and implement the private pure virtual methods listed below:

• ns3::Header::SerializeTo()
• ns3::Header::DeserializeFrom()
• ns3::Header::GetSerializedSize()
• ns3::Header::PrintTo()

Basically, the first three functions are used to serialize and deserialize protocol control information to/from a Buffer. For example, one may define class TCPHeader : public Header. The TCPHeader object will typically consist of some private data (like a sequence number) and public interface access functions (such as checking the bounds of an input). But the underlying representation of the TCPHeader in a Packet Buffer is 20 serialized bytes (plus TCP options). The TCPHeader::SerializeTo() function would therefore be designed to write these 20 bytes properly into the packet, in network byte order. The last function is used to define how the Header object prints itself onto an output stream.

Similarly, user-defined Tags can be appended to the packet. Unlike Headers, Tags are not serialized into a contiguous buffer but are stored in lists. Tags can be flexibly defined to be any type, but there can only be one instance of any particular object type in the Tags buffer at any time.

11.2 Using the packet interface

This section describes how to create and use the ns3::Packet object.

11.2.1 Creating a new packet

The following command will create a new packet with a new unique Id.

  Ptr<Packet> pkt = Create<Packet> ();

What is the Uid (unique Id)? It is an internal id that the system uses to identify packets. It can be fetched via the following method:

  uint32_t uid = pkt->GetUid ();

But please note the following. This uid is an internal uid and cannot be counted on to provide an accurate counter of how many "simulated packets" of a particular protocol are in the system. It is not trivial to make this uid into such a counter, because of questions such as what should the uid be when the packet is sent over broadcast media, or when fragmentation occurs. If a user wants to trace actual packet counts, he or she should look at e.g. the IP ID field or transport sequence numbers, or other packet or frame counters at other protocol layers.

We mentioned above that it is possible to create packets with zero-filled payloads that do not actually require a memory allocation (i.e., the packet may behave, when delays such as serialization or transmission delays are computed, to have a certain number of payload bytes, but the bytes will only be allocated on-demand when needed). The command to do this is, when the packet is created:

  Ptr<Packet> pkt = Create<Packet> (N);

where N is a positive integer.

The packet now has a size of N bytes, which can be verified by the GetSize() method:

  /**
   * \returns the size in bytes of the packet (including the zero-filled
   *          initial payload)
   */
  uint32_t GetSize (void) const;

You can also initialize a packet with a character buffer. The input data is copied and the input buffer is untouched. The constructor applied is:

  Packet (uint8_t const *buffer, uint32_t size);

Here is an example:

  Ptr<Packet> pkt1 = Create<Packet> (reinterpret_cast<const uint8_t*> ("hello"), 5);

Packets are freed when there are no more references to them, as with all ns-3 objects referenced by the Ptr class.

11.2.2 Adding and removing Buffer data

After the initial packet creation (which may possibly create some fake initial bytes of payload), all subsequent buffer data is added by adding objects of class Header or class Trailer. Note that, even if you are in the application layer, handling packets, and want to write application data, you write it as an ns3::Header or ns3::Trailer. If you add a Header, it is prepended to the packet, and if you add a Trailer, it is added to the end of the packet. If you have no data in the packet, then it makes no difference whether you add a Header or Trailer. Since the APIs and classes for header and trailer are pretty much identical, we’ll just look at class Header here.

The first step is to create a new header class. All new Header classes must inherit from class Header, and implement the following methods:

• Serialize ()
• Deserialize ()
• GetSerializedSize ()
• Print ()

To see a simple example of how these are done, look at the UdpHeader class headers src/internet-stack/udp-header.cc. There are many other examples within the source code.

Once you have a header (or you have a preexisting header), the following Packet API can be used to add or remove such headers.

  /**
   * Add header to this packet. This method invokes the
   * Header::GetSerializedSize and Header::Serialize
   * methods to reserve space in the buffer and request the 
   * header to serialize itself in the packet buffer.
   *
   * \param header a reference to the header to add to this packet.
   */
  void AddHeader (const Header & header);
  /**
   * Deserialize and remove the header from the internal buffer.
   * This method invokes Header::Deserialize.
   *
   * \param header a reference to the header to remove from the internal buffer.
   * \returns the number of bytes removed from the packet.
   */
  uint32_t RemoveHeader (Header &header);
  /**
   * Deserialize but does _not_ remove the header from the internal buffer.
   * This method invokes Header::Deserialize.
   *
   * \param header a reference to the header to read from the internal buffer.
   * \returns the number of bytes read from the packet.
   */
  uint32_t PeekHeader (Header &header) const;

For instance, here are the typical operations to add and remove a UDP header.

 // add header
 Ptr<Packet> packet = Create<Packet> ();
 UdpHeader udpHeader;
 // Fill out udpHeader fields appropriately
 packet->AddHeader (udpHeader);
 ...
 // remove header
 UdpHeader udpHeader;
 packet->RemoveHeader (udpHeader); 
 // Read udpHeader fields as needed

11.2.3 Adding and removing Tags

There is a single base class of Tag that all packet tags must derive from. They are used in two different tag lists in the packet; the lists have different semantics and different expected use cases.

As the names imply, ByteTags follow bytes and PacketTags follow packets. What this means is that when operations are done on packets, such as fragmentation, concatenation, and appending or removing headers, the byte tags keep track of which packet bytes they cover. For instance, if a user creates a TCP segment, and applies a ByteTag to the segment, each byte of the TCP segment will be tagged. However, if the next layer down inserts an IPv4 header, this ByteTag will not cover those bytes. The converse is true for the PacketTag; it covers a packet despite the operations on it.

PacketTags are limited in size to 20 bytes. This is a modifiable compile-time constant in src/common/packet-tag-list.h. ByteTags have no such restriction.

Each tag type must subclass ns3::Tag, and only one instance of each Tag type may be in each tag list. Here are a few differences in the behavior of packet tags and byte tags.

• Fragmentation: As mentioned above, when a packet is fragmented, each packet fragment (which is a new packet) will get a copy of all packet tags, and byte tags will follow the new packet boundaries (i.e. if the fragmented packets fragment across a buffer region covered by the byte tag, both packet fragments will still have the appropriate buffer regions byte tagged).
• Concatenation: When packets are combined, two different buffer regions will become one. For byte tags, the byte tags simply follow the respective buffer regions. For packet tags, only the tags on the first packet survive the merge.
• Finding and Printing: Both classes allow you to iterate over all of the tags and print them.
• Removal: Users can add and remove the same packet tag multiple times on a single packet (AddPacketTag () and RemovePacketTag ()). The packet However, once a byte tag is added, it can only be removed by stripping all byte tags from the packet. Removing one of possibly multiple byte tags is not supported by the current API.

As of ns-3.5, Tags are not serialized and deserialized to a buffer when Packet::Serialize () and Packet::Deserialize () are called; this is an open bug.

If a user wants to take an existing packet object and reuse it as a new packet, he or she should remove all byte tags and packet tags before doing so. An example is the UdpEchoServer class, which takes the received packet and "turns it around" to send back to the echo client.

The Packet API for byte tags is given below.

  /**
   * \param tag the new tag to add to this packet
   *
   * Tag each byte included in this packet with the
   * new tag.
   *
   * Note that adding a tag is a const operation which is pretty 
   * un-intuitive. The rationale is that the content and behavior of
   * a packet is _not_ changed when a tag is added to a packet: any
   * code which was not aware of the new tag is going to work just
   * the same if the new tag is added. The real reason why adding a
   * tag was made a const operation is to allow a trace sink which gets
   * a packet to tag the packet, even if the packet is const (and most
   * trace sources should use const packets because it would be
   * totally evil to allow a trace sink to modify the content of a
   * packet).
   */
  void AddByteTag (const Tag &tag) const;
  /**
   * \returns an iterator over the set of byte tags included in this packet.
   */
  ByteTagIterator GetByteTagIterator (void) const;
  /**
   * \param tag the tag to search in this packet
   * \returns true if the requested tag type was found, false otherwise.
   *
   * If the requested tag type is found, it is copied in the user's 
   * provided tag instance.
   */
  bool FindFirstMatchingByteTag (Tag &tag) const;
  
  /**
   * Remove all the tags stored in this packet.
   */
  void RemoveAllByteTags (void);

  /**
   * \param os output stream in which the data should be printed.
   *
   * Iterate over the tags present in this packet, and
   * invoke the Print method of each tag stored in the packet.
   */
  void PrintByteTags (std::ostream &os) const;

The Packet API for packet tags is given below.

  /**
   * \param tag the tag to store in this packet
   *
   * Add a tag to this packet. This method calls the
   * Tag::GetSerializedSize and, then, Tag::Serialize.
   *
   * Note that this method is const, that is, it does not
   * modify the state of this packet, which is fairly
   * un-intuitive.
   */
  void AddPacketTag (const Tag &tag) const;
  /**
   * \param tag the tag to remove from this packet
   * \returns true if the requested tag is found, false
   *          otherwise.
   *
   * Remove a tag from this packet. This method calls
   * Tag::Deserialize if the tag is found.
   */
  bool RemovePacketTag (Tag &tag);
  /**
   * \param tag the tag to search in this packet
   * \returns true if the requested tag is found, false
   *          otherwise.
   *
   * Search a matching tag and call Tag::Deserialize if it is found.
   */
  bool PeekPacketTag (Tag &tag) const;
  /**
   * Remove all packet tags.
   */
  void RemoveAllPacketTags (void);
  
  /**
   * \param os the stream in which we want to print data.
   *
   * Print the list of 'packet' tags.
   *
   * \sa Packet::AddPacketTag, Packet::RemovePacketTag, Packet::PeekPacketTag,
   *  Packet::RemoveAllPacketTags
   */
  void PrintPacketTags (std::ostream &os) const;
  
  /**
   * \returns an object which can be used to iterate over the list of
   *  packet tags.
   */
  PacketTagIterator GetPacketTagIterator (void) const;

Here is a simple example illustrating the use of tags from the code in src/internet-stack/udp-socket-impl.cc:

  Ptr<Packet> p;  // pointer to a pre-existing packet
  SocketIpTtlTag tag
  tag.SetTtl (m_ipMulticastTtl); // Convey the TTL from UDP layer to IP layer
  p->AddPacketTag (tag);

This tag is read at the IP layer, then stripped (src/internet-stack/ipv4-l3-protocol.cc:

  uint8_t ttl = m_defaultTtl;
  SocketIpTtlTag tag;
  bool found = packet->RemovePacketTag (tag);
  if (found)
    {
      ttl = tag.GetTtl ();
    }

11.2.4 Fragmentation and concatenation

Packets may be fragmented or merged together. For example, to fragment a packet p of 90 bytes into two packets, one containing the first 10 bytes and the other containing the remaining 80, one may call the following code:

  Ptr<Packet> frag0 = p->CreateFragment (0, 10);
  Ptr<Packet> frag1 = p->CreateFragment (10, 90);

As discussed above, the packet tags from p will follow to both packet fragments, and the byte tags will follow the byte ranges as needed.

