[simgrid.git] / doc / FAQ.doc

/*! \page faq Frequently Asked Questions

\htmlinclude .FAQ.doc.toc

\section faq_installation Installing the SimGrid library

Many people have been asking me questions on how to use SimGrid. Quite
often, the questions were not really about SimGrid but on the
installation process. This section is intended to help people that are
not familiar with compiling C files under UNIX. If you follow these
instructions and still have some troubles, drop an e-mail to
<simgrid-user@lists.gforge.inria.fr>.

\subsection faq_compiling Compiling SimGrid from an archive

First of all, you need to download the latest version of SimGrid from 
<a href="http://gforge.inria.fr/frs/?group_id=12">here</a>.
Suppose you have uncompressed SimGrid in some temporary location of
your home directory (say <tt>/home/joe/tmp/simgrid-3.0.1 </tt>). The
simplest way to use SimGrid is to install it in your home
directory. Change your directory to
<tt>/home/joe/tmp/simgrid-3.0.1</tt> and type

\verbatim./configure --prefix=$HOME
make
make install
\endverbatim

If at some point, something fails, check the section "\ref
faq_compil_trouble". If it does not help, you can report this problem to the
list but, please, avoid sending a laconic mail like "There is a problem. Is it
okay?". Send the config.log file which is automatically generated by
configure. Try to capture both the standard output and the error output of the
<tt>make</tt> command with <tt>script</tt>. There is no way for us to help you
without the relevant bits of information.

Now, the following directory should have been created : 

      \li <tt>/home/joe/doc/simgrid/html/</tt>
      \li <tt>/home/joe/lib/</tt>
      \li <tt>/home/joe/include/</tt>

SimGrid is not a binary, it is a library. Both a static and a dynamic
version are available. Here is what you can find if you try a <tt>ls
/home/joe/lib</tt>:

\verbatim libsimgrid.a libsimgrid.la libsimgrid.so libsimgrid.so.0 libsimgrid.so.0.0.1
\endverbatim

Thus, there is two ways to link your program with SimGrid:
      \li Either you use the static version, e.g 
\verbatim gcc libsimgrid.a -o MainProgram MainProgram.c
\endverbatim
          In this case, all the SimGrid functions are directly
          included in <tt>MainProgram</tt> (hence a bigger binary).
      \li Either you use the dynamic version (the preferred method)
\verbatim gcc -lsimgrid -o MainProgram MainProgram.c
\endverbatim
          In this case, the SimGrid functions are not included in
          <tt>MainProgram</tt> and you need to set your environment
          variable in such a way that <tt>libsimgrid.so</tt> will be
          found at runtime. This can be done by adding the following
          line in your .bashrc (if you use bash and if you have
          installed the SimGrid libraries in your home directory):
\verbatim export LD_LIBRARY_PATH=$HOME/lib/:$LD_LIBRARY_PATH
\endverbatim


\subsection faq_compiling_cvs Compiling SimGrid from the CVS

The project development takes place in the cvs, where all changes are
commited when they happen. Then every once in a while, we make sure that the
code quality meets our standard and release an archive from the code in the
CVS. We afterward go back to the development in the CVS. So, if you need a
recently added feature and can afford some little problem with the stability
of the lastest features, you may want to use the CVS version instead of a
released one.

For that, you first need to get the "simgrid" module from
<a href="http://gforge.inria.fr/scm/?group_id=12">here</a>. 

You won't find any <tt>configure</tt> and a few other things
(<tt>Makefile.in</tt>'s, documentation, ...) will be missing as well. The
reason for that is that all these files have to be regenerated using the
latest versions of <tt>autoconf</tt>, <tt>libtool</tt>, <tt>automake</tt>
(>1.9) and <tt>doxygen</tt> (>1.4). To generate the <tt>configure</tt> and
the <tt>Makefile.in</tt>'s, you just have to launch the <tt>bootstrap</tt>
command that resides in the top of the source tree. Then just follow the
instructions of Section \ref faq_compiling.

We insist on the fact that you really need the latest versions of
autoconf and automake. Doing this step on exotic architectures/systems
(i.e. anything different from a recent linux distribution) may be
... uncertain. If you want to use the CVS version on another
architecture/system, you should do the previous steps on a perfectly
standard box, then do a <tt>make dist</tt> that will build you a
perfectly portable SimGrid archive.

In summary, the following commands will checkout the CVS, regenerate the
configure script and friends, configure SimGrid and build an archive you can
use on another machine afterward.

\verbatim cvs -d :pserver:anonymous@scm.gforge.inria.fr:/cvsroot/simgrid login
cvs -d :pserver:anonymous@scm.gforge.inria.fr:/cvsroot/simgrid checkout simgrid
cd simgrid
./bootstrap
./configure --enable-maintainer-mode
make dist \endverbatim

\subsection faq_setting_MSG Setting up your own MSG code

Do not build your simulator by modifying the SimGrid examples.  Go
outside the SimGrid source tree and create your own working directory
(say <tt>/home/joe/SimGrid/MyFirstScheduler/</tt>).

Suppose your simulation has the following structure (remember it is
just an example to illustrate a possible way to compile everything;
feel free to organize it as you want).

