@c ============================================================================
@c Begin document body here
@c ============================================================================

@c ============================================================================
@c PART:  The Tracing System
@c ============================================================================
@c The below chapters are under the major heading "The Tracing System"
@c This is similar to the Latex \part command
@c
@c ============================================================================
@c The Tracing System
@c ============================================================================
@node The Tracing System
@chapter The Tracing System

@menu
* Background::
* Overview::
@end menu

@c ============================================================================
@c Background
@c ============================================================================
@node Background
@section Background

As mentioned in the Using the Tracing System section, the whole point of running
an @code{ns-3} simulation is to generate output for study.  You have two basic 
strategies to work with in @code{ns-3}: using generic pre-defined bulk output 
mechanisms and parsing their content to extract interesting information; or 
somehow developing an output mechanism that conveys exactly (and perhaps only) 
the information wanted.

Using pre-defined bulk output mechansims has the advantage of not requiring any
changes to @code{ns-3}, but it does require programming.  Often, pcap or NS_LOG
output messages are gathered during simulation runs and separately run through 
scripts that use grep, sed or awk to parse the messages and reduce and transform
the data to a manageable form.  Programs must be written to do the 
transformation, so this does not come for free.  Of course, if the information
of interest in does not exist in any of the pre-defined output mechanisms,
this approach fails.

If you need to add some tidbit of information to the pre-defined bulk mechanisms,
this can certainly be done; and if you use one of the @code{ns-3} mechanisms, 
you may get your code added as a contribution.

@code{ns-3} provides another mechanism, called Tracing, that avoids some of the 
problems inherent in the bulk output mechanisms.  It has several important 
advantages.  First, you can reduce the amount of data you have to manage by only
tracing the events of interest to you.  Second, if you use this method, you can 
control the format of the output directly so you avoid the postprocessing step 
with sed or awk script.  If you desire, your output can be formatted directly into
a form acceptable by gnuplot, for example.  You can add hooks in the core which 
can then be accessed by other users, but which will produce no information unless
explicitly asked to do so.  For these reasons, we believe that the @code{ns-3}
Tracing system is the best way to get information out of a simulation and is also
therefore one of the most important mechanisms to understand in @command{ns-3}.

@subsection Blunt Instruments
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,

@verbatim
  #include <iostream>
  ...
  void
  SomeFunction (void)
  {
    uint32_t x = SOME_INTERESTING_VALUE;
    ...
    std::cout << "The value of x is " << x << std::endl;
    ...
  } 
@end verbatim

Nobody is going to prevent you from going deep into the core of @code{ns-3} and
adding print statements.  This is insanely easy to do and, after all, you have 
complete control of your own @code{ns-3} branch.  This will probably not turn 
out to be very satisfactory in the long term, though.

As the number of print statements increases in your programs, the task of 
dealing with the large number of outputs will become more and more complicated.  
Eventually, you may feel the need to control what information is being printed 
in some way; perhaps by turning on and off certain categories of prints, or 
increasing or decreasing the amount of information you want.  If you continue 
down this path you may discover that you have re-implemented the @code{NS_LOG}
mechanism.  In order to avoid that, one of the first things you might consider
is using @code{NS_LOG} itself.

We mentioned above that one way to get information out of @code{ns-3} is to 
parse existing NS_LOG output for interesting information.  If you discover that 
some tidbit of information you need is not present in existing log output, you 
could edit the core of @code{ns-3} and simply add your interesting information
to the output stream.  Now, this is certainly better than adding your own
print statements since it follows @code{ns-3} coding conventions and could 
potentially be useful to other people as a patch to the existing core.

Let's pick a random example.  If you wanted to add more logging to the 
@code{ns-3} TCP socket (@code{tcp-socket-impl.cc}) you could just add a new 
message down in the implementation.  Notice that in TcpSocketImpl::ProcessAction()
there is no log message for the @code{ACK_TX} case.  You could simply add one, 
changing the code from:

@verbatim
  bool TcpSocketImpl::ProcessAction (Actions_t a)
  { // These actions do not require a packet or any TCP Headers
    NS_LOG_FUNCTION (this << a);
    switch (a)
    {
      case NO_ACT:
        NS_LOG_LOGIC ("TcpSocketImpl " << this <<" Action: NO_ACT");
        break;
      case ACK_TX:
        SendEmptyPacket (TcpHeader::ACK);
        break;
      ...
@end verbatim

to add a new @code{NS_LOG_LOGIC} in the appropriate @code{case} statement:

@verbatim
  bool TcpSocketImpl::ProcessAction (Actions_t a)
  { // These actions do not require a packet or any TCP Headers
    NS_LOG_FUNCTION (this << a);
    switch (a)
    {
      case NO_ACT:
        NS_LOG_LOGIC ("TcpSocketImpl " << this << " Action: NO_ACT");
        break;
      case ACK_TX:
        NS_LOG_LOGIC ("TcpSocketImpl " << this << " Action: ACK_TX");
        SendEmptyPacket (TcpHeader::ACK);
        break;
      ...
@end verbatim

This may seem fairly simple and satisfying at first glance, but something to
consider is that you will be writing code to add the @code{NS_LOG} statement 
and you will also have to write code (as in grep, sed or awk scripts) to parse
the log output in order to isolate your information.  This is because even 
though you have some control over what is output by the logging system, you 
only have control down to the log component level.  

If you are adding code to an existing module, you will also have to live with the
output that every other developer has found interesting.  You may find that in 
order to get the small amount of information you need, you may have to wade 
through huge amounts of extraneous messages that are of no interest to you.  You
may be forced to save huge log files to disk and process them down to a few lines
whenever you want to do anything.

Since there are no guarantees in @code{ns-3} about the stability of @code{NS_LOG}
messages, you may also discover that pieces of log output on which you depend 
disappear or change between releases.  If you depend on the structure of the 
output, you may find other messages being added or deleted which may affect your
parsing code.

For these reasons, we consider prints to @code{std::cout} and NS_LOG messages 
simple ways to get more information out of @code{ns-3}, but they are really 
unstable and quite blunt instruments.

It is desirable to have a stable facility using stable APIs that allow one to 
reach into the core system and only get the information required.  It is
desirable to be able to do this 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 so the user doesn't have 
to actively go poke around in the system looking for things.

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

@node Overview
@section Overview

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 iteresting 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.  

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.  
Your code looking for trace events from a particular piece of core code could 
happily coexist with other code doing something entirely different from the same
information.

Unless a user connects a trace sink to one of these sources, nothing is output.  By
using the tracing system, both you and other people at the same trace source are 
getting exactly what they want and only what they want out of the system.  Neither
of you are impacting any other user by changing what information is output by the 
system.  If you happen to add a trace source, your work as a good open-source 
citizen may allow other users to provide new utilities that are perhaps very useful
overall, without making any changes to the @code{ns-3} core.  

@node A Simple Low-Level Example
@subsection A Simple Low-Level Example

Let's take a few minutes and walk through a simple tracing example.  We are going
to need a little background on Callbacks to understand what is happening in the
example, so we have to take a small detour right away.

@node Callbacks
@subsubsection Callbacks

The goal of the Callback system in @code{ns-3} 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,

@verbatim
  int (*pfi)(int arg) = 0;
@end verbatim

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, you could
provide a function that looks like,

@verbatim
  int MyFunction (int arg) {}
@end verbatim

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

@verbatim
  pfi = MyFunction;
@end verbatim

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

@verbatim
  int result = (*pfi) (1234);
@end verbatim

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,

@verbatim
  int result = pfi (1234);
@end verbatim

This looks like you are calling a function named ``pfi,'' but the compiler is
smart enough to know to call through the variable @code{pfi} indirectly to
the function @code{MyFunction}.

Conceptually, this is almost exactly how the tracing system will work.
Basically, a trace source @emph{is} a callback.  When a trace sink expresses
interest in receiving trace events, it adds a Callback to a list of Callbacks
internally held by the trace source.  When an interesting event happens, the 
trace source invokes its @code{operator()} providing zero or more parameters.
The @code{operator()} eventually wanders down into the system and does something
remarkably like the indirect call you just saw.  It provides zero or more 
parameters (the call to ``pfi'' above passed one parameter to the target function
@code{MyFunction}.