Now, to put them back together:

  frag0->AddAtEnd (frag1);

Now frag0 should be equivalent to the original packet p. If, however, there were operations on the fragments before being reassembled (such as tag operations or header operations), the new packet will not be the same.

11.2.5 Enabling metadata

We mentioned above that packets, being on-the-wire representations of byte buffers, present a problem to print out in a structured way unless the printing function has access to the context of the header. For instance, consider a tcpdump-like printer that wants to pretty-print the contents of a packet.

To enable this usage, packets may have metadata enabled (disabled by default for performance reasons). This class is used by the Packet class to record every operation performed on the packet’s buffer, and provides an implementation of Packet::Print () method that uses the metadata to analyze the content of the packet’s buffer.

The metadata is also used to perform extensive sanity checks at runtime when performing operations on a Packet. For example, this metadata is used to verify that when you remove a header from a packet, this same header was actually present at the front of the packet. These errors will be detected and will abort the program.

To enable this operation, users will typically insert one or both of these statements at the beginning of their programs:

  Packet::EnablePrinting ();
  Packet::EnableChecking ();

11.3 Sample programs

See samples/main-packet.cc and samples/main-packet-tag.cc.

11.4 Implementation details

11.4.1 Private member variables

A Packet object’s interface provides access to some private data:

  Buffer m_buffer;
  ByteTagList m_byteTagList;
  PacketTagList m_packetTagList;
  PacketMetadata m_metadata;
  mutable uint32_t m_refCount;
  static uint32_t m_globalUid;

Each Packet has a Buffer and two Tags lists, a PacketMetadata object, and a ref count. A static member variable keeps track of the UIDs allocated. The actual uid of the packet is stored in the PacketMetadata.

Note that real network packets do not have a UID; the UID is therefore an instance of data that normally would be stored as a Tag in the packet. However, it was felt that a UID is a special case that is so often used in simulations that it would be more convenient to store it in a member variable.

11.4.2 Buffer implementation

Class Buffer represents a buffer of bytes. Its size is automatically adjusted to hold any data prepended or appended by the user. Its implementation is optimized to ensure that the number of buffer resizes is minimized, by creating new Buffers of the maximum size ever used. The correct maximum size is learned at runtime during use by recording the maximum size of each packet.

Authors of new Header or Trailer classes need to know the public API of the Buffer class. (add summary here)

The byte buffer is implemented as follows:

    struct BufferData {
        uint32_t m_count;
        uint32_t m_size;
        uint32_t m_initialStart;
        uint32_t m_dirtyStart;
        uint32_t m_dirtySize;
        uint8_t m_data[1];
    }; 
    struct BufferData *m_data;
    uint32_t m_zeroAreaSize;
    uint32_t m_start;
    uint32_t m_size;

• BufferData::m_count: reference count for BufferData structure
• BufferData::m_size: size of data buffer stored in BufferData structure
• BufferData::m_initialStart: offset from start of data buffer where data was first inserted
• BufferData::m_dirtyStart: offset from start of buffer where every Buffer which holds a reference to this BufferData instance have written data so far
• BufferData::m_dirtySize: size of area where data has been written so far
• BufferData::m_data: pointer to data buffer
• Buffer::m_zeroAreaSize: size of zero area which extends before m_initialStart
• Buffer::m_start: offset from start of buffer to area used by this buffer
• Buffer::m_size: size of area used by this Buffer in its BufferData structure

Figure 11.2: Implementation overview of a packet’s byte Buffer.

This data structure is summarized in Figure fig:buffer. Each Buffer holds a pointer to an instance of a BufferData. Most Buffers should be able to share the same underlying BufferData and thus simply increase the BufferData’s reference count. If they have to change the content of a BufferData inside the Dirty Area, and if the reference count is not one, they first create a copy of the BufferData and then complete their state-changing operation.

11.4.3 Tags implementation

(XXX revise me)

Tags are implemented by a single pointer which points to the start of a linked list ofTagData data structures. Each TagData structure points to the next TagData in the list (its next pointer contains zero to indicate the end of the linked list). Each TagData contains an integer unique id which identifies the type of the tag stored in the TagData.

struct TagData {
    struct TagData *m_next;
    uint32_t m_id;
    uint32_t m_count;
    uint8_t m_data[Tags::SIZE];
};
class Tags {        
    struct TagData *m_next;
};

Adding a tag is a matter of inserting a new TagData at the head of the linked list. Looking at a tag requires you to find the relevant TagData in the linked list and copy its data into the user data structure. Removing a tag and updating the content of a tag requires a deep copy of the linked list before performing this operation. On the other hand, copying a Packet and its tags is a matter of copying the TagData head pointer and incrementing its reference count.

Tags are found by the unique mapping between the Tag type and its underlying id. This is why at most one instance of any Tag can be stored in a packet. The mapping between Tag type and underlying id is performed by a registration as follows:

/* A sample Tag implementation
 */
struct MyTag {
    uint16_t m_streamId;
};

11.4.4 Memory management

Describe dataless vs. data-full packets.

11.4.5 Copy-on-write semantics

The current implementation of the byte buffers and tag list is based on COW (Copy On Write). An introduction to COW can be found in Scott Meyer’s "More Effective C++", items 17 and 29). This design feature and aspects of the public interface borrows from the packet design of the Georgia Tech Network Simulator. This implementation of COW uses a customized reference counting smart pointer class.

What COW means is that copying packets without modifying them is very cheap (in terms of CPU and memory usage) and modifying them can be also very cheap. What is key for proper COW implementations is being able to detect when a given modification of the state of a packet triggers a full copy of the data prior to the modification: COW systems need to detect when an operation is “dirty” and must therefore invoke a true copy.

Dirty operations:

• ns3::Packet::AddHeader
• ns3::Packet::AddTrailer
• both versions of ns3::Packet::AddAtEnd
• ns3::Packet::RemovePacketTag

Non-dirty operations:

• ns3::Packet::AddPacketTag
• ns3::Packet::PeekPacketTag
• ns3::Packet::RemoveAllPacketTags
• ns3::Packet::AddByteTag
• ns3::Packet::FindFirstMatchingByteTag
• ns3::Packet::RemoveAllByteTags
• ns3::Packet::RemoveHeader
• ns3::Packet::RemoveTrailer
• ns3::Packet::CreateFragment
• ns3::Packet::RemoveAtStart
• ns3::Packet::RemoveAtEnd
• ns3::Packet::CopyData

Dirty operations will always be slower than non-dirty operations, sometimes by several orders of magnitude. However, even the dirty operations have been optimized for common use-cases which means that most of the time, these operations will not trigger data copies and will thus be still very fast.

12. Helpers

The above chapters introduced you to various ns-3 programming concepts such as smart pointers for reference-counted memory management, attributes, namespaces, callbacks, etc. Users who work at this low-level API can interconnect ns-3 objects with fine granulariy. However, a simulation program written entirely using the low-level API would be quite long and tedious to code. For this reason, a separate so-called “helper API” has been overlaid on the core ns-3 API. If you have read the ns-3 tutorial, you will already be familiar with the helper API, since it is the API that new users are typically introduced to first. In this chapter, we introduce the design philosophy of the helper API and contrast it to the low-level API. If you become a heavy user of ns-3, you will likely move back and forth between these APIs even in the same program.

The helper API has a few goals:

1. the rest of src/ has no dependencies on the helper API; anything that can be done with the helper API can be coded also at the low-level API
2. Containers: Often simulations will need to do a number of identical actions to groups of objects. The helper API makes heavy use of containers of similar objects to which similar or identical operations can be performed.
3. The helper API is not generic; it does not strive to maximize code reuse. So, programming constructs such as polymorphism and templates that achieve code reuse are not as prevalent. For instance, there are separate CsmaNetDevice helpers and PointToPointNetDevice helpers but they do not derive from a common NetDevice base class.
4. The helper API typically works with stack-allocated (vs. heap-allocated) objects. For some programs, ns-3 users may not need to worry about any low level Object Create or Ptr handling; they can make do with containers of objects and stack-allocated helpers that operate on them.

The helper API is really all about making ns-3 programs easier to write and read, without taking away the power of the low-level interface. The rest of this chapter provides some examples of the programming conventions of the helper API.

13. Python

For now, please see the Python wiki page at
http://www.nsnam.org/wiki/index.php/NS-3_Python_Bindings.

Part 2: Nodes and NetDevices

14. Node and NetDevices

This chapter describes how ns-3 nodes are put together, and provides a walk-through of how packets traverse an internet-based Node.

Figure 14.1: High-level node architecture.

In ns-3, nodes are instances of class Node. This class may be subclassed, but instead, the conceptual model is that we aggregate or insert objects to it rather than define subclasses.

One might think of a bare ns-3 node as a shell of a computer, to which one may add NetDevices (cards) and other innards including the protocols and applications. fig:node illustrates that Node objects contain a list of Applications (initially, the list is empty), a list of NetDevices (initially, the list is empty), a list of ProtocolHandlers, a unique integer ID, and a system ID (for distributed simulation).

The design tries to avoid putting too many dependencies on the base class Node, Application, or NetDevice for the following:

• IP version, or whether IP is at all even used in the Node.
• implementation details of the IP stack

From a software perspective, the lower interface of applications corresponds to the C-based sockets API. The upper interface of NetDevice objects corresponds to the device independent sublayer of the Linux stack. Everything in between can be aggregated and plumbed together as needed.

Let’s look more closely at the protocol demultiplexer. We want incoming frames at layer-2 to be delivered to the right layer-3 protocol such as Ipv4. The function of this demultiplexer is to register callbacks for receiving packets. The callbacks are indexed based on the EtherType in the layer-2 frame.

Many different types of higher-layer protocols may be connected to the NetDevice, such as IPv4, IPv6, ARP, MPLS, IEEE 802.1x, and packet sockets. Therefore, the use of a callback-based demultiplexer avoids the need to use a common base class for all of these protocols, which is problematic because of the different types of objects (including packet sockets) expected to be registered there.

Each NetDevice delivers packets to a callback with the following signature:

  /**
   * \param device a pointer to the net device which is calling this callback
   * \param packet the packet received
   * \param protocol the 16 bit protocol number associated with this packet.
   *        This protocol number is expected to be the same protocol number
   *        given to the Send method by the user on the sender side.
   * \param address the address of the sender
   * \returns true if the callback could handle the packet successfully, 
   *        false otherwise.
   */
  typedef Callback<bool, Ptr<NetDevice>, Ptr<Packet>, uint16_t,
    const Address &> ReceiveCallback;

There is a function in class Node that matches that signature:

private:
  bool ReceiveFromDevice (Ptr<NetDevice> device, Ptr<Packet>,
                          uint16_t protocol, const Address &from);

However, users do not need to access this function directly. Instead, when users call uint32_t AddDevice (Ptr<NetDevice> device), the implementation of this function sets the callback (and the function returns the ifIndex of the NetDevice on that Node).