      \li <tt>sched.h</tt>: a description of the core of the
          scheduler (i.e. which functions are can be used by the
          agents). For example we could find the following functions
          (master, forwarder, slave).

      \li <tt>sched.c</tt>: a C file including <tt>sched.h</tt> and
          implementing the core of the scheduler. Most of these
          functions use the MSG functions defined in section \ref
          msg_gos_functions.

      \li <tt>masterslave.c</tt>: a C file with the main function, i.e.
          the MSG initialization (MSG_global_init()), the platform
          creation (e.g. with MSG_create_environment()), the
          deployment phase (e.g. with MSG_function_register() and
          MSG_launch_application()) and the call to
          MSG_main()).

To compile such a program, we suggest to use the following
Makefile. It is a generic Makefile that we have used many times with
our students when we teach the C language.

\verbatim
all: masterslave 
masterslave: masterslave.o sched.o

INSTALL_PATH = $$HOME
CC = gcc
PEDANTIC_PARANOID_FREAK =       -O0 -Wshadow -Wcast-align \
                                -Waggregate-return -Wmissing-prototypes -Wmissing-declarations \
                                -Wstrict-prototypes -Wmissing-prototypes -Wmissing-declarations \
                                -Wmissing-noreturn -Wredundant-decls -Wnested-externs \
                                -Wpointer-arith -Wwrite-strings -finline-functions
REASONABLY_CAREFUL_DUDE =       -Wall
NO_PRAYER_FOR_THE_WICKED =      -w -O2 
WARNINGS =                      $(REASONABLY_CAREFUL_DUDE)
CFLAGS = -g $(WARNINGS)

INCLUDES = -I$(INSTALL_PATH)/include
DEFS = -L$(INSTALL_PATH)/lib/
LDADD = -lm -lsimgrid 
LIBS = 

%: %.o
        $(CC) $(INCLUDES) $(DEFS) $(CFLAGS) $^ $(LIBS) $(LDADD) -o $@ 

%.o: %.c
        $(CC) $(INCLUDES) $(DEFS) $(CFLAGS) -c -o $@ $<

clean:
        rm -f $(BIN_FILES) *.o *~
.SUFFIXES:
.PHONY : clean

\endverbatim

The first two lines indicates what should be build when typing make
(<tt>masterslave</tt>) and of which files it is to be made of
(<tt>masterslave.o</tt> and <tt>sched.o</tt>). This makefile assumes
that you have set up correctly your <tt>LD_LIBRARY_PATH</tt> variable
(look, there is a <tt>LDADD = -lm -lsimgrid</tt>). If you prefer using
the static version, remove the <tt>-lsimgrid</tt> and add a
<tt>$(INSTALL_PATH)/lib/libsimgrid.a</tt> on the next line, right
after the <tt>LIBS = </tt>.

More generally, if you have never written a Makefile by yourself, type
in a terminal : <tt>info make</tt> and read the introduction. The
previous example should be enough for a first try but you may want to
perform some more complex compilations...

\subsection faq_setting_GRAS Setting up your own GRAS code

If you use the GRAS interface instead of the MSG one, then previous section
is not the better source of information. Instead, you should check the GRAS
tutorial in general, and the \ref GRAS_tut_tour_setup in particular.

\subsection faq_crosscompile Cross-compiling a Windows DLL of SimGrid from linux

At the moment, we do not distribute Windows pre-compiled version of SimGrid
because the support for this platform is still experimental. We know that
some parts of the GRAS environment do not work, and we think that the others
environments (MSG and SD) have good chances to work, but we didn't test
ourselves. This section explains how we generate the SimGrid DLL so that you
can build it for yourself. First of all, you need to have a version more
recent than 3.1 (ie, a CVS version as time of writting).

In order to cross-compile the package to windows from linux, you need to
install mingw32 (minimalist gnu win32). On Debian, you can do so by
installing the packages mingw32 (compiler), mingw32-binutils (linker and
so), mingw32-runtime.

You can use the VPATH support of configure to compile at the same time for
linux and windows without dupplicating the source nor cleaning the tree
between each. Just run bootstrap (if you use the CVS) to run the autotools.
Then, create a linux and a win directories. Then, type:
\verbatim  cd linux; ../configure --srcdir=.. <usual configure flags>; make; cd ..
cd win;  ../configure --srcdir=.. --host=i586-mingw32msvc <flags>; make; cd ..
\endverbatim
The trick to VPATH builds is to call configure from another directory,
passing it an extra --srcdir argument to tell it where all the sources are.
It will understand you want to use VPATH. Then, the trick to cross-compile
is simply to add a --host argument specifying the target you want to build
for. The i586-mingw32msvc string is what you have to pass to use the mingw32
environment as distributed in Debian.

After that, you can run all make targets from both directories, and test
easily that what you change for one arch does not break the other one. 

It is possible that this VPATH build thing breaks from time to time in the
CVS since it's quite fragile, but it's granted to work in any released
version. If you experience problems, drop us a mail. 

Another possible source of issue is that at the moment, building the
examples request to use the gras_stub_generator tool, which is a compiled
program, not a script. In cross-compilation, you need to cross-execute with
wine for example, which is not really pleasant. We are working on this, but
in the meanwhile, simply don't build the examples in cross-compilation
(<tt>cd src</tt> before running make).
    