The important difference that the tracing system adds is that for each trace
source there is an internal list of Callbacks.  Instead of just making one 
indirect call, a trace source may invoke any number of Callbacks.  When a trace
sink expresses interest in notifications from a trace source, it basically just
arranges to add its own function to the callback list.

If you are interested in more details about how this is actually arranged in 
@code{ns-3}, feel free to peruse the Callback section of the manual.

@node Example Code
@subsubsection Example Code

We have provided some code to implement what is really the simplest example
of tracing that can be assembled.  You can find this code in the tutorial
directory as @code{fourth.cc}.  Let's walk through it.

@verbatim
  /* -*- Mode:C++; c-file-style:"gnu"; indent-tabs-mode:nil; -*- */
  /*
   * This program is free software; you can redistribute it and/or modify
   * it under the terms of the GNU General Public License version 2 as
   * published by the Free Software Foundation;
   *
   * This program 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, write to the Free Software
   * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
   */
  
  #include "ns3/object.h"
  #include "ns3/uinteger.h"
  #include "ns3/traced-value.h"
  #include "ns3/trace-source-accessor.h"
  
  #include <iostream>
  
  using namespace ns3;
@end verbatim

Most of this code should be quite familiar to you.  As mentioned above, the
trace system makes heavy use of the Object and Attribute systems, so you will 
need to include them.  The first two includes above bring in the declarations
for those systems explicitly.  You could use the core module header, but this
illustrates how simple this all really is.  

The file, @code{traced-value.h} brings in the required declarations for tracing
of 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 really means is that you will be able to trace changes to a C++
object made using those operators.

Since the tracing system is integrated with Attributes, and Attributes work
with Objects, there must be an @command{ns-3} @code{Object} for the trace source
to live in.  The next code snippet declares and defines a simple Object we can
work with.

@verbatim
  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<int32_t> m_myInt;
  };
@end verbatim

The two important lines of code, above, with respect to tracing are the 
@code{.AddTraceSource} and the @code{TracedValue} declaration of @code{m_myInt}.

The @code{.AddTraceSource} provides the ``hooks'' used for connecting the trace
source to the outside world through the config system.  The @code{TracedValue} 
declaration provides the infrastructure that overloads the operators mentioned 
above and drives the callback process.

@verbatim
  void
  IntTrace (int32_t oldValue, int32_t newValue)
  {
    std::cout << "Traced " << oldValue << " to " << newValue << std::endl;
  }
@end verbatim

This is the definition of the trace sink.  It corresponds directly to a callback
function.  Once it is connected, this function will be called whenever one of the
overloaded operators of the @code{TracedValue} is executed.

We have now seen the trace source and the trace sink.  What remains is code to
connect the source to the sink.

@verbatim
  int
  main (int argc, char *argv[])
  {
    Ptr<MyObject> myObject = CreateObject<MyObject> ();
    myObject->TraceConnectWithoutContext ("MyInteger", MakeCallback(&IntTrace));
  
    myObject->m_myInt = 1234;
  }
@end verbatim

Here we first create the Object in which the trace source lives.

The next step, the @code{TraceConnectWithoutContext}, forms the connection
between the trace source and the trace sink.  Notice the @code{MakeCallback}
template function.  This function does the magic required to create the
underlying @code{ns-3} Callback object and associate it with the function
@code{IntTrace}.  TraceConnect makes the association between your provided
function and the overloaded @code{operator()} in the traced variable referred 
to by the ``MyInteger'' Attribute.  After this association is made, the trace
source will ``fire'' your provided callback function.

The code to make all of this happen is, of course, non-trivial, but the essence
is that you are arranging for something that looks just like the @code{pfi()}
example above to be called by the trace source.  The declaration of the 
@code{TracedValue<int32_t> m_myInt;} in the Object itself performs the magic 
needed to provide the overloaded operators (++, ---, etc.) that will use the
@code{operator()} to actually invoke the Callback with the desired parameters.
The @code{.AddTraceSource} performs the magic to connect the Callback to the 
Config system, and @code{TraceConnectWithoutContext} performs the magic to
connect your function to the trace source, which is specified by Attribute
name.

Let's ignore the bit about context for now.

Finally, the line,

@verbatim
   myObject->m_myInt = 1234;
@end verbatim

should be interpreted as an invocation of @code{operator=} on the member 
variable @code{m_myInt} with the integer @code{1234} passed as a parameter.

It turns out that this operator is defined (by @code{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 --- @code{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.

If you now build and run this example,

@verbatim
  ./waf --run fourth
@end verbatim

you will see the output from the @code{IntTrace} function execute as soon as the
trace source is hit:

@verbatim
  Traced 0 to 1234
@end verbatim

When we executed the code, @code{myObject->m_myInt = 1234;}, the trace source 
fired and automatically provided the before and after values to the trace sink.
The function @code{IntTrace} then printed this to the standard output.  No 
problem.

@subsection Using the Config Subsystem to Connect to Trace Sources

The @code{TraceConnectWithoutContext} call shown above in the simple example is
actually very rarely used in the system.  More typically, the @code{Config}
subsystem is used to allow selecting a trace source in the system using what is
called a @emph{config path}.  We saw an example of this in the previous section
where we hooked the ``CourseChange'' event when we were playing with 
@code{third.cc}.

Recall that we defined a trace sink to print course change information from the
mobility models of our simulation.  It should now be a lot more clear to you 
what this function is doing.

@verbatim
  void
  CourseChange (std::string context, Ptr<const MobilityModel> model)
  {
    Vector position = model->GetPosition ();
    NS_LOG_UNCOND (context << 
      " x = " << position.x << ", y = " << position.y);
  }
@end verbatim

When we connected the ``CourseChange'' trace source to the above trace sink,
we used what is called a ``Config Path'' to specify the source when we
arranged a connection between the pre-defined trace source and the new trace 
sink:

@verbatim
  std::ostringstream oss;
  oss <<
    "/NodeList/" << wifiStaNodes.Get (nWifi - 1)->GetId () <<
    "/$ns3::MobilityModel/CourseChange";

  Config::Connect (oss.str (), MakeCallback (&CourseChange));
@end verbatim

Let's try and make some sense of what is sometimes considered relatively
mysterious code.  For the purposes of discussion, assume that the node 
number returned by the @code{GetId()} is ``7''.  In this case, the path
above turns out to be,

@verbatim
  "/NodeList/7/$ns3::MobilityModel/CourseChange"
@end verbatim

The last segment of a config path must be an @code{Attribute} of an 
@code{Object}.  In fact, if you had a pointer to the @code{Object} that has the
``CourseChange'' @code{Attribute} handy, you could write this just like we did 
in the previous example.  You know by now that we typically store pointers to 
our nodes in a NodeContainer.  In the @code{third.cc} example, the Nodes of
interest are stored in the @code{wifiStaNodes} NodeContainer.  In fact, while
putting the path together, we used this container to get a Ptr<Node> which we
used to call GetId() on.  We could have used this Ptr<Node> directly to call
a connect method directly:

@verbatim
  Ptr<Object> theObject = wifiStaNodes.Get (nWifi - 1);
  theObject->TraceConnectWithoutContext ("CourseChange", MakeCallback (&CourseChange));
@end verbatim

In the @code{third.cc} example, we actually want an additional ``context'' to 
be delivered along with the Callback parameters (which will be explained below) so we 
could actually use the following equivalent code,

@verbatim
  Ptr<Object> theObject = wifiStaNodes.Get (nWifi - 1);
  theObject->TraceConnect ("CourseChange", MakeCallback (&CourseChange));
@end verbatim

It turns out that the internal code for @code{Config::ConnectWithoutContext} and
@code{Config::Connect} actually do find a Ptr<Object> and call the appropriate
TraceConnect method at the lowest level.

The @code{Config} functions take a path that represents a chain of @code{Object} 
pointers.  Each segment of a path corresponds to an Object Attribute.  The last 
segment is the Attribute of interest, and prior segments must be typed to contain
or find Objects.  The @code{Config} code parses and ``walks'' this path until it 
gets to the final segment of the path.  It then interprets the last segment as
an @code{Attribute} on the last Object it found while walking the path.  The  
@code{Config} functions then call the appropriate @code{TraceConnect} or 
@code{TraceConnectWithoutContext} method on the final Object.  Let's see what 
happens in a bit more detail when the above path is walked.