But what does the ReceiveFromDevice function do? Here, it looks up another callback, in its list of callbacks, corresponding to the matching EtherType. This callback is called a ProtocolHandler, and is specified as follows:

  typedef Callback<void, Ptr<NetDevice>, Ptr<Packet>, uint16_t,
    const Address &> ProtocolHandler;

Upper-layer protocols or objects are expected to provide such a function. and register it with the list of ProtocolHandlers by calling Node::RegisterProtocolHandler (); For instance, if Ipv4 is aggregated to a Node, then the Ipv4 receive function can be registered with the protocol handler by calling:

  RegisterProtocolHandler (
    MakeCallback (&Ipv4L3Protocol::Receive, ipv4), 
    Ipv4L3Protocol::PROT_NUMBER, 0);

and likewise for Ipv6, ARP, etc.

14.1 NodeList

Every Node created is automatically added to the ns-3 NodeList. The NodeList class provides an Add() method and C++ iterators to allow one to walk the node list or fetch a Node pointer by its integer identifier.

The following chapters provide reference information on the available NetDevices in ns-3.

15. Simple NetDevice

16. PointToPoint NetDevice

This is the introduction to PointToPoint NetDevice chapter, to complement the PointToPoint model doxygen.

16.1 Overview of the PointToPoint model

The ns-3 point-to-point model is of a very simple point to point data link connecting exactly two PointToPointNetDevice devices over an PointToPointChannel. This can be viewed as equivalent to a full duplex RS-232 or RS-422 link with null modem and no handshaking.

Data is encapsulated in the Point-to-Point Protocol (PPP – RFC 1661), however the Link Control Protocol (LCP) and associated state machine is not implemented. The PPP link is assumed to be established and authenticated at all times.

Data is not framed, therefore Address and Control fields will not be found. Since the data is not framed, there is no need to provide Flag Sequence and Control Escape octets, nor is a Frame Check Sequence appended. All that is required to implement non-framed PPP is to prepend the PPP protocol number for IP Version 4 which is the sixteen-bit number 0x21 (see http://www.iana.org/assignments/ppp-numbers).

The PointToPointNetDevice provides following Attributes:

• Address: The ns3::Mac48Address of the device (if desired);
• DataRate: The data rate (ns3::DataRate) of the device;
• TxQueue: The transmit queue (ns3::Queue) used by the device;
• InterframeGap: The optional ns3::Time to wait between "frames";
• Rx: A trace source for received packets;
• Drop: A trace source for dropped packets.

The PointToPointNetDevice models a transmitter section that puts bits on a corresponding channel "wire." ‘The DataRate attribute specifies the number of bits per second that the device will simulate sending over the channel. In reality no bits are sent, but an event is scheduled for an elapsed time consistent with the number of bits in each packet and the specified DataRate. The implication here is that the receiving device models a receiver section that can receive any any data rate. Therefore there is no need, nor way to set a receive data rate in this model. By setting the DataRate on the transmitter of both devices connected to a given PointToPointChannel one can model a symmetric channel; or by setting different DataRates one can model an asymmetric channel (e.g., ADSL).

The PointToPointNetDevice supports the assignment of a "receive error model." This is an ErrorModel object that is used to simulate data corruption on the link.

16.2 Point-to-Point Channel Model

The point to point net devices are connected via an PointToPointChannel. This channel models two wires transmitting bits at the data rate specified by the source net device. There is no overhead beyond the eight bits per byte of the packet sent. That is, we do not model Flag Sequences, Frame Check Sequences nor do we "escape" any data.

The PointToPointChannel provides following Attributes:

• Delay: An ns3::Time specifying the speed of light transmission delay for the channel.

16.3 Using the PointToPointNetDevice

The PointToPoint net devices and channels are typically created and configured using the associated PointToPointHelper object. The various ns3 device helpers generally work in a similar way, and their use is seen in many of our example programs and is also covered in the ns-3 tutorial.

The conceptual model of interest is that of a bare computer “husk” into which you plug net devices. The bare computers are created using a NodeContainer helper. You just ask this helper to create as many computers (we call them Nodes) as you need on your network:

  NodeContainer nodes;
  nodes.Create (2);

Once you have your nodes, you need to instantiate a PointToPointHelper and set any attributes you may want to change. Note that since this is a point-to-point (as compared to a point-to-multipoint) there may only be two nodes with associated net devices connected by a PointToPointChannel.

  PointToPointHelper pointToPoint;
  pointToPoint.SetDeviceAttribute ("DataRate", StringValue ("5Mbps"));
  pointToPoint.SetChannelAttribute ("Delay", StringValue ("2ms"));

Once the attributes are set, all that remains is to create the devices and install them on the required nodes, and to connect the devices together using a PointToPoint channel. When we create the net devices, we add them to a container to allow you to use them in the future. This all takes just one line of code.

  NetDeviceContainer devices = pointToPoint.Install (nodes);

16.4 PointToPoint Tracing

Like all ns-3 devices, the Point-to-Point Model provides a number of trace sources. These trace sources can be hooked using your own custom trace code, or you can use our helper functions to arrange for tracing to be enabled on devices you specify.

16.4.1 Upper-Level (MAC) Hooks

From the point of view of tracing in the net device, there are several interesting points to insert trace hooks. A convention inherited from other simulators is that packets destined for transmission onto attached networks pass through a single "transmit queue" in the net device. We provide trace hooks at this point in packet flow, which corresponds (abstractly) only to a transition from the network to data link layer, and call them collectively the device MAC hooks.

When a packet is sent to the Point-to-Point net device for transmission it always passes through the transmit queue. The transmit queue in the PointToPointNetDevice inherits from Queue, and therefore inherits three trace sources:

• An Enqueue operation source (see ns3::Queue::m_traceEnqueue);
• A Dequeue operation source (see ns3::Queue::m_traceDequeue);
• A Drop operation source (see ns3::Queue::m_traceDrop).

The upper-level (MAC) trace hooks for the PointToPointNetDevice are, in fact, exactly these three trace sources on the single transmit queue of the device.

The m_traceEnqueue event is triggered when a packet is placed on the transmit queue. This happens at the time that ns3::PointtoPointNetDevice::Send or ns3::PointToPointNetDevice::SendFrom is called by a higher layer to queue a packet for transmission. An Enqueue trace event firing should be interpreted as only indicating that a higher level protocol has sent a packet to the device.

The m_traceDequeue event is triggered when a packet is removed from the transmit queue. Dequeues from the transmit queue can happen in two situations: 1) If the underlying channel is idle when PointToPointNetDevice::Send is called, a packet is dequeued from the transmit queue and immediately transmitted; 2) a packet may be dequeued and immediately transmitted in an internal TransmitCompleteEvent that functions much like a transmit complete interrupt service routine. An Dequeue trace event firing may be viewed as indicating that the PointToPointNetDevice has begun transmitting a packet.

16.4.2 Lower-Level (PHY) Hooks

Similar to the upper level trace hooks, there are trace hooks available at the lower levels of the net device. We call these the PHY hooks. These events fire from the device methods that talk directly to the PointToPointChannel.

The trace source m_dropTrace is called to indicate a packet that is dropped by the device. This happens when a packet is discarded as corrupt due to a receive error model indication (see ns3::ErrorModel and the associated attribute "ReceiveErrorModel").

The other low-level trace source fires on reception of a packet (see ns3::PointToPointNetDevice::m_rxTrace) from the PointToPointChannel.

17. CSMA NetDevice

This is the introduction to CSMA NetDevice chapter, to complement the Csma model doxygen.

17.1 Overview of the CSMA model

The ns-3 CSMA device models a simple bus network in the spirit of Ethernet. Although it does not model any real physical network you could ever build or buy, it does provide some very useful functionality.

Typically when one thinks of a bus network Ethernet or IEEE 802.3 comes to mind. Ethernet uses CSMA/CD (Carrier Sense Multiple Access with Collision Detection with exponentially increasing backoff to contend for the shared transmission medium. The ns-3 CSMA device models only a portion of this process, using the nature of the globally available channel to provide instantaneous (faster than light) carrier sense and priority-based collision "avoidance." Collisions in the sense of Ethernet never happen and so the ns-3 CSMA device does not model collision detection, nor will any transmission in progress be "jammed."

17.1.1 CSMA Layer Model

There are a number of conventions in use for describing layered communications architectures in the literature and in textbooks. The most common layering model is the ISO seven layer reference model. In this view the CsmaNetDevice and CsmaChannel pair occupies the lowest two layers – at the physical (layer one), and data link (layer two) positions. Another important reference model is that specified by RFC 1122, "Requirements for Internet Hosts – Communication Layers." In this view the CsmaNetDevice and CsmaChannel pair occupies the lowest layer – the link layer. There is also a seemingly endless litany of alternative descriptions found in textbooks and in the literature. We adopt the naming conventions used in the IEEE 802 standards which speak of LLC, MAC, MII and PHY layering. These acronyms are defined as:

• LLC: Logical Link Control;
• MAC: Media Access Control;
• MII: Media Independent Interface;
• PHY: Physical Layer.

In this case the LLC and MAC are sublayers of the OSI data link layer and the MII and PHY are sublayers of the OSI physical layer.

The "top" of the CSMA device defines the transition from the network layer to the data link layer. This transition is performed by higher layers by calling either CsmaNetDevice::Send or CsmaNetDevice::SendFrom.

In contrast to the IEEE 802.3 standards, there is no precisely specified PHY in the CSMA model in the sense of wire types, signals or pinouts. The "bottom" interface of the CsmaNetDevice can be thought of as as a kind of Media Independent Interface (MII) as seen in the "Fast Ethernet" (IEEE 802.3u) specifications. This MII interface fits into a corresponding media independent interface on the CsmaChannel. You will not find the equivalent of a 10BASE-T or a 1000BASE-LX PHY.

The CsmaNetDevice calls the CsmaChannel through a media independent interface. There is a method defined to tell the channel when to start "wiggling the wires" using the method CsmaChannel::TransmitStart, and a method to tell the channel when the transmission process is done and the channel should begin propagating the last bit across the "wire": CsmaChannel::TransmitEnd.

When the TransmitEnd method is executed, the channel will model a single uniform signal propagation delay in the medium and deliver copes of the packet to each of the devices attached to the packet via the CsmaNetDevice::Receive method.

There is a "pin" in the device media independent interface corresponding to "COL" (collision). The state of the channel may be sensed by calling CsmaChannel::GetState. Each device will look at this "pin" before starting a send and will perform appropriate backoff operations if required.

Properly received packets are forwarded up to higher levels from the CsmaNetDevice via a callback mechanism. The callback function is initialized by the higher layer (when the net device is attached) using CsmaNetDevice::SetReceiveCallback and is invoked upon "proper" reception of a packet by the net device in order to forward the packet up the protocol stack.

17.2 CSMA Channel Model

The class CsmaChannel models the actual transmission medium. There is no fixed limit for the number of devices connected to the channel. The CsmaChannel models a data rate and a speed-of-light delay which can be accessed via the attributes "DataRate" and "Delay" respectively. The data rate provided to the channel is used to set the data rates used by the transmitter sections of the CSMA devices connected to the channel. There is no way to independently set data rates in the devices. Since the data rate is only used to calculate a delay time, there is no limitation (other than by the data type holding the value) on the speed at which CSMA channels and devices can operate; and no restriction based on any kind of PHY characteristics.