Program (cross-)compiled with mingw32 do request an extra DLL at run-time to be
usable. For example, if you want to test your build with wine, you should do
the following to put this library where wine looks for DLLs.
\verbatim  cp /usr/share/doc/mingw32-runtime/mingwm10.dll.gz ~/.wine/c/windows/system/
gunzip ~/.wine/c/windows/system/mingwm10.dll.gz
\endverbatim

The DLL is builded in src/.libs, and installed in the <prefix>/bin directory
when you run make install. 

If you want to use it in a native project on windows, you need to use 
simgrid.dll and mingwm10.dll. For each DLL, you need to build .def file
under linux (listing the defined symbols), and convert it into a .lib file
under windows (specifying this in a way that windows compilers like). To
generate the def files, run (under linux):
\verbatim echo "LIBRARY libsimgrid-0.dll" > simgrid.def
echo EXPORTS >> simgrid.def
nm libsimgrid-0.dll | grep ' T _' | sed 's/.* T _//' >> simgrid.def
nm libsimgrid-0.dll | grep ' D _' | sed 's/.* D _//' | sed 's/$/ DATA/' >> simgrid.def

echo "LIBRARY mingwm10.dll" > mingwm10.def
echo EXPORTS >> mingwm10.def
nm mingwm10.dll | grep ' T _' | sed 's/.* T _//' >> mingwm10.def
nm mingwm10.dll | grep ' D _' | sed 's/.* D _//' | sed 's/$/ DATA/' >> mingwm10.def
\endverbatim

To create the import .lib files, use the <tt>lib</tt> windows tool (from
MSVC) the following way to produce simgrid.lib and mingwm10.lib
\verbatim lib /def:simgrid.def
lib /def:mingwm10.def
\endverbatim

If you happen to use Borland C Builder, the right command line is the
following (note that you don't need any file.def to get this working).
\verbatim implib simgrid.lib libsimgrid-0.dll
implib mingwm10.lib mingwm10.dll
\endverbatim

Then, set the following parameters in Visual C++ 2005:
Linker -> Input -> Additional dependencies = simgrid.lib mingwm10.lib

Just in case you wonder how to generate a DLL from libtool in another
project, we added -no-undefined to any lib*_la_LDFLAGS variables so that
libtool accepts to generate a dynamic library under windows. Then, to make
it true, we pass any dependencies (such as -lws2 under windows or -lpthread
on need) on the linking line. Passing such deps is a good idea anyway so
that they get noted in the library itself, avoiding the users to know about
our dependencies and put them manually on their compilation line. Then we
added the AC_LIBTOOL_WIN32_DLL macro just before AC_PROG_LIBTOOL in the
configure.ac. It means that we exported any symbols which need to be.
Nowadays, functions get automatically exported, so we don't need to load our
header files with tons of __declspec(dllexport) cruft. We only need to do so
for data, but there is no public data in SimGrid so we are good.

\section faq_simgrid I'm new to SimGrid. I have some questions. Where should I start?

You are at the right place... Having a look to these
<a href="http://graal.ens-lyon.fr/~alegrand/articles/slides_g5k_simul.pdf">slides</a>
(or to these
<a href="http://graal.ens-lyon.fr/~alegrand/articles/Simgrid-Introduction.pdf">"obsolete" slides</a>)
may give you some insights on what SimGrid can help you to do and what
are its limitations. Then you definitely should read the \ref
MSG_examples. There is also a mailing list: <simgrid-user@lists.gforge.inria.fr>.

\subsection faq_generic Building a generic simulator

Please read carefully the \ref MSG_examples. You'll find in \ref
MSG_ex_master_slave a very simple consisting of a master (that owns a bunch of
tasks and distributes them) , some slaves (that process tasks whenever
they receive one) and some forwarder agents (that simply pass the
tasks they receive to some slaves).

\subsection faq_visualization Visualizing the schedule

It is sometime convenient to "see" how the agents are behaving. If you
like colors, you can use <tt>tools/MSG_visualization/colorize.pl </tt>
as a filter to your MSG outputs. It works directly with INFO. Beware,
INFO() prints on stderr. Do not forget to redirect if you want to
filter (e.g. with bash): 
\verbatim 
./msg_test small_platform.xml small_deployment.xml 2>&1 | ../../tools/MSG_visualization/colorize.pl
\endverbatim

We also have a more graphical output. Have a look at MSG_paje_output(). It 
generates an input to <a href="http://www-id.imag.fr/Logiciels/paje/">Paje</a>.
<center>
\htmlonly
 <a href="Paje_MSG_screenshot.jpg"><img src="Paje_MSG_screenshot_thn.jpg"></a>
\endhtmlonly
</center>

\subsection faq_postmortem_analysis Online/postmortem analysis

Vizualization with Paje can be seen as a kind of postmortem
analysis. However, as soon as you start playing with big simulations,
you'll realize that processing such output is kind of tricky. There is
so much generic informations that it is hard to find the information
you are looking for.