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/7'' refers to the eighth node in the
list of nodes created during the simulation.  This reference is actually a 
@code{Ptr<Node>} and so is a subclass of an @code{ns3::Object}.  

As described in the Object Model section of the @code{ns-3} manual, we support
Object Aggregation.  This allows us to form an association between different 
Objects without any programming.  Each Object in an Aggregation can be reached 
from the other Objects.  

The next path segment being walked begins with the ``$'' character.  This 
indicates to the config system that a @code{GetObject} call should be made 
looking for the type that follows.  It turns out that the MobilityHelper used in 
@code{third.cc} arranges to Aggregate, or associate, a mobility model to each of 
the wireless Nodes.  When you add the ``$'' you are asking for another Object that
has presumably been previously aggregated.  You can think of this as switching
pointers from the original Ptr<Node> as specified by ``/NodeList/7'' to its 
associated mobility model --- which is of type ``$ns3::MobilityModel''.  If you
are familiar with @code{GetObject}, we have asked the system to do the following:

@verbatim
  Ptr<MobilityModel> mobilityModel = node->GetObject<MobilityModel> ()
@end verbatim

We are now at the last Object in the path, so we turn our attention to the 
Attributes of that Object.  The @code{MobilityModel} class defines an Attribute 
called ``CourseChange.''  You can see this by looking at the source code in
@code{src/mobility/mobility-model.cc} and searching for ``CourseChange'' in your
favorite editor.  You should find,

@verbatim
  .AddTraceSource (``CourseChange'',
                   ``The value of the position and/or velocity vector changed'',
                   MakeTraceSourceAccessor (&MobilityModel::m_courseChangeTrace))
@end verbatim

which should look very familiar at this point.  

If you look for the corresponding declaration of the underlying traced variable 
in @code{mobility-model.h} you will find

@verbatim
  TracedCallback<Ptr<const MobilityModel> > m_courseChangeTrace;
@end verbatim

The type declaration @code{TracedCallback} identifies @code{m_courseChangeTrace}
as a special list of Callbacks that can be hooked using the Config functions 
described above.

The @code{MobilityModel} class is designed to be a base class providing a common
interface for all of the specific subclasses.  If you search down to the end of 
the file, you will see a method defined called @code{NotifyCourseChange()}:

@verbatim
  void
  MobilityModel::NotifyCourseChange (void) const
  {
    m_courseChangeTrace(this);
  }
@end verbatim

Derived classes will call into this method whenever they do a course change to
support tracing.  This method invokes @code{operator()} on the underlying 
@code{m_courseChangeTrace}, which will, in turn, invoke all of the registered
Callbacks, calling all of the trace sinks that have registered interest in the
trace source by calling a Config function.

So, in the @code{third.cc} example we looked at, whenever a course change is 
made in one of the @code{RandomWalk2dMobilityModel} instances installed, there
will be a @code{NotifyCourseChange()} call which calls up into the 
@code{MobilityModel} base class.  As seen above, this invokes @code{operator()}
on @code{m_courseChangeTrace}, which in turn, calls any registered trace sinks.
In the example, the only code registering an interest was the code that provided
the config path.  Therefore, the @code{CourseChange} function that was hooked 
from Node number seven will be the only Callback called.

The final piece of the puzzle is the ``context.''  Recall that we saw an output 
looking something like the following from @code{third.cc}:

@verbatim
  /NodeList/7/$ns3::MobilityModel/CourseChange x = 7.27897, y = 2.22677
@end verbatim

The first part of the output is the context.  It is simply the path through
which the config code located the trace source.  In the case we have been looking at
there can be any number of trace sources in the system corresponding to any number
of nodes with mobility models.  There needs to be some way to identify which trace
source is actually the one that fired the Callback.  An easy way is to request a 
trace context when you @code{Config::Connect}.

@subsection How to Find and Connect Trace Sources, and Discover Callback Signatures

The first question that inevitably comes up for new users of the Tracing system is,
``okay, I know that there must be trace sources in the simulation core, but how do
I find out what trace sources are available to me''?  

The second question is, ``okay, I found a trace source, how do I figure out the
config path to use when I connect to it''? 

The third question is, ``okay, I found a trace source, how do I figure out what 
the return type and formal arguments of my callback function need to be''?

The fourth question is, ``okay, I typed that all in and got this incredibly bizarre
error message, what in the world does it mean''?

@subsection What Trace Sources are Available?

The answer to this question is found in the @code{ns-3} Doxygen.  Go to the 
@code{ns-3} web site @uref{http://www.nsnam.org/getting_started.html,,``here''}
and select the ``Doxygen (stable)'' link ``Documentation'' on the navigation
bar to the left side of the page.  Expand the ``Modules'' book in the NS-3 
documentation tree a the upper left by clicking the ``+'' box.  Now, expand
the ``Core'' book in the tree by clicking its ``+'' box.  You should now
see three extremely useful links:

@itemize @bullet
@item The list of all trace sources
@item The list of all attributes
@item The list of all global values
@end itemize

The list of interest to us here is ``the list of all trace sources.''  Go 
ahead and select that link.  You will see, perhaps not too surprisingly, a
list of all of the trace sources available in the @code{ns-3} core.

As an example, scroll down to @code{ns3::MobilityModel}.  You will find
an entry for

@verbatim
  CourseChange: The value of the position and/or velocity vector changed 
@end verbatim

You should recognize this as the trace source we used in the @code{third.cc}
example.  Perusing this list will be helpful.

@subsection What String do I use to Connect?

The easiest way to do this is to grep around in the @code{ns-3} codebase for someone
who has already figured it out,  You should always try to copy someone else's
working code before you start to write your own.  Try something like:

@verbatim
  find . -name '*.cc' | xargs grep CourseChange | grep Connect
@end verbatim

and you may find your answer along with working code.  For example, in this
case, @code{./ns-3-dev/examples/wireless/mixed-wireless.cc} has something
just waiting for you to use:

@verbatim
  Config::Connect (``/NodeList/*/$ns3::MobilityModel/CourseChange'', 
    MakeCallback (&CourseChangeCallback));
@end verbatim

If you cannot find any examples in the distribution, you can find this out
from the @code{ns-3} Doxygen.  It will probably be simplest just to walk 
through the ``CourseChanged'' example.

Let's assume that you have just found the ``CourseChanged'' trace source in 
``The list of all trace sources'' and you want to figure out how to connect to
it.  You know that you are using (again, from the @code{third.cc} example) an
@code{ns3::RandomWalk2dMobilityModel}.  So open the ``Class List'' book in
the NS-3 documentation tree by clicking its ``+'' box.  You will now see a
list of all of the classes in @code{ns-3}.  Scroll down until you see the
entry for @code{ns3::RandomWalk2dMobilityModel} and follow that link.
You should now be looking at the ``ns3::RandomWalk2dMobilityModel Class 
Reference.''

If you now scroll down to the ``Member Function Documentation'' section, you
will see documentation for the @code{GetTypeId} function.  You constructed one
of these in the simple tracing example above:

@verbatim
    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;
    }
@end verbatim

As mentioned above, this is the bit of code that connected the Config 
and Attribute systems to the underlying trace source.  This is also the
place where you should start looking for information about the way to 
connect. 

You are looking at the same information for the RandomWalk2dMobilityModel; and
the information you want is now right there in front of you in the Doxygen:

@verbatim
  This object is accessible through the following paths with Config::Set and Config::Connect: 

  /NodeList/[i]/$ns3::MobilityModel/$ns3::RandomWalk2dMobilityModel 
@end verbatim

The documentation tells you how to get to the @code{RandomWalk2dMobilityModel} 
Object.  Compare the string above with the string we actually used in the 
example code:

@verbatim
  "/NodeList/7/$ns3::MobilityModel"
@end verbatim

The difference is due to the fact that two @code{GetObject} calls are implied 
in the string found in the documentation.  The first, for @code{$ns3::MobilityModel}
will query the aggregation for the base class.  The second implied 
@code{GetObject} call, for @code{$ns3::RandomWalk2dMobilityModel}, is used to ``cast''
the base class to the concrete imlementation class.  The documentation shows 
both of these operations for you.  It turns out that the actual Attribute you are
going to be looking for is found in the base class as we have seen.