The CsmaChannel has three states, IDLE, TRANSMITTING and PROPAGATING. These three states are "seen" instantaneously by all devices on the channel. By this we mean that if one device begins or ends a simulated transmission, all devices on the channel are immediately aware of the change in state. There is no time during which one device may see an IDLE channel while another device physically further away in the collision domain may have begun transmitting with the associated signals not propagated down the channel to other devices. Thus there is no need for collision detection in the CsmaChannel model and it is not implemented in any way.

We do, as the name indicates, have a Carrier Sense aspect to the model. Since the simulator is single threaded, access to the common channel will be serialized by the simulator. This provides a deterministic mechanism for contending for the channel. The channel is allocated (transitioned from state IDLE to state TRANSMITTING) on a first-come first-served basis. The channel always goes through a three state process:

  IDLE -> TRANSMITTING -> PROPAGATING -> IDLE

The TRANSMITTING state models the time during which the source net device is actually wiggling the signals on the wire. The PROPAGATING state models the time after the last bit was sent, when the signal is propagating down the wire to the "far end."

The transition to the TRANSMITTING state is driven by a call to CsmaChannel::TransmitStart which is called by the net device that transmits the packet. It is the responsibility of that device to end the transmission with a call to CsmaChannel::TransmitEnd at the appropriate simulation time that reflects the time elapsed to put all of the packet bits on the wire. When TransmitEnd is called, the channel schedules an event corresponding to a single speed-of-light delay. This delay applies to all net devices on the channel identically. You can think of a symmetrical hub in which the packet bits propagate to a central location and then back out equal length cables to the other devices on the channel. The single “speed of light” delay then corresponds to the time it takes for: 1) a signal to propagate from one CsmaNetDevice through its cable to the hub; plus 2) the time it takes for the hub to forward the packet out a port; plus 3) the time it takes for the signal in question to propagate to the destination net device.

The CsmaChannel models a broadcast medium so the packet is delivered to all of the devices on the channel (including the source) at the end of the propagation time. It is the responsibility of the sending device to determine whether or not it receives a packet broadcast over the channel.

The CsmaChannel provides following Attributes:

• DataRate: The bitrate for packet transmission on connected devices;
• Delay: The speed of light transmission delay for the channel.

17.3 CSMA Net Device Model

The CSMA network device appears somewhat like an Ethernet device. The CsmaNetDevice provides following Attributes:

• Address: The Mac48Address of the device;
• SendEnable: Enable packet transmission if true;
• ReceiveEnable: Enable packet reception if true;
• EncapsulationMode: Type of link layer encapsulation to use;
• RxErrorModel: The receive error model;
• TxQueue: The transmit queue used by the device;
• InterframeGap: The optional time to wait between "frames";
• Rx: A trace source for received packets;
• Drop: A trace source for dropped packets.

The CsmaNetDevice supports the assignment of a "receive error model." This is an ErrorModel object that is used to simulate data corruption on the link.

Packets sent over the CsmaNetDevice are always routed through the transmit queue to provide a trace hook for packets sent out over the network. This transmit queue can be set (via attribute) to model different queuing strategies.

Also configurable by attribute is the encapsulation method used by the device. Every packet gets an EthernetHeader that includes the destination and source MAC addresses, and a length/type field. Every packet also gets an EthernetTrailer which includes the FCS. Data in the packet may be encapsulated in different ways.

By default, or by setting the "EncapsulationMode" attribute to "Dix", the encapsulation is according to the DEC, Intel, Xerox standard. This is sometimes called EthernetII framing and is the familiar destination MAC, source MAC, EtherType, Data, CRC format.

If the "EncapsulationMode" attribute is set to "Llc", the encapsulation is by LLC SNAP. In this case, a SNAP header is added that contains the EtherType (IP or ARP).

The other implemented encapsulation modes are IP_ARP (set "EncapsulationMode" to "IpArp") in which the length type of the Ethernet header receives the protocol number of the packet; or ETHERNET_V1 (set "EncapsulationMode" to "EthernetV1") in which the length type of the Ethernet header receives the length of the packet. A "Raw" encapsulation mode is defined but not implemented – use of the RAW mode results in an assertion.

Note that all net devices on a channel must be set to the same encapsulation mode for correct results. The encapsulation mode is not sensed at the receiver.

The CsmaNetDevice implements a random exponential backoff algorithm that is executed if the channel is determined to be busy (TRANSMITTING or PPROPAGATING) when the device wants to start propagating. This results in a random delay of up to pow (2, retries) - 1 microseconds before a retry is attempted. The default maximum number of retries is 1000.

17.4 Using the CsmaNetDevice

The CSMA net devices and channels are typically created and configured using the associated CsmaHelper object. The various ns3 device helpers generally work in a similar way, and their use is seen in many of our example programs.

The conceptual model of interest is that of a bare computer “husk” into which you plug net devices. The bare computers are created using a NodeContainer helper. You just ask this helper to create as many computers (we call them Nodes) as you need on your network:

  NodeContainer csmaNodes;
  csmaNodes.Create (nCsmaNodes);

Once you have your nodes, you need to instantiate a CsmaHelper and set any attributes you may want to change.

  CsmaHelper csma;
  csma.SetChannelAttribute ("DataRate", StringValue ("100Mbps"));
  csma.SetChannelAttribute ("Delay", TimeValue (NanoSeconds (6560)));

  csma.SetDeviceAttribute ("EncapsulationMode", StringValue ("Dix"));
  csma.SetDeviceAttribute ("FrameSize", UintegerValue (2000));

Once the attributes are set, all that remains is to create the devices and install them on the required nodes, and to connect the devices together using a CSMA channel. When we create the net devices, we add them to a container to allow you to use them in the future. This all takes just one line of code.

  NetDeviceContainer csmaDevices = csma.Install (csmaNodes);

17.5 CSMA Tracing

Like all ns-3 devices, the CSMA Model provides a number of trace sources. These trace sources can be hooked using your own custom trace code, or you can use our helper functions to arrange for tracing to be enabled on devices you specify.

17.5.1 Upper-Level (MAC) Hooks

From the point of view of tracing in the net device, there are several interesting points to insert trace hooks. A convention inherited from other simulators is that packets destined for transmission onto attached networks pass through a single "transmit queue" in the net device. We provide trace hooks at this point in packet flow, which corresponds (abstractly) only to a transition from the network to data link layer, and call them collectively the device MAC hooks.

When a packet is sent to the CSMA net device for transmission it always passes through the transmit queue. The transmit queue in the CsmaNetDevice inherits from Queue, and therefore inherits three trace sources:

• An Enqueue operation source (see Queue::m_traceEnqueue);
• A Dequeue operation source (see Queue::m_traceDequeue);
• A Drop operation source (see Queue::m_traceDrop).

The upper-level (MAC) trace hooks for the CsmaNetDevice are, in fact, exactly these three trace sources on the single transmit queue of the device.

The m_traceEnqueue event is triggered when a packet is placed on the transmit queue. This happens at the time that CsmaNetDevice::Send or CsmaNetDevice::SendFrom is called by a higher layer to queue a packet for transmission.

The m_traceDequeue event is triggered when a packet is removed from the transmit queue. Dequeues from the transmit queue can happen in three situations: 1) If the underlying channel is idle when the CsmaNetDevice::Send or CsmaNetDevice::SendFrom is called, a packet is dequeued from the transmit queue and immediately transmitted; 2) If the underlying channel is idle, a packet may be dequeued and immediately transmitted in an internal TransmitCompleteEvent that functions much like a transmit complete interrupt service routine; or 3) from the random exponential backoff handler if a timeout is detected.

Case (3) implies that a packet is dequeued from the transmit queue if it is unable to be transmitted according to the backoff rules. It is important to understand that this will appear as a Dequeued packet and it is easy to incorrectly assume that the packet was transmitted since it passed through the transmit queue. In fact, a packet is actually dropped by the net device in this case. The reason for this behavior is due to the definition of the Queue Drop event. The m_traceDrop event is, by definition, fired when a packet cannot be enqueued on the transmit queue because it is full. This event only fires if the queue is full and we do not overload this event to indicate that the CsmaChannel is "full."

17.5.2 Lower-Level (PHY) Hooks

Similar to the upper level trace hooks, there are trace hooks available at the lower levels of the net device. We call these the PHY hooks. These events fire from the device methods that talk directly to the CsmaChannel.

The trace source m_dropTrace is called to indicate a packet that is dropped by the device. This happens in two cases: First, if the receive side of the net device is not enabled (see CsmaNetDevice::m_receiveEnable and the associated attribute "ReceiveEnable").

The m_dropTrace is also used to indicate that a packet was discarded as corrupt if a receive error model is used (see CsmaNetDevice::m_receiveErrorModel and the associated attribute "ReceiveErrorModel").

The other low-level trace source fires on reception of an accepted packet (see CsmaNetDevice::m_rxTrace). A packet is accepted if it is destined for the broadcast address, a multicast address, or to the MAC address assigned to the net device.

18. Wifi NetDevice

ns-3 nodes can contain a collection of NetDevice objects, much like an actual computer contains separate interface cards for Ethernet, Wifi, Bluetooth, etc. This chapter describes the ns-3 WifiNetDevice and related models. By adding WifiNetDevice objects to ns-3 nodes, one can create models of 802.11-based infrastructure and ad hoc networks.

18.1 Overview of the model

The WifiNetDevice models a wireless network interface controller based on the IEEE 802.11 standard. We will go into more detail below but in brief, ns-3 provides models for these aspects of 802.11:

• basic 802.11 DCF with infrastructure and adhoc modes
• 802.11a and 802.11b physical layers
• QoS-based EDCA and queueing extensions of 802.11e
• various propagation loss models including Nakagami, Rayleigh, Friis, LogDistance, FixedRss, Random
• two propagation delay models, a distance-based and random model
• various rate control algorithms including Aarf, Arf, Cara, Onoe, Rraa, ConstantRate, and Minstrel
• 802.11s (mesh), described in another chapter

The set of 802.11 models provided in ns-3 attempts to provide an accurate MAC-level implementation of the 802.11 specification and to provide a not-so-slow PHY-level model of the 802.11a specification.

The implementation is modular and provides roughly four levels of models:

• the PHY layer models
• the so-called MAC low models: they implement DCF and EDCAF
• the so-called MAC high models: they implement the MAC-level beacon generation, probing, and association state machines, and
• a set of Rate control algorithms used by the MAC low models

There are presently six MAC high models, three for non-QoS MACs and three for QoS MACs.

• non-QoS MACs:
  1. a simple adhoc state machine that does not perform any kind of beacon generation, probing, or association. This state machine is implemented by the ns3::AdhocWifiMac class.
  2. an active probing and association state machine that handles automatic re-association whenever too many beacons are missed is implemented by the ns3::NqstaWifiMac class.
  3. an access point that generates periodic beacons, and that accepts every attempt to associate. This AP state machine is implemented by the ns3::NqapWifiMac class.
• QoS MACs:
  1. a simple adhoc state machine like above but also able to manage QoS traffic. This state machine is implemented by ns3::QadhocWifiMac class.
  2. a station state machine like above but also able to manage QoS traffic. Implemented by ns3::QstaWifiMac.
  3. a QoS access point state machine like above implemented by ns3::QapWifiMac.