As a matter of fact, loging really depends on simulations (e.g. what
kind of events is important...). That is why we do not propose a big
dump of your whole simulation (it would slow everything down) but give
you neat tools to structure you logs. Have a look at \ref XBT_log. In
fact, rather than a post-mortem analysis, you may want to do it on the
fly. The process you are running can do whatever you want. Have you
thought about adding a global structure where you directly compute the
informations that are really important rather than writing everything
down and then processing huge files ?

\subsection faq_C Argh! Do I really have to code in C ?

Up until now, there is no binding for other languages. If you use C++,
you should be able to use the SimGrid library as a standard C library
and everything should work fine (simply <i>link</i> against this
library; recompiling SimGrid with a C++ compiler won't work and it
wouldn't help if you could).

In fact, the bindings needed to allow one to use SimGrid from Perl,
Python, Java, etc. are double-layered.  The first layer would allow
you to call for example the MSG_task_get_name(task) function while
what you really want is a proper object wrapping allowing you to call
task->name(). That's the purpose of the second layer.  The first one
is granted with C++ but can be done with tools like 
<a href="www.swig.org/">swig</a> for other languages like Perl, Ruby,
Python, CAML. None of us really need the second one (which is a bit
more demanding and cannot be automatically generated) yet and there is
no real point in doing the first one without the second. :)

As usual, you're welcome to participate.

\section faq_MIA How to ....? Is there a function in the API to simply ....?

Here is the deal. The whole SimGrid project (MSG, SURF, GRAS, ...) is
meant to be kept as simple and generic as possible. We cannot add
functions for everybody's need when these functions can easily be
built from the ones already in the API. Most of the time, it is
possible and when it was not possible we always have upgraded the API
accordingly. When somebody asks us a question like "How to do that ?
Is there a function in the API to simply do this ?", we're always glad
to answer and help. However if we don't need this code for our own
need, there is no chance we're going to write it... it's your job! :)
The counterpart to our answers is that once you come up with a neat
implementation of this feature (task duplication, RPC, thread
synchronization, ...), you should send it to us and we will be glad to
add it to the distribution. Thus, other people will take advantage of
it (and we don't have to answer this question again and again ;).

You'll find in this section a few "Missing In Action" features. Many
people have asked about it and we have given hints on how to simply do
it with MSG. Feel free to contribute...

\subsection faq_MIA_examples I want some more complex examples!

Many people have come to ask me a more complex example and each time,
they have realized afterward that the basics were in the previous three
examples. 

Of course they have often been needing more complex functions like
MSG_process_suspend(), MSG_process_resume() and
MSG_process_isSuspended() (to perform synchronization), or
MSG_task_Iprobe() and MSG_process_sleep() (to avoid blocking
receptions), or even MSG_process_create() (to design asynchronous
communications or computations). But the examples are sufficient to
start.

We know. We should add some more examples, but not really some more
complex ones... We should add some examples that illustrate some other
functionalities (like how to simply encode asynchronous
communications, RPC, process migrations, thread synchronization, ...)
and we will do it when we will have a little bit more time. We have
tried to document the examples so that they are understandable. Tell
us if something is not clear and once again feel free to participate!
:)

\subsection faq_MIA_taskdup Missing in action: Task duplication/replication

There is no task duplication in MSG. When you create a task, you can
process it or send it somewhere else. As soon as a process has sent
this task, he doesn't have this task anymore. It's gone. The receiver
process has got the task. However, you could decide upon receiving to
create a "copy" of a task but you have to handle by yourself the
semantic associated to this "duplication".

As we already told, we prefer keeping the API as simple as
possible. This kind of feature is rather easy to implement by users
and the semantic you associate really depends on people. Having a
*generic* task duplication mechanism is not that trivial (in
particular because of the data field). That is why I would recommand
that you write it by yourself even if I can give you advice on how to
do it.

You have the following functions to get informations about a task:
MSG_task_get_name(), MSG_task_get_compute_duration(),
MSG_task_get_remaining_computation(), MSG_task_get_data_size(),
and MSG_task_get_data().

You could use a dictionnary (#xbt_dict_t) of dynars (#xbt_dynar_t). If
you still don't see how to do it, please come back to us...

\subsection faq_MIA_asynchronous I want to do asynchronous communications in MSG

Up until now, there is no asynchronous communications in MSG. However,
you can create as many process as you want so you should be able to do
whatever you want... I've written a queue module to help implementing
some asynchronous communications at low cost (creating thousands of
process only to handle communications may be problematic in term of
performance at some point). I'll add it in the distribution asap.

\subsection faq_MIA_thread_synchronization I need to synchronize my MSG processes

You obviously cannot use pthread_mutexes of pthread_conds. The best
thing would be to propose similar structures. Unfortunately, we
haven't found time to do it yet. However you can try to play with
MSG_process_suspend() and MSG_process_resume(). You can even do some
synchronization with fake communications (using MSG_task_get(),
MSG_task_put() and MSG_task_Iprobe()).

\subsection faq_MIA_host_load Where is the get_host_load function hidden in MSG?

There is no such thing because its semantic wouldn't be really
clear. Of course, it is something about the amount of host throughput,
but there is as many definition of "host load" as people asking for
this function. First, you have to remember that resource availability
may vary over time, which make any load notion harder to define.