Look further down in the @code{GetTypeId} doxygen.  You will find,

@verbatim
  No TraceSources defined for this type.
  TraceSources defined in parent class ns3::MobilityModel:

  CourseChange: The value of the position and/or velocity vector changed 
  Reimplemented from ns3::MobilityModel
@end verbatim

This is exactly what you need to know.  The trace source of interest is found in
@code{ns3::MobilityModel} (which you knew anyway).  The interesting thing this
bit of Doxygen tells you is that you don't need that extra cast in the config
path above to get to the concrete class, since the trace source is actually in
the base class.  Therefore the additional @code{GetObject} is not required and
you simply use the path:

@verbatim
  /NodeList/[i]/$ns3::MobilityModel
@end verbatim

which perfectly matches the example path:

@verbatim
  /NodeList/7/$ns3::MobilityModel
@end verbatim

@subsection What Return Value and Formal Arguments?

The easiest way to do this is to grep around in the @code{ns-3} codebase for someone
who has already figured it out,  You should always try to copy someone else's
working code.  Try something like:

@verbatim
  find . -name '*.cc' | xargs grep CourseChange | grep Connect
@end verbatim

and you may find your answer along with working code.  For example, in this
case, @code{./ns-3-dev/examples/wireless/mixed-wireless.cc} has something
just waiting for you to use.  You will find

@verbatim
  Config::Connect (``/NodeList/*/$ns3::MobilityModel/CourseChange'', 
    MakeCallback (&CourseChangeCallback));
@end verbatim

as a result of your grep.  The @code{MakeCallback} should indicate to you that
there is a callback function there which you can use.  Sure enough, there is:

@verbatim
  static void
  CourseChangeCallback (std::string path, Ptr<const MobilityModel> model)
  {
    ...
  }
@end verbatim

@subsubsection Take my Word for It

If there are no examples to work from, this can be, well, challenging to 
actually figure out from the source code.

Before embarking on a walkthrough of the code, I'll be kind and just tell you
a simple way to figure this out:  The return value of your callback will always 
be void.  The formal parameter list for a @code{TracedCallback} can be found 
from the template parameter list in the declaration.  Recall that for our
current example, this is in @code{mobility-model.h}, where we have previously
found:

@verbatim
  TracedCallback<Ptr<const MobilityModel> > m_courseChangeTrace;
@end verbatim

There is a one-to-one correspondence between the template parameter list in 
the declaration and the formal arguments of the callback function.  Here,
there is one template parameter, which is a @code{Ptr<const MobilityModel>}.
This tells you that you need a function that returns void and takes a
a @code{Ptr<const MobilityModel>}.  For example,

@verbatim
  void
  CourseChangeCallback (Ptr<const MobilityModel> model)
  {
    ...
  }
@end verbatim

That's all you need if you want to @code{Config::ConnectWithoutContext}.  If
you want a context, you need to @code{Config::Connect} and use a Callback 
function that takes a string context, then the required argument.

@verbatim
  void
  CourseChangeCallback (std::string path, Ptr<const MobilityModel> model)
  {
    ...
  }
@end verbatim

If you want to ensure that your @code{CourseChangeCallback} is only visible
in your local file, you can add the keyword @code{static} and come up with:

@verbatim
  static void
  CourseChangeCallback (std::string path, Ptr<const MobilityModel> model)
  {
    ...
  }
@end verbatim

which is exactly what we used in the @code{third.cc} example.

@subsubsection The Hard Way

This section is entirely optional.  It is going to be a bumpy ride, especially
for those unfamiliar with the details of templates.  However, if you get through
this, you will have a very good handle on a lot of the @code{ns-3} low level
idioms.

So, again, let's figure out what signature of callback function is required for
the ``CourseChange'' Attribute.  This is going to be painful, but you only need
to do this once.  After you get through this, you will be able to just look at
a @code{TracedCallback} and understand it.

The first thing we need to look at is the declaration of the trace source.
Recall that this is in @code{mobility-model.h}, where we have previously
found:

@verbatim
  TracedCallback<Ptr<const MobilityModel> > m_courseChangeTrace;
@end verbatim

This declaration is for a template.  The template parameter is inside the
angle-brackets, so we are really interested in finding out what that
@code{TracedCallback<>} is.  If you have absolutely no idea where this might
be found, grep is your friend.

We are probably going to be interested in some kind of declaration in the 
@code{ns-3} source, so first change into the @code{src} directory.  Then, 
we know this declaration is going to have to be in some kind of header file,
so just grep for it using:

@verbatim
  find . -name '*.h' | xargs grep TracedCallback
@end verbatim

You'll see 124 lines fly by (I piped this through wc to see how bad it was).
Although that may seem like it, that's not really a lot.  Just pipe the output
through more and start scanning through it.  On the first page, you will see
some very suspiciously template-looking stuff.

@verbatim
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::TracedCallback ()
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::ConnectWithoutContext (c ...
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::Connect (const CallbackB ...
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::DisconnectWithoutContext ...
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::Disconnect (const Callba ...
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (void) const ...
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1) const ...
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1, T2 a2 ...
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1, T2 a2 ...
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1, T2 a2 ...
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1, T2 a2 ...
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1, T2 a2 ...
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::operator() (T1 a1, T2 a2 ...
@end verbatim

It turns out that all of this comes from the header file 
@code{traced-callback.h} which sounds very promising.  You can then take a
look at @code{mobility-model.h} and see that there is a line which confirms
this hunch:

@verbatim
  #include "ns3/traced-callback.h"
@end verbatim

Of course, you could have gone at this from the other direction and started
by looking at the includes in @code{mobility-model.h} and noticing the 
include of @code{traced-callback.h} and inferring that this must be the file
you want.

In either case, the next step is to take a look at @code{src/core/traced-callback.h}
in your favorite editor to see what is happening.

You will see a comment at the top of the file that should be comforting:

@verbatim
  An ns3::TracedCallback has almost exactly the same API as a normal ns3::Callback but
  instead of forwarding calls to a single function (as an ns3::Callback normally does),
  it forwards calls to a chain of ns3::Callback.
@end verbatim

This should sound very familiar and let you know you are on the right track.

Just after this comment, you will find,

@verbatim
  template<typename T1 = empty, typename T2 = empty, 
           typename T3 = empty, typename T4 = empty,
           typename T5 = empty, typename T6 = empty,
           typename T7 = empty, typename T8 = empty>
  class TracedCallback 
  {
    ...
@end verbatim

This tells you that TracedCallback is a templated class.  It has eight possible
type parameters with default values.  Go back and compare this with the 
declaration you are trying to understand:

@verbatim
  TracedCallback<Ptr<const MobilityModel> > m_courseChangeTrace;
@end verbatim

The @code{typename T1} in the templated class declaration corresponds to the 
@code{Ptr<const MobilityModel>} in the declaration above.  All of the other
type parameters are left as defaults.  Looking at the constructor really
doesn't tell you much.  The one place where you have seen a connection made
between your Callback function and the tracing system is in the @code{Connect}
and @code{ConnectWithoutContext} functions.  If you scroll down, you will see
a @code{ConnectWithoutContext} method here:

@verbatim
  template<typename T1, typename T2, 
           typename T3, typename T4,
           typename T5, typename T6,
           typename T7, typename T8>
  void 
  TracedCallback<T1,T2,T3,T4,T5,T6,T7,T8>::ConnectWithoutContext ...
  {
    Callback<void,T1,T2,T3,T4,T5,T6,T7,T8> cb;
    cb.Assign (callback);
    m_callbackList.push_back (cb);
  }
@end verbatim

You are now in the belly of the beast.  When the template is instantiated for
the declaration above, the compiler will replace @code{T1} with 
@code{Ptr<const MobilityModel>}.  