With QoS MAC models is possible to work with traffic belonging to four different access classes: AC_VO for voice traffic, AC_VI for video traffic, AC_BE for best-effort traffic and AC_BK for background traffic. In order to determine MSDU’s access class, every packet forwarded down to these MAC layers should be marked using ns3::QosTag in order to set a TID (traffic id) for that packet otherwise it will be considered belonging to AC_BE access class.

The MAC low layer is split into three components:

1. ns3::MacLow which takes care of RTS/CTS/DATA/ACK transactions.
2. ns3::DcfManager and ns3::DcfState which implements the DCF and EDCAF functions.
3. ns3::DcaTxop or ns3::EdcaTxopN which handle the packet queue, packet fragmentation, and packet retransmissions if they are needed. ns3::DcaTxop object is used by non-QoS high MACs. ns3::EdcaTxopN is used by QoS high MACs and performs also QoS operations like 802.11n MSDU aggregation.

There are also several rate control algorithms that can be used by the Mac low layer:

• ns3::ArfMacStations
• ns3::AArfMacStations
• ns3::IdealMacStations
• ns3::CrMacStations
• ns3::OnoeMacStations
• ns3::AmrrMacStations

The PHY layer implements a single model in the ns3::WifiPhy class: the physical layer model implemented there is described fully in a paper entitled "Yet Another Network Simulator". Validation results for 802.11b are available in this technical report.

In ns-3, nodes can have multiple WifiNetDevices on separate channels, and the WifiNetDevice can coexist with other device types; this removes an architectural limitation found in ns-2. Presently, however, there is no model for cross-channel interference or coupling.

The source code for the Wifi NetDevice lives in the directory src/devices/wifi.

Figure 18.1: Wifi NetDevice architecture.

18.2 Using the WifiNetDevice

The modularity provided by the implementation makes low-level configuration of the WifiNetDevice powerful but complex. For this reason, we provide some helper classes to perform common operations in a simple matter, and leverage the ns-3 attribute system to allow users to control the parametrization of the underlying models.

Users who use the low-level ns-3 API and who wish to add a WifiNetDevice to their node must create an instance of a WifiNetDevice, plus a number of constituent objects, and bind them together appropriately (the WifiNetDevice is very modular in this regard, for future extensibility). At the low-level API, this can be done with about 20 lines of code (see ns3::WifiHelper::Install, and ns3::YansWifiPhyHelper::Create). They also must create, at some point, a WifiChannel, which also contains a number of constituent objects (see ns3::YansWifiChannelHelper::Create).

However, a few helpers are available for users to add these devices and channels with only a few lines of code, if they are willing to use defaults, and the helpers provide additional API to allow the passing of attribute values to change default values. The scripts in src/examples can be browsed to see how this is done.

18.2.1 YansWifiChannelHelper

The YansWifiChannelHelper has an unusual name. Readers may wonder why it is named this way. The reference is to the yans simulator, from which this model is taken. The helper can be used to create a WifiChannel with a default PropagationLoss and PropagationDelay model. Specifically, the default is a channel model with a propagation delay equal to a constant, the speed of light, and a propagation loss based on a log distance model with a reference loss of 46.6777 dB at reference distance of 1m.

Users will typically type code such as:

  YansWifiChannelHelper wifiChannelHelper = YansWifiChannelHelper::Default ();
  Ptr<WifiChannel> wifiChannel = wifiChannelHelper.Create ();

to get the defaults. Note the distinction above in creating a helper object vs. an actual simulation object. In ns-3, helper objects (used at the helper API only) are created on the stack (they could also be created with operator new and later deleted). However, the actual ns-3 objects typically inherit from class ns3::Object and are assigned to a smart pointer. See the chapter on Object model for a discussion of the ns-3 object model, if you are not familiar with it.

Todo: Add notes about how to configure attributes with this helper API

18.2.2 YansWifiPhyHelper

Physical devices (base class ns3::Phy) connect to ns3::Channel models in ns-3. We need to create Phy objects appropriate for the YansWifiChannel; here the YansWifiPhyHelper will do the work.

The YansWifiPhyHelper class configures an object factory to create instances of a YansWifiPhy and adds some other objects to it, including possibly a supplemental ErrorRateModel and a pointer to a MobilityModel. The user code is typically:

  YansWifiPhyHelper wifiPhyHelper = YansWifiPhyHelper::Default ();
  wifiPhyHelper.SetChannel (wifiChannel);

Note that we haven’t actually created any WifiPhy objects yet; we’ve just prepared the YansWifiPhyHelper by telling it which channel it is connected to. The phy objects are created in the next step.

18.2.3 NqosWifiMacHelper and QosWifiMacHelper

The ns3::NqosWifiMacHelper and ns3::QosWifiMacHelper configure an object factory to create instances of a ns3::WifiMac. They are used to configure MAC parameters like type of MAC, values of contention windows and so on. Setting up a non-QoS MAC layers the object we use is ns3::NqosWifiMacHelper. For example the following user code configures a non-QoS MAC sta and changes its default values for contention window and Aifsn:

  NqosWifiMacHelper wifiMacHelper = NqosWifiMacHelper::Default ();
  Ssid ssid = Ssid ("ns-3-ssid");
  wifiMacHelper.SetType ("ns3::NqstaWifiMac", "Ssid", SsidValue (ssid), 
"ActiveProbing", BooleanValue (false));
  wifiMacHelper.SetDcaParameters ("MinCw", UintegerValue (20), "Aifsn", 
UintegerValue (3));

Setting up a QoS MACs we use a ns3::QosWifiMacHelper instead. This object could be also used to change default EDCA parameters, and to set a possible MSDU aggregator for a particular access class in order to use 802.11n MSDU aggregation feature. A possible user code:

  QosWifiMacHelper wifiMacHelper = QosWifiMacHelper::Default ();
  wifiMacHelper.SetType ("ns3::QapWifiMac", 
    "Ssid", SsidValue (ssid), 
    "BeaconGeneration", BooleanValue (true), 
    "BeaconInterval", TimeValue (Seconds (2.5)));
  wifiMacHelper.SetMsduAggregatorForAc (AC_VO, "ns3::MsduStandardAggregator", 
    "MaxAmsduSize", UintegerValue (3839));

Call to QosWifiMacHelper::Default () is needed in order to set default EDCA parameters properly for all access classes; otherwise we should set them one by one.

18.2.4 WifiHelper

We’re now ready to create WifiNetDevices. First, let’s create a WifiHelper with default settings:

  WifiHelper wifiHelper = WifiHelper::Default ();

What does this do? It sets the RemoteStationManager to ns3::ArfWifiManager. Now, let’s use the wifiPhyHelper and wifiMacHelper created above to install WifiNetDevices on a set of nodes in a NodeContainer "c":

  NetDeviceContainer wifiContainer = WifiHelper::Install (wifiPhyHelper, wifiMacHelper, c);

This creates the WifiNetDevice which includes also a WifiRemoteStationManager, a WifiMac, and a WifiPhy (connected to the matching WifiChannel).

There are many ns-3 Attributes that can be set on the above helpers to deviate from the default behavior; the example scripts show how to do some of this reconfiguration.

18.2.5 AdHoc WifiNetDevice configuration

This is a typical example of how a user might configure an adhoc network.

18.2.6 Infrastructure (Access Point and clients) WifiNetDevice configuration

This is a typical example of how a user might configure an access point and a set of clients.

18.3 The WifiChannel and WifiPhy models

The WifiChannel subclass can be used to connect together a set of ns3::WifiNetDevice network interfaces. The class ns3::WifiPhy is the object within the WifiNetDevice that receives bits from the channel. A WifiChannel contains a ns3::PropagationLossModel and a ns3::PropagationDelayModel which can be overridden by the WifiChannel::SetPropagationLossModel and the WifiChannel::SetPropagationDelayModel methods. By default, no propagation models are set.

The WifiPhy models an 802.11a channel, in terms of frequency, modulation, and bit rates, and interacts with the PropagationLossModel and PropagationDelayModel found in the channel.

This section summarizes the description of the BER calculations found in the yans paper taking into account the Forward Error Correction present in 802.11a and describes the algorithm we implemented to decide whether or not a packet can be successfully received. See "Yet Another Network Simulator" for more details.

The PHY layer can be in one of three states:

1. TX: the PHY is currently transmitting a signal on behalf of its associated MAC
2. RX: the PHY is synchronized on a signal and is waiting until it has received its last bit to forward it to the MAC.
3. IDLE: the PHY is not in the TX or RX states.

When the first bit of a new packet is received while the PHY is not IDLE (that is, it is already synchronized on the reception of another earlier packet or it is sending data itself), the received packet is dropped. Otherwise, if the PHY is IDLE, we calculate the received energy of the first bit of this new signal and compare it against our Energy Detection threshold (as defined by the Clear Channel Assessment function mode 1). If the energy of the packet k is higher, then the PHY moves to RX state and schedules an event when the last bit of the packet is expected to be received. Otherwise, the PHY stays in IDLE state and drops the packet.

The energy of the received signal is assumed to be zero outside of the reception interval of packet k and is calculated from the transmission power with a path-loss propagation model in the reception interval. where the path loss exponent, n, is chosen equal to 3, the reference distance, d_0 is choosen equal to 1.0m and the reference energy is based based on a Friis propagation model.

When the last bit of the packet upon which the PHY is synchronized is received, we need to calculate the probability that the packet is received with any error to decide whether or not the packet on which we were synchronized could be successfully received or not: a random number is drawn from a uniform distribution and is compared against the probability of error.

To evaluate the probability of error, we start from the piecewise linear functions shown in Figure fig:snir and calculate the SNIR function.

Figure 18.2: SNIR function over time

From the SNIR function we can derive bit error rates for BPSK and QAM modulations. Then, for each interval l where BER is constant, we define the upper bound of a probability that an error is present in the chunk of bits located in the interval l for packet k. If we assume an AWGN channel, binary convolutional coding (which is the case in 802.11a) and hard-decision Viterbi decoding, the error rate is thus derived, and the packet error probability for packet k can be computed..

18.3.1 WifiChannel configuration

WifiChannel models include both a PropagationDelayModel and a PropagationLossModel. The following PropagationDelayModels are available:

• ConstantSpeedPropagationDelayModel
• RandomPropagationDelayModel

The following PropagationLossModels are available:

• RandomPropagationLossModel
• FriisPropagationLossModel
• LogDistancePropagationLossModel
• JakesPropagationLossModel
• CompositePropagationLossModel

18.4 The MAC model

The 802.11 Distributed Coordination Function is used to calculate when to grant access to the transmission medium. While implementing the DCF would have been particularly easy if we had used a recurring timer that expired every slot, we chose to use the method described in (missing reference here from Yans paper) where the backoff timer duration is lazily calculated whenever needed since it is claimed to have much better performance than the simpler recurring timer solution.