It may be instantaneous value or an average one. Moreover it may be only the
power of the computer, or may take the background load into account, or may
even take the currently running tasks into account. In some SURF models,
communications have an influence on computational power. Should it be taken
into account too?

So, we decided not to include such a function into MSG and let people do it
thereselves so that they get the value matching exactly what they mean. One
possibility is to run active measurement as in next code snippet. It is very
close from what you would have to do out of the simulator, and thus gives
you information that you could also get in real settings to not hinder the
realism of your simulation. 

\verbatim
double get_host_load() {
   m_task_t task = MSG_task_create("test", 0.001, 0, NULL);
   double date = MSG_get_clock();

   MSG_task_execute(task);
   date = MSG_get_clock() - date;
   MSG_task_destroy(task);
   return (0.001/date);
}
\endverbatim

Of course, it may not match your personal definition of "host load". In this
case, please detail what you mean on the mailing list, and we will extend
this FAQ section to fit your taste if possible.

\subsection faq_MIA_communication_time How can I get the *real* communication time ?  

Communications are synchronous and thus if you simply get the time
before and after a communication, you'll only get the transmission
time and the time spent to really communicate (it will also take into
account the time spent waiting for the other party to be
ready). However, getting the *real* communication time is not really
hard either. The following solution is a good starting point.

\verbatim
int sender()
{
  m_task_t task = MSG_task_create("Task", task_comp_size, task_comm_size, 
                                  calloc(1,sizeof(double)));
  *((double*) task->data) = MSG_get_clock();
  MSG_task_put(task, slaves[i % slaves_count], PORT_22);
  INFO0("Send completed");
  return 0;
}
int receiver()
{
  m_task_t task = NULL;
  double time1,time2;

  time1 = MSG_get_clock();
  a = MSG_task_get(&(task), PORT_22);
  time2 = MSG_get_clock();
  if(time1<*((double *)task->data))
     time1 = *((double *) task->data);
  INFO1("Communication time :  \"%f\" ", time2-time1);
  free(task->data);
  MSG_task_destroy(task);
  return 0;
}
\endverbatim

\subsection faq_MIA_batch_scheduler Is there a native support for batch schedulers in SimGrid ?

No, there is no native support for batch schedulers and none is
planned because this is a very specific need (and doing it in a
generic way is thus very hard). However some people have implemented
their own batch schedulers. Vincent Garonne wrote one during his PhD
and put his code in the contrib directory of our CVS so that other can
keep working on it. You may find inspinring ideas in it.

\subsection faq_MIA_checkpointing I need a checkpointing thing

Actually, it depends on whether you want to checkpoint the simulation, or to
simulate checkpoints. 

The first one could help if your simulation is a long standing process you
want to keep running even on hardware issues. It could also help to
<i>rewind</i> the simulation by jumping sometimes on an old checkpoint to
cancel recent calculations.\n 
Unfortunately, such thing will probably never exist in SG. One would have to
duplicate all data structures because doing a rewind at the simulator level
is very very hard (not talking about the malloc free operations that might
have been done in between). Instead, you may be interested in the Libckpt
library (http://www.cs.utk.edu/~plank/plank/www/libckpt.html). This is the
checkpointing solution used in the condor project, for example. It makes it
easy to create checkpoints (at the OS level, creating something like core
files), and rerunning them on need.

If you want to simulate checkpoints instead, it means that you want the
state of an executing task (in particular, the progress made towards
completion) to be saved somewhere.  So if a host (and the task executing on
it) fails (cf. #MSG_HOST_FAILURE), then the task can be restarted
from the last checkpoint.\n

Actually, such a thing does not exists in SimGrid either, but it's just
because we don't think it is fundamental and it may be done in the user code
at relatively low cost. You could for example use a watcher that
periodically get the remaining amount of things to do (using
MSG_task_get_remaining_computation()), or fragment the task in smaller
subtasks.

\section faq_SG Where has SG disappeared?!?

OK, it's time to explain what's happening to the SimGrid project. Let's
start with a little bit of history.

* Historically, SimGrid was a low-level toolkit for scheduling with
classical models such as DAGs. That was SimGrid v.1.* aka SG, written
by Henri Casanova. I (Arnaud) had been using it in its earliest
versions during an internship at UCSD.

Then we have realized that encoding distributed algorithm in SG was a
real pain.

* So we have built MSG on top of SG and have released SimGrid v.2.*. MSG
offered a very basic API to encode a distributed application easily.
However encoding MSG on top of SG was not really convenient and did not
use the DAG part since the control of the task synchronization was done
on top of MSG and no more in SG. We have been playing a little bit with
MSG. We have realized that:

   \li 1) the platform modeling was quite flexible and could be "almost"
       automated (e.g. using random generator and post-annotations);

   \li 2) SG was the bottleneck because of the way we were using
       it. We needed to simulate concurrent transfers, complex load
       sharing mechanisms. Many optimizations (e.g. trace integration)
       were totally inefficient when combined with MSG and made extending SG
       to implement new sharing policies, parallel tasks models, or failures
       (many people were asking for these kind of features) a real pain;

   \li 3) the application modeling was not really easy. Even though the
       application modeling depends on people's applications, we thought
       we could improve things here. One of our target here was realistic
distributed applications ranging from computer sensor networks like
the NWS to peer-to-peer applications;