@verbatim
  void 
  TracedCallback<Ptr<const MobilityModel>::ConnectWithoutContext ... cb
  {
    Callback<void, Ptr<const MobilityModel> > cb;
    cb.Assign (callback);
    m_callbackList.push_back (cb);
  }
@end verbatim

You can now see the implementation of everything we've been talking about.  The
code creates a Callback of the right type and assigns your function to it.  This
is the equivalent of the @code{pfi = MyFunction} we discussed at the start of
this section.  The code then adds the Callback to the list of Callbacks for 
this source.  The only thing left is to look at the definition of Callback.
Using the same grep trick as we used to find @code{TracedCallback}, you will be
able to find that the file @code{./core/callback.h} is the one we need to look at.

If you look down through the file, you will see a lot of probably almost
incomprehensible template code.  You will eventually come to some Doxygen for
the Callback template class, though.  Fortunately, there is some English:

@verbatim
  This class template implements the Functor Design Pattern.
  It is used to declare the type of a Callback:
   - the first non-optional template argument represents
     the return type of the callback.
   - the second optional template argument represents
     the type of the first argument to the callback.
   - the third optional template argument represents
     the type of the second argument to the callback.
   - the fourth optional template argument represents
     the type of the third argument to the callback.
   - the fifth optional template argument represents
     the type of the fourth argument to the callback.
   - the sixth optional template argument represents
     the type of the fifth argument to the callback.
@end verbatim

We are trying to figure out what the

@verbatim
    Callback<void, Ptr<const MobilityModel> > cb;
@end verbatim

declaration means.  Now we are in a position to understand that the first
(non-optional) parameter, @code{void}, represents the return type of the 
Callback.  The second (non-optional) parameter, @code{Ptr<const MobilityModel>} 
represents the first argument to the callback.

The Callback in question is your function to receive the trace events.  From
this you can infer that you need a function that returns @code{void} and takes
a @code{Ptr<const MobilityModel>}.  For example,

@verbatim
  void
  CourseChangeCallback (Ptr<const MobilityModel> model)
  {
    ...
  }
@end verbatim

That's all you need if you want to @code{Config::ConnectWithoutContext}.  If
you want a context, you need to @code{Config::Connect} and use a Callback 
function that takes a string context.  This is because the @code{Connect}
function will provide the context for you.  You'll need:

@verbatim
  void
  CourseChangeCallback (std::string path, Ptr<const MobilityModel> model)
  {
    ...
  }
@end verbatim

If you want to ensure that your @code{CourseChangeCallback} is only visible
in your local file, you can add the keyword @code{static} and come up with:

@verbatim
  static void
  CourseChangeCallback (std::string path, Ptr<const MobilityModel> model)
  {
    ...
  }
@end verbatim

which is exactly what we used in the @code{third.cc} example.  Perhaps you
should now go back and reread the previous section (Take My Word for It).

If you are interested in more details regarding the implementation of 
Callbacks, feel free to take a look at the @code{ns-3} manual.  They are one
of the most frequently used constructs in the low-level parts of @code{ns-3}.
It is, in my opinion, a quite elegant thing.

@subsection What About TracedValue?

Earlier in this section, we presented a simple piece of code that used a
@code{TracedValue<int32_t>} to demonstrate the basics of the tracing code.
We just glossed over the way to find the return type and formal arguments
for the @code{TracedValue}.  Rather than go through the whole exercise, we
will just point you at the correct file, @code{src/core/traced-value.h} and
to the important piece of code:

@verbatim
  template <typename T>
  class TracedValue
  {
  public:
    ...
    void Set (const T &v) {
      if (m_v != v)
        {
  	m_cb (m_v, v);
  	m_v = v;
        }
    }
    ...
  private:
    T m_v;
    TracedCallback<T,T> m_cb;
  };
@end verbatim

Here you see that the @code{TracedValue} is templated, of course.  In the simple
example case at the start of the section, the typename is int32_t.  This means 
that the member variable being traced (@code{m_v} in the private section of the 
class) will be an @code{int32_t m_v}.  The @code{Set} method will take a 
@code{const int32_t &v} as a parameter.  You should now be able to understand 
that the @code{Set} code will fire the @code{m_cb} callback with two parameters:
the first being the current value of the @code{TracedValue}; and the second 
being the new value being set.

The callback, @code{m_cb} is declared as a @code{TracedCallback<T, T>} which
will correspond to a @code{TracedCallback<int32_t, int32_t>} when the class is 
instantiated.

Recall that the callback target of a TracedCallback always returns @code{void}.  
Further recall that there is a one-to-one correspondence between the template 
parameter list in the declaration and the formal arguments of the callback 
function.  Therefore the callback will need to have a function signature that 
looks like:

@verbatim
  void
  MyCallback (int32_t oldValue, int32_t newValue)
  {
    ...
  }
@end verbatim

It probably won't surprise you that this is exactly what we provided in that 
simple example we covered so long ago:

@verbatim
  void
  IntTrace (int32_t oldValue, int32_t newValue)
  {
    std::cout << "Traced " << oldValue << " to " << newValue << std::endl;
  }
@end verbatim

@c ============================================================================
@c A Real Example
@c ============================================================================
@node A Real Example
@section A Real Example

Let's do an example taken from one of the best-known books on TCP around.  
``TCP/IP Illustrated, Volume 1: The Protocols,'' by W. Richard Stevens is a 
classic.  I just flipped the book open and ran across a nice plot of both the 
congestion window and sequence numbers versus time on page 366.  Stevens calls 
this, ``Figure 21.10. Value of cwnd and send sequence number while data is being 
transmitted.''  Let's just recreate the cwnd part of that plot in @command{ns-3}
using the tracing system and @code{gnuplot}.

@subsection Are There Trace Sources Available?

The first thing to think about is how we want to get the data out.  What is it
that we need to trace?  The first thing to do is to consult ``The list of all
trace sources'' to see what we have to work with.  Recall that this is found
in the @command{ns-3} Doxygen in the ``Core'' Module section.  If you scroll
through the list, you will eventually find:

@verbatim
  ns3::TcpSocketImpl
  CongestionWindow: The TCP connection's congestion window
@end verbatim

It turns out that the @command{ns-3} TCP implementation lives (mostly) in the 
file @code{src/internet-stack/tcp-socket-impl.cc}.  If you don't know this a 
priori, you can use the recursive grep trick:

@verbatim
  find . -name '*.cc' | xargs grep -i tcp
@end verbatim

You will find page after page of instances of tcp pointing you to that file. 

If you open @code{src/internet-stack/tcp-socket-impl.cc} in your favorite 
editor, you will see right up at the top of the file, the following declarations:

@verbatim
  TypeId
  TcpSocketImpl::GetTypeId ()
  {
    static TypeId tid = TypeId(``ns3::TcpSocketImpl'')
      .SetParent<TcpSocket> ()
      .AddTraceSource (``CongestionWindow'',
                       ``The TCP connection's congestion window'',
                       MakeTraceSourceAccessor (&TcpSocketImpl::m_cWnd))
      ;
    return tid;
  }
@end verbatim

This should tell you to look for the declaration of @code{m_cWnd} in the header
file @code{src/internet-stack/tcp-socket-impl.h}.  If you open this file in your
favorite editor, you will find:

@verbatim
  TracedValue<uint32_t> m_cWnd; //Congestion window
@end verbatim

You should now understand this code completely.  If we have a pointer to the 
@code{TcpSocketImpl}, we can @code{TraceConnect} to the ``CongestionWindow'' trace 
source if we provide an appropriate callback target.  This is the same kind of
trace source that we saw in the simple example at the start of this section,
except that we are talking about @code{uint32_t} instead of @code{int32_t}.

We now know that we need to provide a callback that returns void and takes 
two @code{uint32_t} parameters, the first being the old value and the second 
being the new value:

@verbatim
  void
  CwndTrace (uint32_t oldValue, uint32_t newValue)
  {
    ...
  }
@end verbatim

@subsection What Script to Use?