The higher-level MAC functions are implemented in a set of other C++ classes and deal with:

• packet fragmentation and defragmentation,
• use of the rts/cts protocol,
• rate control algorithm,
• connection and disconnection to and from an Access Point,
• the MAC transmission queue,
• beacon generation,
• msdu aggregation,
• etc.

18.5 Wifi Attributes

The WifiNetDevice makes heavy use of the ns-3 Attributes subsystem for configuration and default value management. Presently, approximately 100 values are stored in this system.

For instance, class ns-3::WifiMac exports these attributes:

• CtsTimeout: When this timeout expires, the RTS/CTS handshake has failed.
• AckTimeout: When this timeout expires, the DATA/ACK handshake has failed.
• Sifs: The value of the SIFS constant.
• EifsNoDifs: The value of EIFS-DIFS
• Slot: The duration of a Slot.
• Pifs: The value of the PIFS constant.
• MaxPropagationDelay: The maximum propagation delay. Unused for now.
• MaxMsduSize: The maximum size of an MSDU accepted by the MAC layer.This value conforms to the specification.
• Ssid: The ssid we want to belong to.

18.6 Wifi Tracing

This needs revised/updating based on the latest Doxygen

ns-3 has a sophisticated tracing infrastructure that allows users to hook into existing trace sources, or to define and export new ones.

Wifi-related trace sources that are available by default include:

• ns3::WifiNetDevice
  • Rx: Received payload from the MAC layer.
  • Tx: Send payload to the MAC layer.
• ns3::WifiPhy
  • State: The WifiPhy state
  • RxOk: A packet has been received successfully.
  • RxError: A packet has been received unsuccessfully.
  • Tx: Packet transmission is starting.

Briefly, this means, for example, that a user can hook a processing function to the "State" tracing hook above and be notified whenever the WifiPhy model changes state.

19. Mesh NetDevice

The Mesh NetDevice based on 802.11s was added in ns-3.6. An overview presentation by Kirill Andreev was published at the wns-3 workshop in 2009:
http://www.nsnam.org/wiki/index.php/Wns3-2009

20. Bridge NetDevice

Some examples of the use of Bridge NetDevice can be found in examples/csma/ directory.

Part 3: Emulation

21. Emulation

ns-3 has been designed for integration into testbed and virtual machine environments. We have addressed this need by providing two kinds of net devices. The first kind, which we call an Emu NetDevice allows ns-3 simulations to send data on a “real” network. The second kind, called a Tap NetDevice allows a “real” host to participate in an ns-3 simulation as if it were one of the simulated nodes. An ns-3 simulation may be constructed with any combination of simulated, Emu, or Tap devices.

One of the use-cases we want to support is that of a testbed. A concrete example of an environment of this kind is the ORBIT testbed. ORBIT is a laboratory emulator/field trial network arranged as a two dimensional grid of 400 802.11 radio nodes. We integrate with ORBIT by using their “imaging” process to load and run ns-3 simulations on the ORBIT array. We use our Emu NetDevices to drive the hardware in the testbed and we can accumulate results either using the ns-3 tracing and logging functions, or the native ORBIT data gathering techniques. See http://www.orbit-lab.org/ for details on the ORBIT testbed.

A simulation of this kind is shown in the following figure:

Figure 21.1: Example Implementation of Testbed Emulation.

You can see that there are separate hosts, each running a subset of a “global” simulation. Instead of an ns-3 channel connecting the hosts, we use real hardware provided by the testbed. This allows ns-3 applications and protocol stacks attached to a simulation node to communicate over real hardware.

We expect the primary use for this configuration will be to generate repeatable experimental results in a real-world network environment that includes all of the ns-3 tracing, logging, visualization and statistics gathering tools.

In what can be viewed as essentially an inverse configuration, we allow “real” machines running native applications and protocol stacks to integrate with an ns-3 simulation. This allows for the simulation of large networks connected to a real machine, and also enables virtualization. A simulation of this kind is shown in the following figure:

Figure 21.2: Implementation overview of emulated channel.

Here, you will see that there is a single host with a number of virtual machines running on it. An ns-3 simulation is shown running in the virtual machine shown in the center of the figure. This simulation has a number of nodes with associated ns-3 applications and protocol stacks that are talking to an ns-3 channel through native simulated ns-3 net devices.

There are also two virtual machines shown at the far left and far right of the figure. These VMs are running native (Linux) applications and protocol stacks. The VM is connected into the simulation by a Linux Tap net device. The user-mode handler for the Tap device is instantiated in the simulation and attached to a proxy node that represents the native VM in the simulation. These handlers allow the Tap devices on the native VMs to behave as if they were ns-3 net devices in the simulation VM. This, in turn, allows the native software and protocol suites in the native VMs to believe that they are connected to the simulated ns-3 channel.

We expect the typical use case for this environment will be to analyze the behavior of native applications and protocol suites in the presence of large simulated ns-3 networks.

22. Emu NetDevice

22.1 Behavior

The Emu net device allows a simulation node to send and receive packets over a real network. The emulated net device relies on a specified interface being in promiscuous mode. It opens a raw socket and binds to that interface. We perform MAC spoofing to separate simulation network traffic from other network traffic that may be flowing to and from the host.

One can use the Emu net device in a testbed situation where the host on which the simulation is running has a specific interface of interest which drives the testbed hardware. You would also need to set this specific interface into promiscuous mode and provide an appropriate device name to the ns-3 emulated net device. An example of this environment is the ORBIT testbed as described above.

The Emu net device only works if the underlying interface is up and in promiscuous mode. Packets will be sent out over the device, but we use MAC spoofing. The MAC addresses will be generated (by default) using the Organizationally Unique Identifier (OUI) 00:00:00 as a base. This vendor code is not assigned to any organization and so should not conflict with any real hardware.

It is always up to the user to determine that using these MAC addresses is okay on your network and won’t conflict with anything else (including another simulation using Emu devices) on your network. If you are using the emulated net device in separate simulations you must consider global MAC address assignment issues and ensure that MAC addresses are unique across all simulations. The emulated net device respects the MAC address provided in the SetAddress method so you can do this manually. For larger simulations, you may want to set the OUI in the MAC address allocation function.

IP addresses corresponding to the emulated net devices are the addresses generated in the simulation, which are generated in the usual way via helper functions. Since we are using MAC spoofing, there will not be a conflict between ns-3 network stacks and any native network stacks.

The emulated net device comes with a helper function as all ns-3 devices do. One unique aspect is that there is no channel associated with the underlying medium. We really have no idea what this external medium is, and so have not made an effort to model it abstractly. The primary thing to be aware of is the implication this has for IPv4 global routing. The global router module attempts to walk the channels looking for adjacent networks. Since there is no channel, the global router will be unable to do this and you must then use a dynamic routing protocol such as OLSR to include routing in Emu-based networks.

22.2 Usage

Any mixing of ns-3 objects with real objects will typically require that ns-3 compute checksums in its protocols. By default, checksums are not computed by ns-3. To enable checksums (e.g. UDP, TCP, IP), users must set the attribute ChecksumEnabled to true, such as follows:

GlobalValue::Bind ("ChecksumEnabled", BooleanValue (true));

The usage of the Emu net device is straightforward once the network of simulations has been configured. Since most of the work involved in working with this device is in network configuration before even starting a simulation, you may want to take a moment to review a couple of HOWTO pages on the ns-3 wiki that describe how to set up a virtual test network using VMware and how to run a set of example (client server) simulations that use Emu net devices.

• http://www.nsnam.org/wiki/index.php/HOWTO_use_VMware_to_set_up_virtual_networks_(Windows)
• http://www.nsnam.org/wiki/index.php/HOWTO_use_ns-3_scripts_to_drive_real_hardware_(experimental)

Once you are over the configuration hurdle, the script changes required to use an Emu device are trivial. The main structural difference is that you will need to create an ns-3 simulation script for each node. In the case of the HOWTOs above, there is one client script and one server script. The only “challenge” is to get the addresses set correctly.

Just as with all other ns-3 net devices, we provide a helper class for the Emu net device. The following code snippet illustrates how one would declare an EmuHelper and use it to set the “DeviceName” attribute to “eth1” and install Emu devices on a group of nodes. You would do this on both the client and server side in the case of the HOWTO seen above.

  EmuHelper emu;
  emu.SetAttribute ("DeviceName", StringValue ("eth1"));
  NetDeviceContainer d = emu.Install (n);

The only other change that may be required is to make sure that the address spaces (MAC and IP) on the client and server simulations are compatible. First the MAC address is set to a unique well-known value in both places (illustrated here for one side).

  //
  // We've got the devices in place.  Since we're using MAC address 
  // spoofing under the sheets, we need to make sure that the MAC addresses
  // we have assigned to our devices are unique.  Ns-3 will happily
  // automatically assign the same MAC address to the devices in both halves
  // of our two-script pair, so let's go ahead and just manually change them
  // to something we ensure is unique.
  //
  Ptr<NetDevice> nd = d.Get (0);
  Ptr<EmuNetDevice> ed = nd->GetObject<EmuNetDevice> ();
  ed->SetAddress ("00:00:00:00:00:02");

And then the IP address of the client or server is set in the usual way using helpers.

  //
  // We've got the "hardware" in place.  Now we need to add IP addresses.
  // This is the server half of a two-script pair.  We need to make sure
  // that the addressing in both of these applications is consistent, so
  // we use provide an initial address in both cases.  Here, the client 
  // will reside on one machine running ns-3 with one node having ns-3
  // with IP address "10.1.1.2" and talk to a server script running in 
  // another ns-3 on another computer that has an ns-3 node with IP 
  // address "10.1.1.3"
  //
  Ipv4AddressHelper ipv4;
  ipv4.SetBase ("10.1.1.0", "255.255.255.0", "0.0.0.2");
  Ipv4InterfaceContainer i = ipv4.Assign (d);

You will use application helpers to generate traffic exactly as you do in any ns-3 simulation script. Note that the server address shown below in a snippet from the client, must correspond to the IP address assigned to the server node similarly to the snippet above.

  uint32_t packetSize = 1024;
  uint32_t maxPacketCount = 2000;
  Time interPacketInterval = Seconds (0.001);
  UdpEchoClientHelper client ("10.1.1.3", 9);
  client.SetAttribute ("MaxPackets", UintegerValue (maxPacketCount));
  client.SetAttribute ("Interval", TimeValue (interPacketInterval));
  client.SetAttribute ("PacketSize", UintegerValue (packetSize));
  ApplicationContainer apps = client.Install (n.Get (0));
  apps.Start (Seconds (1.0));
  apps.Stop (Seconds (2.0));

The Emu net device and helper provide access to ASCII and pcap tracing functionality just as other ns-3 net devices to. You enable tracing similarly to these other net devices:

  EmuHelper::EnablePcapAll ("emu-udp-echo-client");

To see an example of a client script using the Emu net device, see examples/emu-udp-echo-client.cc and examples/emu-udp-echo-server.cc in the repository http://code.nsnam.org/craigdo/ns-3-emu/.