* So we have been planning mainly two things for SimGrid 3:

   \li 1) I have proposed to get rid of SG and to re-implement a new kernel
       that would be faster and more flexible. That is what I did in the
end of 2004: SURF. SURF is based on a fast max-min linear solver
using O(1) data-structures. I have quickly replaced SG by SURF in
MSG and the result has been that on the MSG example, the new
version was more than 10 times faster while we had gain a lot of
flexibility. I think I could still easily make MSG faster but I
have to work on MSG now (e.g. using some of the O(1)
data-structures I've been using to build SURF) since it has become
the bottleneck. Some MSG functions have been removed from the API
but they were mainly intended to build the platform by hand (they
had appeared in the earliest versions of MSG) and were therefore
not useful anymore since we are providing a complete mechanism to
automatically build the platform and deploy the agents on it.;

   \li 2) GRAS is a new project Martin and I have come up with. The idea is
to have a programming environment that let you program real
distributed applications while letting you the ability to run it in
the simulator without having to change the slightest line of your
code. From the simulation point of view, GRAS performs the
application modeling automatically... Up until now, GRAS works on
top MSG for historical reasons but I'm going to make it work
directly on top of SURF so that it can use all the flex and the
speed provided by SURF.

Those two things are working, but we want to make everything as clean as
possible before releasing SimGrid v.3.

So what about those nice DAGs we used to have in SimGrid v.1.? They're
not anymore in SimGrid v.3. At least not in their original form... Let
me recall you the way SimGrid 3 is organized:

\verbatim
________________
|   User code  |
|______________|
| | MSG | GRAS |
| -------------|
| |   SURF     |
| -------------|
|     XBT      |
----------------
\endverbatim

XBT is our tool box and now, you should have an idea of what the other
ones are. As you can see, the primitive SG is not here
anymore. However we have written a brand new and cleaner API for this
purpose: \ref SD_API. It is built directly on top of SURF and provides
an API rather close to the old SG:

\verbatim
______________________
|    User code       |
|____________________|
| | MSG | GRAS | SD  |
| -------------------|
| |      SURF        |
| -------------------|
|        XBT         |
----------------------
\endverbatim

The nice thing is that, as it is writen on top of SURF, it seamlessly
support DAG of parallel tasks as well as complex communications
patterns. Some old codes using SG are currently under rewrite using
\ref SD_API to check that all needful functions are provided.

\subsection faq_SG_DAG How to implement a distributed dynamic scheduler of DAGs.

Distributed is somehow "contagious". If you start making distributed
decisions, there is no way to handle DAGs directly anymore (unless I
am missing something). You have to encode your DAGs in term of
communicating process to make the whole scheduling process
distributed. Here is an example of how you could do that. Assume T1
has to be done before T2.

\verbatim
 int your_agent(int argc, char *argv[] {
   ...
   T1 = MSG_task_create(...);
   T2 = MSG_task_create(...);
   ...
   while(1) {
     ...
     if(cond) MSG_task_execute(T1);
     ...
     if((MSG_task_get_remaining_computation(T1)=0.0) && (you_re_in_a_good_mood))
        MSG_task_execute(T2)
     else {
        /* do something else */
     }
   }
 }
\endverbatim
 
If you decide that the distributed part is not that much important and that
DAG is really the level of abstraction you want to work with, then you should
give a try to \ref SD_API.

\section faq_dynamic Dynamic resources and platform building

\subsection faq_platform Building a realistic platform

We can speak more than an hour on this subject and we still do not have
the right answer, just some ideas. You can read the following
<a href="http://graal.ens-lyon.fr/~alegrand/articles/Simgrid-Introduction.pdf">slides</a>.
It may give you some hints. You can also have a look at the
<tt>tools/platform_generation/</tt> directory. There is a perl-script
we use to annotate a Tiers generated platform.

\subsection faq_SURF_dynamic How can I have variable resource availability?

A nice feature of SimGrid is that it enables you to seamlessly have
resources whose availability change over time. When you build a
platform, you generally declare CPUs like that:

\verbatim
  <cpu name="Cpu A" power="100.00"/>
\endverbatim 

If you want the availability of "CPU A" to change over time, the only
thing you have to do is change this definition like that:

\verbatim
  <cpu name="Cpu A" power="100.00" availability_file="trace_A.txt" state_file="trace_A_failure.txt"/>
\endverbatim

For CPUs, availability files are expressed in fraction of available
power. Let's have a look at what "trace_A.txt" may look like:

\verbatim
PERIODICITY 1.0
0.0 1.0
11.0 0.5
20.0 0.9
\endverbatim

At time 0, our CPU will deliver 100 Mflop/s. At time 11.0, it will
deliver only 50 Mflop/s until time 20.0 where it will will start
delivering 90 Mflop/s. Last at time 21.0 (20.0 plus the periodicity
1.0), we'll be back to the beginning and it will deliver 100Mflop/s.

Now let's look at the state file:
\verbatim
PERIODICITY 10.0
1.0 -1.0
2.0 1.0
\endverbatim

A negative value means "off" while a positive one means "on". At time
1.0, the CPU is on. At time 1.0, it is turned off and at time 2.0, it
is turned on again until time 12 (2.0 plus the periodicity 10.0). It
will be turned on again at time 13.0 until time 23.0, and so on.