It's always best to try and find working code laying around that you can 
modify, rather than starting from scratch.  So the first order of business now
is to find some code that already hooks the ``CongestionWindow'' trace source
and see if we can modify it.  As usual, grep is your friend:

@verbatim
  find . -name '*.cc' | xargs grep CongestionWindow
@end verbatim

This will point out a couple of promising candidates: 
@code{examples/tcp/tcp-large-transfer.cc} and 
@code{src/test/ns3tcp/ns3tcp-cwnd-test-suite.cc}.

We haven't visited any of the test code yet, so let's take a look there.  You
will typically find that test code is fairly minimal, so this is probably a
very good bet.  Open @code{src/test/ns3tcp/ns3tcp-cwnd-test-suite.cc} in your
favorite editor and search for ``CongestionWindow.''  You will find,

@verbatim
  ns3TcpSocket->TraceConnectWithoutContext (``CongestionWindow'', 
    MakeCallback (&Ns3TcpCwndTestCase1::CwndChange, this));
@end verbatim

This should look very familiar to you.  We mentioned above that if we had a
pointer to the @code{TcpSocketImpl}, we could @code{TraceConnect} to the 
``CongestionWindow'' trace source.  That's exactly what we have here; so it
turns out that this line of code does exactly what we want.  Let's go ahead
and extract the code we need from this function 
(@code{Ns3TcpCwndTestCase1::DoRun (void)}).  If you look at this function,
you will find that it looks just like an @code{ns-3} script.  It turns out that
is exactly what it is.  It is a script run by the test framework, so we can just
pull it out and wrap it in @code{main} instead of in @code{DoRun}.  Rather than
walk through this, step, by step, we have provided the file that results from
porting this test back to a native @code{ns-3} script --
@code{examples/tutorial/fifth.cc}.  

@subsection A Common Problem and Solution

The @code{fifth.cc} example demonstrates an extremely important rule that you 
must understand before using any kind of @code{Attribute}:  you must ensure 
that the target of a @code{Config} command exists before trying to use it.
This is no different than saying an object must be instantiated before trying
to call it.  Although this may seem obvious when stated this way, it does
trip up many people trying to use the system for the first time.

Let's return to basics for a moment.  There are three basic time periods that
exist in any @command{ns-3} script.  The first time period is sometimes called 
``Configuration Time'' or ``Setup Time,'' and is in force during the period 
when the @code{main} function of your script is running, but before 
@code{Simulator::Run} is called.  The second time period  is sometimes called
``Simulation Time'' and is in force during the time period when 
@code{Simulator::Run} is actively executing its events.  After it completes
executing the simulation,  @code{Simulator::Run} will return control back to 
the @code{main} function.  When this happens, the script enters what can be 
called ``Teardown Time,'' which is when the structures and objects created 
during setup and taken apart and released.

Perhaps the most common mistake made in trying to use the tracing system is 
assuming that entities constructed dynamically during simulation time are
available during configuration time.  In particular, an @command{ns-3}
@code{Socket} is a dynamic object often created by @code{Applications} to
communicate between @code{Nodes}.  An @command{ns-3} @code{Application} 
always has a ``Start Time'' and a ``Stop Time'' associated with it.  In the
vast majority of cases, an @code{Application} will not attempt to create 
a dynamic object until its @code{StartApplication} method is called.  This
is to ensure that the simulation is completely configured before the app
tries to do anything (what would happen if it tried to connect to a node
that didn't exist yet during configuration time).  The answer to this 
issue is to 1) create a simulator event that is run after the dynamic object
is created and hook the trace when that event is executed; or 2) create the
dynamic object at configuration time, hook it then, and give the object to
the system to use during simulation time.  We took the second approach in
the @code{fifth.cc} example.  This decision required us to create the 
@code{MyApp} @code{Application}, the entire purpose of which is to 
take a @code{Socket} as a parameter.  

@subsection A fifth.cc Walkthrough

Now, let's take a look at the example program we constructed by dissecting
the congestion window test.  Open @code{examples/tutorial/fifth.cc} in your
favorite editor.  You should see some familiar looking code:

@verbatim
  /* -*- Mode:C++; c-file-style:''gnu''; indent-tabs-mode:nil; -*- */
  /*
   * This program is free software; you can redistribute it and/or modify
   * it under the terms of the GNU General Public License version 2 as
   * published by the Free Software Foundation;
   *
   * This program 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, write to the Free Software
   * Foundation, Include., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
   */
  
  #include <fstream>
  #include "ns3/core-module.h"
  #include "ns3/common-module.h"
  #include "ns3/simulator-module.h"
  #include "ns3/node-module.h"
  #include "ns3/helper-module.h"
  
  using namespace ns3;
  
  NS_LOG_COMPONENT_DEFINE ("FifthScriptExample");
@end verbatim

This has all been covered, so we won't rehash it.  The next lines of source are
the network illustration and a comment addressing the problem described above
with @code{Socket}.

@verbatim
  // ===========================================================================
  //
  //         node 0                 node 1
  //   +----------------+    +----------------+
  //   |    ns-3 TCP    |    |    ns-3 TCP    |
  //   +----------------+    +----------------+
  //   |    10.1.1.1    |    |    10.1.1.2    |
  //   +----------------+    +----------------+
  //   | point-to-point |    | point-to-point |
  //   +----------------+    +----------------+
  //           |                     |
  //           +---------------------+
  //                5 Mbps, 2 ms
  //
  //
  // We want to look at changes in the ns-3 TCP congestion window.  We need
  // to crank up a flow and hook the CongestionWindow attribute on the socket
  // of the sender.  Normally one would use an on-off application to generate a
  // flow, but this has a couple of problems.  First, the socket of the on-off
  // application is not created until Application Start time, so we wouldn't be
  // able to hook the socket (now) at configuration time.  Second, even if we
  // could arrange a call after start time, the socket is not public so we
  // couldn't get at it.
  //
  // So, we can cook up a simple version of the on-off application that does what
  // we want.  On the plus side we don't need all of the complexity of the on-off
  // application.  On the minus side, we don't have a helper, so we have to get
  // a little more involved in the details, but this is trivial.
  //
  // So first, we create a socket and do the trace connect on it; then we pass
  // this socket into the constructor of our simple application which we then
  // install in the source node.
  // ===========================================================================
  //
@end verbatim

This should also be self-explanatory.  

The next part is the declaration of the @code{MyApp} @code{Application} that
we put together to allow the @code{Socket} to be created at configuration time.

@verbatim
  class MyApp : public Application
  {
  public:
  
    MyApp ();
    virtual ~MyApp();
  
    void Setup (Ptr<Socket> socket, Address address, uint32_t packetSize, 
      uint32_t nPackets, DataRate dataRate);
  
  private:
    virtual void StartApplication (void);
    virtual void StopApplication (void);
  
    void ScheduleTx (void);
    void SendPacket (void);
  
    Ptr<Socket>     m_socket;
    Address         m_peer;
    uint32_t        m_packetSize;
    uint32_t        m_nPackets;
    DataRate        m_dataRate;
    EventId         m_sendEvent;
    bool            m_running;
    uint32_t        m_packetsSent;
  };
@end verbatim

You can see that this class inherits from the @command{ns-3} @code{Application}
class.  Take a look at @code{src/node/application.h} if you are interested in 
what is inherited.  The @code{MyApp} class is obligated to override the 
@code{StartApplication} and @code{StopApplication} methods.  These methods are
called when the corresponding base class @code{Start} and @code{Stop} methods are
called during simulation time.  

@subsubsection How Applications are Started and Stopped

It is worthwhile to spend a bit of time explaining how events actually get 
started in the system.  The most common way to start pumping events is to start
an @code{Application}.  This is done as the result of the following (hopefully)
familar lines of an @command{ns-3} script:

@verbatim
  ApplicationContainer apps = ...
  apps.Start (Seconds (1.0));
  apps.Stop (Seconds (10.0));
@end verbatim

The application container code (see @code{src/helper/application-container.h} if
you are interested) loops through its contained applications and calls,

@verbatim
  app->Start (startTime);
  app->Stop (stopTime);
@end verbatim

on each of them.  The @code{Start} method of an @code{Application} calls 
@code{Application::ScheduleStart} (see @code{src/helper/application-container.cc})
which, in turn, schedules an event to start the @code{Application}:

@verbatim
  Simulator::Schedule (startTime, &Application::StartApplication, this);
@end verbatim

Since @code{MyApp} inherits from @code{Application} and overrides 
@code{StartApplication}, this bit of code causes the simulator to execute
something that is effectively like,

@verbatim
  this->StartApplication (startTime);
@end verbatim

where the @code{this} pointer, if you have kept it all straight, is the pointer
to the @code{Application} in the container.  It is then expected that another 
event will be scheduled in the overridden @code{StartApplication} that will 
begin doing some application-specific function, like sending packets.  