22.3 Implementation

Perhaps the most unusual part of the Emu and Tap device implementation relates to the requirement for executing some of the code with super-user permissions. Rather than force the user to execute the entire simulation as root, we provide a small “creator” program that runs as root and does any required high-permission sockets work.

We do a similar thing for both the Emu and the Tap devices. The high-level view is that the CreateSocket method creates a local interprocess (Unix) socket, forks, and executes the small creation program. The small program, which runs as suid root, creates a raw socket and sends back the raw socket file descriptor over the Unix socket that is passed to it as a parameter. The raw socket is passed as a control message (sometimes called ancillary data) of type SCM_RIGHTS.

The Emu net device uses the ns-3 threading and multithreaded real-time scheduler extensions. The interesting work in the Emu device is done when the net device is started (EmuNetDevice::StartDevice ()). An attribute (“Start”) provides a simulation time at which to spin up the net device. At this specified time (which defaults to t=0), the socket creation function is called and executes as described above. You may also specify a time at which to stop the device using the “Stop” attribute.

Once the (promiscuous mode) socket is created, we bind it to an interface name also provided as an attribute (“DeviceName”) that is stored internally as m_deviceName:

  struct ifreq ifr;
  bzero (&ifr, sizeof(ifr));
  strncpy ((char *)ifr.ifr_name, m_deviceName.c_str (), IFNAMSIZ);

  int32_t rc = ioctl (m_sock, SIOCGIFINDEX, &ifr);

  struct sockaddr_ll ll;
  bzero (&ll, sizeof(ll));

  ll.sll_family = AF_PACKET;
  ll.sll_ifindex = m_sll_ifindex;
  ll.sll_protocol = htons(ETH_P_ALL);

  rc = bind (m_sock, (struct sockaddr *)&ll, sizeof (ll));

After the promiscuous raw socket is set up, a separate thread is spawned to do reads from that socket and the link state is set to Up.

  m_readThread = Create<SystemThread> (
    MakeCallback (&EmuNetDevice::ReadThread, this));
  m_readThread->Start ();

  NotifyLinkUp ();

The EmuNetDevice::ReadThread function basically just sits in an infinite loop reading from the promiscuous mode raw socket and scheduling packet receptions using the real-time simulator extensions.

  for (;;)
    {
      ...

      len = recvfrom (m_sock, buf, bufferSize, 0, (struct sockaddr *)&addr, 
        &addrSize);

      ...

      DynamicCast<RealtimeSimulatorImpl> (Simulator::GetImplementation ())->
        ScheduleRealtimeNow (
          MakeEvent (&EmuNetDevice::ForwardUp, this, buf, len));

      ...
    }

The line starting with our templated DynamicCast function probably deserves a comment. It gains access to the simulator implementation object using the Simulator::GetImplementation method and then casts to the real-time simulator implementation to use the real-time schedule method ScheduleRealtimeNow. This function will cause a handler for the newly received packet to be scheduled for execution at the current real time clock value. This will, in turn cause the simulation clock to be advanced to that real time value when the scheduled event (EmuNetDevice::ForwardUp) is fired.

The ForwardUp function operates as most other similar ns-3 net device methods do. The packet is first filtered based on the destination address. In the case of the Emu device, the MAC destination address will be the address of the Emu device and not the hardware address of the real device. Headers are then stripped off and the trace hooks are hit. Finally, the packet is passed up the ns-3 protocol stack using the receive callback function of the net device.

Sending a packet is equally straightforward as shown below. The first thing we do is to add the ethernet header and trailer to the ns-3 Packet we are sending. The source address corresponds to the address of the Emu device and not the underlying native device MAC address. This is where the MAC address spoofing is done. The trailer is added and we enqueue and dequeue the packet from the net device queue to hit the trace hooks.

  header.SetSource (source);
  header.SetDestination (destination);
  header.SetLengthType (packet->GetSize ());
  packet->AddHeader (header);

  EthernetTrailer trailer;
  trailer.CalcFcs (packet);
  packet->AddTrailer (trailer);

  m_queue->Enqueue (packet);
  packet = m_queue->Dequeue ();

  struct sockaddr_ll ll;
  bzero (&ll, sizeof (ll));

  ll.sll_family = AF_PACKET;
  ll.sll_ifindex = m_sll_ifindex;
  ll.sll_protocol = htons(ETH_P_ALL);

  rc = sendto (m_sock, packet->PeekData (), packet->GetSize (), 0, 
    reinterpret_cast<struct sockaddr *> (&ll), sizeof (ll));

Finally, we simply send the packet to the raw socket which puts it out on the real network.

23. Tap NetDevice

The Tap NetDevice can be used to allow a host system or virtual machines to interact with a simulation. See
examples/tap/tap-wifi-dumbbell.cc for an example.

Part 4: Internet Models

24. Sockets APIs

The sockets API is a long-standing API used by user-space applications to access network services in the kernel. A “socket” is an abstraction, like a Unix file handle, that allows applications to connect to other Internet hosts and exchange reliable byte streams and unreliable datagrams, among other services.

ns-3 provides two types of sockets APIs, and it is important to understand the differences between them. The first is a native ns-3 API, while the second uses the services of the native API to provide a POSIX-like API as part of an overall application process. Both APIs strive to be close to the typical sockets API that application writers on Unix systems are accustomed to, but the POSIX variant is much closer to a real system’s sockets API.

24.1 ns-3 sockets API

The native sockets API for ns-3 provides an interface to various types of transport protocols (TCP, UDP) as well as to packet sockets and, in the future, Netlink-like sockets. However, users are cautioned to understand that the semantics are not the exact same as one finds in a real system (for an API which is very much aligned to real systems, see the next section).

class ns3::Socket is defined in src/node/socket.cc,h. Readers will note that many public member functions are aligned with real sockets function calls, and all other things being equal, we have tried to align with a Posix sockets API. However, note that:

• ns-3 applications handle a smart pointer to a Socket object, not a file descriptor;
• there is no notion of synchronous API or a “blocking” API; in fact, the model for interaction between application and socket is one of asynchronous I/O, which is not typically found in real systems (more on this below);
• the C-style socket address structures are not used;
• the API is not a complete sockets API, such as supporting all socket options or all function variants;
• many calls use ns3::Packet class to transfer data between application and socket. This may seem peculiar to pass “Packets” across a stream socket API, but think of these packets as just fancy byte buffers at this level (more on this also below).

24.1.1 Basic operation and calls

Figure 24.1: Implementation overview of native sockets API

24.1.1.1 Creating sockets

An application that wants to use sockets must first create one. On real systems using a C-based API, this is accomplished by calling socket():

     int
     socket(int domain, int type, int protocol);

which creates a socket in the system and returns an integer descriptor.

In ns-3, we have no equivalent of a system call at the lower layers, so we adopt the following model. There are certain factory objects that can create sockets. Each factory is capable of creating one type of socket, and if sockets of a particular type are able to be created on a given node, then a factory that can create such sockets must be aggregated to the Node.

  static Ptr<Socket> CreateSocket (Ptr<Node> node, TypeId tid);

Examples of TypeIds to pass to this method are TcpSocketFactory, PacketSocketFactory, and UdpSocketFactory.

This method returns a smart pointer to a Socket object. Here is an example:

  Ptr<Node> n0;
  // Do some stuff to build up the Node's internet stack
  Ptr<Socket> localSocket = Socket::CreateSocket (n0, TcpSocketFactory::GetTypeId ());

In some ns-3 code, sockets will not be explicitly created by user’s main programs, if an ns-3 application does it. For instance, for class ns3::OnOffApplication, the function StartApplication() performs the socket creation, and the application holds the socket pointer.

24.1.1.2 Using sockets

Below is a typical sequence of socket calls for a TCP client in a real implementation:

• sock = socket(PF_INET, SOCK_STREAM, IPPROTO_TCP);
• bind(sock, ...);
• connect(sock, ...);
• send(sock, ...);
• recv(sock, ...);
• close(sock);

There are analogs to all of these calls in ns-3, but we will focus on two aspects here. First, most usage of sockets in real systems requires a way to manage I/O between the application and kernel. These models include blocking sockets, signal-based I/O, and non-blocking sockets with polling. In ns-3, we make use of the callback mechanisms to support a fourth mode, which is analogous to POSIX asynchronous I/O.

In this model, on the sending side, if the send() call were to fail because of insufficient buffers, the application suspends the sending of more data until a function registered at the SetSendCallback() callback is invoked. An application can also ask the socket how much space is available by calling GetTxAvailable (). A typical sequence of events for sending data (ignoring connection setup) might be:

• SetSendCallback (MakeCallback(&HandleSendCallback));
• Send ();
• Send ();
• ...
• // Send fails because buffer is full
• (wait until HandleSendCallback() is called)
• (HandleSendCallback() is called by socket, since space now available)
• Send (); // Start sending again

Similarly, on the receive side, the socket user does not block on a call to recv(). Instead, the application sets a callback with SetRecvCallback () in which the socket will notify the application when (and how much) there is data to be read, and the application then calls Recv() to read the data until no more can be read.

24.1.2 Packet vs. buffer variants

There are two basic variants of Send() and Recv() supported:

  virtual int Send (Ptr<Packet> p) = 0;
  int Send (const uint8_t* buf, uint32_t size);

  Ptr<Packet> Recv (void);
  int Recv (uint8_t* buf, uint32_t size);

The non-Packet variants are provided for legacy API reasons. When calling the raw buffer variant of Send(), the buffer is immediately written into a Packet and the Send (Ptr<Packet> p) is invoked.

Users may find it semantically odd to pass a Packet to a stream socket such as TCP. However, do not let the name bother you; think of ns3::Packet to be a fancy byte buffer. There are a few reasons why the Packet variants are more likely to be preferred in ns-3:

• Users can use the Tags facility of packets to, for example, encode a flow ID or other helper data at the application layer.
• Users can exploit the copy-on-write implementation to avoid memory copies (on the receive side, the conversion back to a uint8_t* buf may sometimes incur an additional copy).
• Use of Packet is more aligned with the rest of the ns-3 API

24.1.3 Sending dummy data

Sometimes, users want the simulator to just pretend that there is an actual data payload in the packet (e.g. to calculate transmission delay) but do not want to actually produce or consume the data. This is straightforward to support in ns-3; have applications call Create<Packet> (size); instead of Create<Packet> (buffer, size);. Similarly, passing in a zero to the pointer argument in the raw buffer variants has the same effect. Note that, if some subsequent code tries to read the Packet data buffer, the fake buffer will be converted to a real (zeroed) buffer on the spot, and the efficiency will be lost there.

24.1.4 Socket options

to be completed

24.1.5 Socket errno

to be completed

24.1.6 Example programs

to be completed

24.2 POSIX-like sockets API

25. Internet Stack

25.1 Internet stack aggregation

A bare class Node is not very useful as-is; other objects must be aggregated to it to provide useful node functionality.

The ns-3 source code directory src/internet-stack provides implementation of TCP/IPv4- and IPv6-related components. These include IPv4, ARP, UDP, TCP, IPv6, Neighbor Discovery, and other related protocols.