Now, let's look how the same kind of thing can be done for network
links. A usual declaration looks like:

\verbatim
  <network_link name="LinkA" bandwidth="10.0" latency="0.2"/>
\endverbatim

You have at your disposal the following options: bandwidth_file,
latency_file and state_file. The only difference with CPUs is that
bandwidth_file and latency_file do not express fraction of available
power but are expressed directly in Mb/s and seconds.

\subsection faq_flexml_bypassing How can I have some C functions do what the platform file does?

So you want to bypass the XML files parser, uh? Maybe doin some parameter
sweep experiments on your simulations or so? This is possible, but it's not
really easy. Here is how it goes.

For this, you have to first remember that the XML parsing in SimGrid is done
using a tool called FleXML. Given a DTD, this gives a flex-based parser. If
you want to bypass the parser, you need to provide some code mimicking what
it does and replacing it in its interactions with the SURF code. So, let's
have a look at these interactions.

FleXML parser are close to classical SAX parsers. It means that a
well-formed SimGrid platform XML file might result in the following
"events":

  - start "platform_description"
  - start "cpu" with attributes name="host1" power="1.0"
  - end "cpu"
  - start "cpu" with attributes name="host2" power="2.0"
  - end "cpu"
  - start "network_link" with ...
  - end "network_link"
  - start "route" with ...
  - end "route"
  - start "route" with ...
  - end "route"
  - end "platform_description"

The communication from the parser to the SURF code uses two means:
Attributes get copied into some global variables, and a surf-provided
function gets called by the parser for each event. For example, the event
  - start "cpu" with attributes name="host1" power="1.0"

let the parser do the equivalent of:
\verbatim
  strcpy("host1",A_cpu_name);
  A_cpu_power = 1.0;
  (*STag_cpu_fun)();
\endverbatim

In SURF, we attach callbacks to the different events by initializing the
pointer functions to some the right surf functions. Example in
workstation_KCCFLN05.c (surf_parse_open() ends up calling surf_parse()):
\verbatim
  // Building the routes
  surf_parse_reset_parser();
  STag_route_fun=parse_route_set_endpoints;
  ETag_route_element_fun=parse_route_elem;
  ETag_route_fun=parse_route_set_route;
  surf_parse_open(file);
  xbt_assert1((!surf_parse()),"Parse error in %s",file);
  surf_parse_close();
\endverbatim
    
So, to bypass the FleXML parser, you need to write your own version of the
surf_parse function, which should do the following:
   - Call the corresponding STag_<tag>_fun function to simulate tag start
   - Fill the A_<tag>_<attribute> variables with the wanted values
   - Call the corresponding ETag_<tag>_fun function to simulate tag end
   - (do the same for the next set of values, and loop)

Then, tell SimGrid that you want to use your own "parser" instead of the stock one:
\verbatim
  surf_parse = surf_parse_bypass;
  MSG_create_environment(NULL);
\endverbatim

An example of this trick is distributed in the file examples/msg/msg_test_surfxml_bypassed.c

\section faq_limits Pushing the limits

\subsection faq_context_1000 I want thousands of simulated processes

SimGrid can use either pthreads library or the UNIX98 contextes. On most
systems, the number of pthreads is limited and then your simulation may be
limited for a stupid reason. This is especially true with the current linux
pthreads, and I cannot get more than 2000 simulated processes with pthreads
on my box. The UNIX98 contexts allow me to raise the limit to 25,000
simulated processes on my laptop.

The <tt>--with-context</tt> option of the <tt>./configure</tt> script allows
you to choose between UNIX98 contextes (<tt>--with-context=ucontext</tt>)
and the pthread version ( (<tt>--with-context=pthread</tt>). The default
value is ucontext when the script detect a working UNIX98 context
implementation. On Windows boxes, the provided value is discarded and an
adapted version is picked up.

We experienced some issues with contextes on some rare systems (solaris 8
and lower or old alpha linuxes comes to mind). The main problem is that the
configure script detect the contextes as being functional when it's not
true. If you happen to use such a system, switch manually to the pthread
version, and provide us with a good patch for the configure script so that
it is done automatically ;)

\subsection faq_context_10000 I want hundred thousands of simulated processes

As explained above, SimGrid can use UNIX98 contextes to represent and handle
the simulated processes. Thanks to this, the main limitation to the number
of simulated processes becomes the available memory. 

Here are some tricks I had to use in order to run a token ring between
25,000 processes on my laptop (1Gb memory, 1.5Gb swap).

 - First of all, make sure your code runs for a few hundreds processes
   before trying to push the limit. Make sure it's valgrind-clean, ie that
   valgrind does not report neither memory error nor memory leaks. Indeed,
   numerous simulated processes result in *fat* simulation hindering debugging.

 - It was really boring to write 25,000 entries in the deployment file, so I wrote 
   a little script <tt>examples/gras/tokenS/make_deployment.pl</tt>, which you may
   want to adapt to your case. You could also think about hijacking
   the SURFXML parser (have look at \ref faq_flexml_bypassing).