@code{StopApplication} operates in a similar manner and tells the @code{Application}
to stop generating events.

@subsubsection The MyApp Application

The @code{MyApp} @code{Application} needs a constructor and a destructor,
of course:

@verbatim
  MyApp::MyApp ()
    : m_socket (0),
      m_peer (),
      m_packetSize (0),
      m_nPackets (0),
      m_dataRate (0),
      m_sendEvent (),
      m_running (false),
      m_packetsSent (0)
  {
  }
  
  MyApp::~MyApp()
  {
    m_socket = 0;
  }
@end verbatim

The existence of the next bit of code is the whole reason why we wrote this
@code{Application} in the first place.

@verbatim
void
MyApp::Setup (Ptr<Socket> socket, Address address, uint32_t packetSize, 
                     uint32_t nPackets, DataRate dataRate)
{
  m_socket = socket;
  m_peer = address;
  m_packetSize = packetSize;
  m_nPackets = nPackets;
  m_dataRate = dataRate;
}
@end verbatim

This code should be pretty self-explanatory.  We are just initializing member
variables.  The important one from the perspective of tracing is the 
@code{Ptr<Socket> socket} which we needed to provide to the application 
during configuration time.  Recall that we are going to create the @code{Socket}
as a @code{TcpSocket} (which is implemented by @code{TcpSocketImpl}) and hook 
its ``CongestionWindow'' trace source before passing it to the @code{Setup}
method.

@verbatim
  void
  MyApp::StartApplication (void)
  {
    m_running = true;
    m_packetsSent = 0;
    m_socket->Bind ();
    m_socket->Connect (m_peer);
    SendPacket ();
  }
@end verbatim

The above code is the overridden implementation @code{Application::StartApplication}
that will be automatically called by the simulator to start our @code{Application}
running.  You can see that it does a @code{Socket} @code{Bind} operation.  If
you are familiar with Berkeley Sockets this shouldn't be a surprise.  It performs
the required work on the local side of the connection just as you might expect.
The following @code{Connect} will do what is required to establish a connection 
with the TCP at @code{Address} m_peer.  It should now be clear why we need to defer
a lot of this to simulation time, since the @code{Connect} is going to need a fully
functioning network to complete.  After the @code{Connect}, the @code{Application} 
then starts creating simulation events by calling @code{SendPacket}.

The next bit of code explains to the @code{Application} how to stop creating 
simulation events.

@verbatim
  void
  MyApp::StopApplication (void)
  {
    m_running = false;
  
    if (m_sendEvent.IsRunning ())
      {
        Simulator::Cancel (m_sendEvent);
      }
  
    if (m_socket)
      {
        m_socket->Close ();
      }
  }
@end verbatim

Every time a simulation event is scheduled, an @code{Event} is created.  If the 
@code{Event} is pending execution or executing, its method @code{IsRunning} will
return @code{true}.  In this code, if @code{IsRunning()} returns true, we 
@code{Cancel} the event which removes it from the simulator event queue.  By 
doing this, we break the chain of events that the @code{Application} is using to
keep sending its @code{Packets} and the @code{Application} goes quiet.  After we 
quiet the @code{Application} we @code{Close} the socket which tears down the TCP 
connection.

The socket is actually deleted in the destructor when the @code{m_socket = 0} is
executed.  This removes the last reference to the underlying Ptr<Socket> which 
causes the destructor of that Object to be called.

Recall that @code{StartApplication} called @code{SendPacket} to start the 
chain of events that describes the @code{Application} behavior.

@verbatim
  void
  MyApp::SendPacket (void)
  {
    Ptr<Packet> packet = Create<Packet> (m_packetSize);
    m_socket->Send (packet);
  
    if (++m_packetsSent < m_nPackets)
      {
        ScheduleTx ();
      }
  }
@end verbatim

Here, you see that @code{SendPacket} does just that.  It creates a @code{Packet}
and then does a @code{Send} which, if you know Berkeley Sockets, is probably 
just what you expected to see.

It is the responsibility of the @code{Application} to keep scheduling the 
chain of events, so the next lines call @code{ScheduleTx} to schedule another
transmit event (a @code{SendPacket}) until the @code{Application} decides it
has sent enough.

@verbatim
  void
  MyApp::ScheduleTx (void)
  {
    if (m_running)
      {
        Time tNext (Seconds (m_packetSize * 8 / static_cast<double> (m_dataRate.GetBitRate ())));
        m_sendEvent = Simulator::Schedule (tNext, &MyApp::SendPacket, this);
      }
  }
@end verbatim

Here, you see that @code{ScheduleTx} does exactly that.  If the @code{Applciation}
is running (if @code{StopApplication} has not been called) it will schedule a 
new event, which calls @code{SendPacket} again.  The alert reader will spot
something that also trips up new users.  The data rate of an @code{Application} is
just that.  It has nothing to do with the data rate of an underlying @code{Channel}.
This is the rate at which the @code{Application} produces bits.  It does not take
into account any overhead for the various protocols or channels that it uses to 
transport the data.  If you set the data rate of an @code{Application} to the same
data rate as your underlying @code{Channel} you will eventually get a buffer overflow.

@subsubsection The Trace Sinks

The whole point of this exercise is to get trace callbacks from TCP indicating the
congestion window has been updated.  The next piece of code implements the 
corresponding trace sink:

@verbatim
  static void
  CwndChange (uint32_t oldCwnd, uint32_t newCwnd)
  {
    NS_LOG_UNCOND (Simulator::Now ().GetSeconds () << ``\t'' << newCwnd);
  }
@end verbatim

This should be very familiar to you now, so we won't dwell on the details.  This
function just logs the current simulation time and the new value of the 
congestion window every time it is changed.  You can probably imagine that you
could load the resulting output into a graphics program (gnuplot or Excel) and
immediately see a nice graph of the congestion window behavior over time.

We added a new trace sink to show where packets are dropped.  We are going to 
add an error model to this code also, so we wanted to demonstrate this working.

@verbatim
  static void
  RxDrop (Ptr<const Packet> p)
  {
    NS_LOG_UNCOND ("RxDrop at " << Simulator::Now ().GetSeconds ());
  }
@end verbatim

This trace sink will be connected to the ``PhyRxDrop'' trace source of the 
point-to-point NetDevice.  This trace source fires when a packet is dropped
by the physical layer of a @code{NetDevice}.  If you take a small detour to the
source (@code{src/devices/point-to-point/point-to-point-net-device.cc}) you will
see that this trace source refers to @code{PointToPointNetDevice::m_phyRxDropTrace}.
If you then look in @code{src/devices/point-to-point/point-to-point-net-device.h}
for this member variable, you will find that it is declared as a
@code{TracedCallback<Ptr<const Packet> >}.  This should tell you that the
callback target should be a function that returns void and takes a single
parameter which is a @code{Ptr<const Packet>} -- just what we have above.

@subsubsection The Main Program

The following code should be very familiar to you by now:

@verbatim
  int
  main (int argc, char *argv[])
  {
    NodeContainer nodes;
    nodes.Create (2);
  
    PointToPointHelper pointToPoint;
    pointToPoint.SetDeviceAttribute ("DataRate", StringValue ("5Mbps"));
    pointToPoint.SetChannelAttribute ("Delay", StringValue ("2ms"));
  
    NetDeviceContainer devices;
    devices = pointToPoint.Install (nodes);
@end verbatim

This creates two nodes with a point-to-point channel between them, just as
shown in the illustration at the start of the file.

The next few lines of code show something new.  If we trace a connection that
behaves perfectly, we will end up with a monotonically increasing congestion
window.  To see any interesting behavior, we really want to introduce link 
errors which will drop packets, cause duplicate ACKs and trigger the more
interesting behaviors of the congestion window.