Internet Nodes are not subclasses of class Node; they are simply Nodes that have had a bunch of IPv4-related objects aggregated to them. They can be put together by hand, or via a helper function InternetStackHelper::Install () which does the following to all nodes passed in as arguments:

void
InternetStackHelper::Install (Ptr<Node> node) const
{
  if (node->GetObject<Ipv4> () != 0)
    {
      NS_FATAL_ERROR ("InternetStackHelper::Install(): Aggregating "
                      "an InternetStack to a node with an existing Ipv4 object");
      return;
    

  CreateAndAggregateObjectFromTypeId (node, "ns3::ArpL3Protocol");
  CreateAndAggregateObjectFromTypeId (node, "ns3::Ipv4L3Protocol");
  CreateAndAggregateObjectFromTypeId (node, "ns3::Icmpv4L4Protocol");
  CreateAndAggregateObjectFromTypeId (node, "ns3::UdpL4Protocol");
  node->AggregateObject (m_tcpFactory.Create<Object> ());
  Ptr<PacketSocketFactory> factory = CreateObject<PacketSocketFactory> ();
  node->AggregateObject (factory);
  // Set routing
  Ptr<Ipv4> ipv4 = node->GetObject<Ipv4> ();
  Ptr<Ipv4RoutingProtocol> ipv4Routing = m_routing->Create (node);
  ipv4->SetRoutingProtocol (ipv4Routing);

Where multiple implementations exist in ns-3 (TCP, IP routing), these objects are added by a factory object (TCP) or by a routing helper (m_routing).

Note that the routing protocol is configured and set outside this function. By default, the following protocols are added to Ipv4:

InternetStackHelper::InternetStackHelper ()
{
  SetTcp ("ns3::TcpL4Protocol");
  static Ipv4StaticRoutingHelper staticRouting;
  static Ipv4GlobalRoutingHelper globalRouting;
  static Ipv4ListRoutingHelper listRouting;
  listRouting.Add (staticRouting, 0);
  listRouting.Add (globalRouting, -10);
  SetRoutingHelper (listRouting);
}

By default, IPv4 and IPv6 are enabled.

25.1.1 Internet Node structure

An IPv4-capable Node (an ns-3 Node augmented by aggregation to have one or more IP stacks) has the following internal structure.

25.1.1.1 Layer-3 protocols

At the lowest layer, sitting above the NetDevices, are the "layer 3" protocols, including IPv4, IPv6 (in the future), and ARP. The class Ipv4L3Protocol is an implementation class whose public interface is typically class Ipv4 (found in src/node directory), but the Ipv4L3Protocol public API is also used internally in the src/internet-stack directory at present.

In class Ipv4L3Protocol, one method described below is Receive ():

 /**
   * Lower layer calls this method after calling L3Demux::Lookup
   * The ARP subclass needs to know from which NetDevice this
   * packet is coming to:
   *    - implement a per-NetDevice ARP cache
   *    - send back arp replies on the right device
   */
  void Receive( Ptr<NetDevice> device, Ptr<const Packet> p, uint16_t protocol, 
const Address &from, const Address &to, NetDevice::PacketType packetType);

First, note that the Receive () function has a matching signature to the ReceiveCallback in the class Node. This function pointer is inserted into the Node’s protocol handler when AddInterface () is called. The actual registration is done with a statement such as: follows:

 RegisterProtocolHandler ( MakeCallback (&Ipv4Protocol::Receive, ipv4),
    Ipv4L3Protocol::PROT_NUMBER, 0);

The Ipv4L3Protocol object is aggregated to the Node; there is only one such Ipv4L3Protocol object. Higher-layer protocols that have a packet to send down to the Ipv4L3Protocol object can call GetObject<Ipv4L3Protocol> () to obtain a pointer, as follows:

  Ptr<Ipv4L3Protocol> ipv4 = m_node->GetObject<Ipv4L3Protocol> ();
  if (ipv4 != 0)
    {
      ipv4->Send (packet, saddr, daddr, PROT_NUMBER);
    }

This class nicely demonstrates two techniques we exploit in ns-3 to bind objects together: callbacks, and object aggregation.

Once IPv4 routing has determined that a packet is for the local node, it forwards it up the stack. This is done with the following function:

void
Ipv4L3Protocol::LocalDeliver (Ptr<const Packet> packet, Ipv4Header const&ip, uint32_t iif)

The first step is to find the right Ipv4L4Protocol object , based on IP protocol number. For instance, TCP is registered in the demux as protocol number 6. Finally, the Receive() function on the Ipv4L4Protocol (such as TcpL4Protocol::Receive is called.

We have not yet introduced the class Ipv4Interface. Basically, each NetDevice is paired with an IPv4 representation of such device. In Linux, this class Ipv4Interface roughly corresponds to the struct in_device; the main purpose is to provide address-family specific information (addresses) about an interface.

The IPv6 implementation follows a similar architecture.

25.1.1.2 Layer-4 protocols and sockets

We next describe how the transport protocols, sockets, and applications tie together. In summary, each transport protocol implementation is a socket factory. An application that needs a new socket

For instance, to create a UDP socket, an application would use a code snippet such as the following:

      Ptr<Udp> udpSocketFactory = GetNode ()->GetObject<Udp> ();
      Ptr<Socket> m_socket = socketFactory->CreateSocket ();
      m_socket->Bind (m_local_address);
      ...

The above will query the node to get a pointer to its UDP socket factory, will create one such socket, and will use the socket with an API similar to the C-based sockets API, such as Connect () and Send (). See the chapter on ns-3 sockets for more information.

We have described so far a socket factory (e.g. class Udp) and a socket, which may be specialized (e.g., class UdpSocket). There are a few more key objects that relate to the specialized task of demultiplexing a packet to one or more receiving sockets. The key object in this task is class Ipv4EndPointDemux. This demultiplexer stores objects of class Ipv4EndPoint. This class holds the addressing/port tuple (local port, local address, destination port, destination address) associated with the socket, and a receive callback. This receive callback has a receive function registered by the socket. The Lookup () function to Ipv4EndPointDemux returns a list of Ipv4EndPoint objects (there may be a list since more than one socket may match the packet). The layer-4 protocol copies the packet to each Ipv4EndPoint and calls its ForwardUp () method, which then calls the Receive () function registered by the socket.

An issue that arises when working with the sockets API on real systems is the need to manage the reading from a socket, using some type of I/O (e.g., blocking, non-blocking, asynchronous, ...). ns-3 implements an asynchronous model for socket I/O; the application sets a callback to be notified of received data ready to be read, and the callback is invoked by the transport protocol when data is available. This callback is specified as follows:

  void Socket::SetRecvCallback (Callback<void, Ptr<Socket>, 
    Ptr<Packet>, const Address&> receivedData);

The data being received is conveyed in the Packet data buffer. An example usage is in class PacketSink:

  m_socket->SetRecvCallback (MakeCallback(&PacketSink::HandleRead, this));

To summarize, internally, the UDP implementation is organized as follows:

• a UdpImpl class that implements the UDP socket factory functionality
• a UdpL4Protocol class that implements the protocol logic that is socket-independent
• a UdpSocketImpl class that implements socket-specific aspects of UDP
• a class called Ipv4EndPoint that stores the addressing tuple (local port, local address, destination port, destination address) associated with the socket, and a receive callback for the socket.

25.1.2 Ipv4-capable node interfaces

Many of the implementation details, or internal objects themselves, of Ipv4-capable Node objects are not exposed at the simulator public API. This allows for different implementations; for instance, replacing the native ns-3 models with ported TCP/IP stack code.

The C++ public APIs of all of these objects is found in the src/node directory, including principally:

• socket.h
• tcp.h
• udp.h
• ipv4.h

These are typically base class objects that implement the default values used in the implementation, implement access methods to get/set state variables, host attributes, and implement publicly-available methods exposed to clients such as CreateSocket.

25.1.3 Example path of a packet

These two figures show an example stack trace of how packets flow through the Internet Node objects.

Figure 25.1: Send path of a packet.

Figure 25.2: Receive path of a packet.

26. IPv4

27. IPv6

IPv6 models are being added to ns-3. A paper on the IPv6 models was published in WNS2 2008:
http://lsiit.u-strasbg.fr/Publications/2008/VMM08/

28. Routing overview

ns-3 is intended to support traditional routing approaches and protocols, support ports of open source routing implementations, and facilitate research into unorthodox routing techniques. The overall routing architecture is described below in Routing architecture. Users who wish to just read about how to configure global routing for wired topologies can read Global centralized routing. Unicast routing protocols are described in Unicast routing. Multicast routing is documented in Multicast routing.

28.1 Routing architecture

Figure 28.1: Overview of routing

fig:routing shows the overall routing architecture for Ipv4. The key objects are Ipv4L3Protocol, Ipv4RoutingProtocol(s) (a class to which all routing/forwarding has been delegated from Ipv4L3Protocol), and Ipv4Route(s).

Ipv4L3Protocol must have at least one Ipv4RoutingProtocol added to it at simulation setup time. This is done explicitly by calling Ipv4::SetRoutingProtocol ().

The abstract base class Ipv4RoutingProtocol () declares a minimal interface, consisting of two methods: RouteOutput () and RouteInput (). For packets traveling outbound from a host, the transport protocol will query Ipv4 for the Ipv4RoutingProtocol object interface, and will request a route via Ipv4RoutingProtocol::RouteOutput (). A Ptr to Ipv4Route object is returned. This is analagous to a dst_cache entry in Linux. The Ipv4Route is carried down to the Ipv4L3Protocol to avoid a second lookup there. However, some cases (e.g. Ipv4 raw sockets) will require a call to RouteOutput() directly from Ipv4L3Protocol.

For packets received inbound for forwarding or delivery, the following steps occur. Ipv4L3Protocol::Receive() calls Ipv4RoutingProtocol::RouteInput(). This passes the packet ownership to the Ipv4RoutingProtocol object. There are four callbacks associated with this call:

• LocalDeliver
• UnicastForward
• MulticastForward
• Error

The Ipv4RoutingProtocol must eventually call one of these callbacks for each packet that it takes responsibility for. This is basically how the input routing process works in Linux.

Figure 28.2: Ipv4Routing specialization

This overall architecture is designed to support different routing approaches, including (in the future) a Linux-like policy-based routing implementation, proactive and on-demand routing protocols, and simple routing protocols for when the simulation user does not really care about routing.

fig:routing-specialization illustrates how multiple routing protocols derive from this base class. A class Ipv4ListRouting (implementation class Ipv4ListRoutingImpl) provides the existing list routing approach in ns-3. Its API is the same as base class Ipv4Routing except for the ability to add multiple prioritized routing protocols (Ipv4ListRouting::AddRoutingProtocol(), Ipv4ListRouting::GetRoutingProtocol()).

The details of these routing protocols are described below in Unicast routing. For now, we will first start with a basic unicast routing capability that is intended to globally build routing tables at simulation time t=0 for simulation users who do not care about dynamic routing.