@command{ns-3} provides @code{ErrorModel} objects which can be attached to
@code{Channels}.  We are using the @code{RateErrorModel} which allows us
to introduce errors into a @code{Channel} at a given @emph{rate}. 

@verbatim
  Ptr<RateErrorModel> em = CreateObjectWithAttributes<RateErrorModel> (
    "RanVar", RandomVariableValue (UniformVariable (0., 1.)),
    "ErrorRate", DoubleValue (0.00001));
  devices.Get (1)->SetAttribute ("ReceiveErrorModel", PointerValue (em));
@end verbatim

The above code instantiates a @code{RateErrorModel} Object.  Rather than 
using the two-step process of instantiating it and then setting Attributes,
we use the convenience function @code{CreateObjectWithAttributes} which
allows us to do both at the same time.  We set the ``RanVar'' 
@code{Attribute} to a random variable that generates a uniform distribution
from 0 to 1.  We also set the ``ErrorRate'' @code{Attribute}.
We then set the resulting instantiated @code{RateErrorModel} as the error
model used by the point-to-point @code{NetDevice}.  This will give us some
retransmissions and make our plot a little more interesting.

@verbatim
  InternetStackHelper stack;
  stack.Install (nodes);

  Ipv4AddressHelper address;
  address.SetBase (``10.1.1.0'', ``255.255.255.252'');
  Ipv4InterfaceContainer interfaces = address.Assign (devices);
@end verbatim

The above code should be familiar.  It installs internet stacks on our two
nodes and creates interfaces and assigns IP addresses for the point-to-point
devices.

Since we are using TCP, we need something on the destination node to receive
TCP connections and data.  The @code{PacketSink} @code{Application} is commonly
used in @command{ns-3} for that purpose.

@verbatim
  uint16_t sinkPort = 8080;
  Address sinkAddress (InetSocketAddress(interfaces.GetAddress (1), sinkPort));
  PacketSinkHelper packetSinkHelper ("ns3::TcpSocketFactory", 
    InetSocketAddress (Ipv4Address::GetAny (), sinkPort));
  ApplicationContainer sinkApps = packetSinkHelper.Install (nodes.Get (1));
  sinkApps.Start (Seconds (0.));
  sinkApps.Stop (Seconds (20.));
@end verbatim

This should all be familiar, with the exception of,

@verbatim
  PacketSinkHelper packetSinkHelper ("ns3::TcpSocketFactory", 
    InetSocketAddress (Ipv4Address::GetAny (), sinkPort));
@end verbatim

This code instantiates a @code{PacketSinkHelper} and tells it to create sockets
using the class @code{ns3::TcpSocketFactory}.  This class implements a design 
pattern called ``object factory'' which is a commonly used mechanism for 
specifying a class used to create objects in an abstract way.  Here, instead of 
having to create the objects themselves, you provide the @code{ PacketSinkHelper}
a string that specifies a @code{TypeId} string used to create an object which 
can then be used, in turn, to create instances of the Objects created by the 
factory.

The remaining parameter tells the @code{Application} which address and port it
should @code{Bind} to.

The next two lines of code will create the socket and connect the trace source.

@verbatim
  Ptr<Socket> ns3TcpSocket = Socket::CreateSocket (nodes.Get (0), 
    TcpSocketFactory::GetTypeId ());
  ns3TcpSocket->TraceConnectWithoutContext (``CongestionWindow'', 
    MakeCallback (&CwndChange));
@end verbatim

The first statement calls the static member function @code{Socket::CreateSocket}
and provides a @code{Node} and an explicit @code{TypeId} for the object factory
used to create the socket.  This is a slightly lower level call than the 
@code{PacketSinkHelper} call above, and uses an explicit C++ type instead of 
one referred to by a string.  Otherwise, it is conceptually the same thing.

Once the @code{TcpSocket} is created and attached to the @code{Node}, we can
use @code{TraceConnectWithoutContext} to connect the CongestionWindow trace 
source to our trace sink.

Recall that we coded an @code{Application} so we could take that @code{Socket}
we just made (during configuration time) and use it in simulation time.  We now 
have to instantiate that @code{Application}.  We didn't go to any trouble to
create a helper to manage the @code{Application} so we are going to have to 
create and install it ``manually.''  This is actually quite easy:

@verbatim
  Ptr<MyApp> app = CreateObject<MyApp> ();
  app->Setup (ns3TcpSocket, sinkAddress, 1040, 1000, DataRate ("1Mbps"));
  nodes.Get (0)->AddApplication (app);
  app->Start (Seconds (1.));
  app->Stop (Seconds (20.));
@end verbatim

The first line creates an @code{Object} of type @code{MyApp} -- our
@code{Application}.  The second line tells the @code{Application} what
@code{Socket} to use, what address to connect to, how much data to send 
at each send event, how many send events to generate and the rate at which
to produce data from those events.

Next, we manually add the @code{MyApp Application} to the source node
and explicitly call the @code{Start} and @code{Stop} methods on the 
@code{Application} to tell it when to start and stop doing its thing.

We need to actually do the connect from the receiver point-to-point @code{NetDevice}
to our callback now.

@verbatim
  devices.Get (1)->TraceConnectWithoutContext("PhyRxDrop", MakeCallback (&RxDrop));
@end verbatim

It should now be obvious that we are getting a reference to the receiving 
@code{Node NetDevice} from its container and connecting the trace source defined
by the attribute ``PhyRxDrop'' on that device to the trace sink @code{RxDrop}.

Finally, we tell the simulator to override any @code{Applications} and just
stop processing events at 20 seconds into the simulation.

@verbatim
    Simulator::Stop (Seconds(20));
    Simulator::Run ();
    Simulator::Destroy ();

    return 0;
  }
@end verbatim

Recall that as soon as @code{Simulator::Run} is called, configuration time
ends, and simulation time begins.  All of the work we orchestrated by 
creating the @code{Application} and teaching it how to connect and send
data actually happens during this function call.

As soon as @code{Simulator::Run} returns, the simulation is complete and
we enter the teardown phase.  In this case, @code{Simulator::Destroy} takes
care of the gory details and we just return a success code after it completes.

@subsection Running fifth.cc

Since we have provided the file @code{fifth.cc} for you, if you have built
your distribution (in debug mode since it uses NS_LOG -- recall that optimized
builds optimize out NS_LOGs) it will be waiting for you to run.

@verbatim
  ./waf --run fifth
  Waf: Entering directory `/home/craigdo/repos/ns-3-allinone-dev/ns-3-dev/build
  Waf: Leaving directory `/home/craigdo/repos/ns-3-allinone-dev/ns-3-dev/build'
  'build' finished successfully (0.684s)
  1.20919 1072
  1.21511 1608
  1.22103 2144
  ...
  1.2471  8040
  1.24895 8576
  1.2508  9112
  RxDrop at 1.25151
  ...
@end verbatim

You can probably see immediately a downside of using prints of any kind in your
traces.  We get those extraneous waf messages printed all over our interesting
information along with those RxDrop messages.  We will remedy that soon, but I'm
sure you can't wait to see the results of all of this work.  Let's redirect that
output to a file called @code{cwnd.dat}:

@verbatim
  ./waf --run fifth > cwnd.dat 2>&1
@end verbatim

Now edit up ``cwnd.dat'' in your favorite editor and remove the waf build status
and drop lines, leaving only the traced data (you could also comment out the
@code{TraceConnectWithoutContext("PhyRxDrop", MakeCallback (&RxDrop));} in the
script to get rid of the drop prints just as easily. 

You can now run gnuplot (if you have it installed) and tell it to generate some 
pretty pictures:

@verbatim
  gnuplot> set terminal png size 640,480
  gnuplot> set output "cwnd.png"
  gnuplot> plot "cwnd.dat" using 1:2 title 'Congestion Window' with linespoints
  gnuplot> exit
@end verbatim

You should now have a graph of the congestion window versus time sitting in the 
file ``cwnd.png'' in all of its glory, that looks like:

@sp 1
@center @image{figures/cwnd,,,,png}