 - The deployment file became quite big, so I had to do what is in the FAQ
   entry \ref faq_flexml_limit

 - Each UNIX98 context has its own stack entry. As debugging this is quite 
   hairly, the default value is a bit overestimated so that user don't get 
   into trouble about this. You want to tune this size to increse the number 
   of processes. This is the <tt>STACK_SIZE</tt> define in 
   <tt>src/xbt/context_private.h</tt>, which is 128kb by default.
   Reduce this as much as you can, but be warned that if this value is too 
   low, you'll get a segfault. The token ring example, which is quite simple, 
   runs with 40kb stacks.

\section faq_troubleshooting Troubleshooting

\subsection faq_compil_trouble ./configure fails!

We now only one reason for the configure to fail:

 - <b>You are using a borken build environment</b>\n
   If symptom is that configure complains about gcc not being able to build
   executables, you are probably missing the libc6-dev package. Damn Ubuntu.

If you experience other kind of issue, please get in touch with us. We are
always interested in improving our portability to new systems.

\subsection faq_distcheck_fails Dude! "make check" fails on my machine!

Don't assume we never run this target, because we do. Really. Promise!

There is several reasons which may cause the make check to fail on your
machine:

 - <b>You are using a borken libc (probably concerning the contextes)</b>.\n
   The symptom is that the "make check" fails within the examples/msg directory.\n
   By default, SimGrid uses something called ucontexts. This is part of the
   libc, but it's quite undertested. For example, some (old) versions of the
   glibc on alpha do not implement these functions, but provide the stubs
   (which return ENOSYS: not implemented). It fools our detection mecanism
   and leads to segfaults.\n
   On some x86_64, the pointer to function is stored into a integer, but int
   are 32bits only on this arch while pointers are 64bits. Our detection
   mecanism also fails to detect the problem, which leads to segfaults.\n
   In both cases, there is not much we can do to fix the bug. We are working
   on a workaround for x86_64 machines, but in the meanwhile, you can
   compile with --with-context=pthread to avoid ucontext completely. You'll
   be a bit more limitated in the number of simulated processes you can start
   concurently, but 5000 processes is still enough for most purposes, isn't
   it?\n
   This limitation is the reason why we insist on using this piece of ...
   software even if it's so troublesome.\n
   <b>=> use --with-pthread on AMD64 architecture that do not have an 
   ultra-recent libc.</b>
   
 - <b>There is a bug in SimGrid we aren't aware of</b>.\n
   If none of the above apply, please drop us a mail on the mailing list so
   that we can check it out.

\subsection faq_longjmp longjmp madness in valgrind

This is when valgrind starts complaining about longjmp things, just like:

\verbatim ==21434== Conditional jump or move depends on uninitialised value(s)
==21434==    at 0x420DBE5: longjmp (longjmp.c:33)
==21434==
==21434== Use of uninitialised value of size 4
==21434==    at 0x420DC3A: __longjmp (__longjmp.S:48)
\endverbatim

or even when it reports scary things like:

\verbatim ==24023== Warning: client switching stacks?  SP change: 0xBE3FF618 --> 0xBE7FF710
x86->IR: unhandled instruction bytes: 0xF4 0xC7 0x83 0xD0
==24023==          to suppress, use: --max-stackframe=4194552 or greater
==24023== Your program just tried to execute an instruction that Valgrind
==24023== did not recognise.  There are two possible reasons for this.
==24023== 1. Your program has a bug and erroneously jumped to a non-code
==24023==    location.  If you are running Memcheck and you just saw a
==24023==    warning about a bad jump, it's probably your program's fault.
==24023== 2. The instruction is legitimate but Valgrind doesn't handle it,
==24023==    i.e. it's Valgrind's fault.  If you think this is the case or
==24023==    you are not sure, please let us know.
==24023== Either way, Valgrind will now raise a SIGILL signal which will
==24023== probably kill your program.
==24023==
==24023== Process terminating with default action of signal 4 (SIGILL)
==24023==  Illegal opcode at address 0x420D234
==24023==    at 0x420D234: abort (abort.c:124)
\endverbatim

This is the sign that you didn't used the exception mecanism well. Most
probably, you have a <tt>return;</tt> somewhere within a <tt>TRY{}</tt>
block. This is <b>evil</b>, and you must not do this. Did you read the section
about \ref XBT_ex??

\subsection faq_valgrind Valgrind spits tons of errors!

It may happen that valgrind, the memory debugger beloved by any decent C
programmer, spits tons of warnings like the following :
\verbatim ==8414== Conditional jump or move depends on uninitialised value(s)
==8414==    at 0x400882D: (within /lib/ld-2.3.6.so)
==8414==    by 0x414EDE9: (within /lib/tls/i686/cmov/libc-2.3.6.so)
==8414==    by 0x400B105: (within /lib/ld-2.3.6.so)
==8414==    by 0x414F937: _dl_open (in /lib/tls/i686/cmov/libc-2.3.6.so)
==8414==    by 0x4150F4C: (within /lib/tls/i686/cmov/libc-2.3.6.so)
==8414==    by 0x400B105: (within /lib/ld-2.3.6.so)
==8414==    by 0x415102D: __libc_dlopen_mode (in /lib/tls/i686/cmov/libc-2.3.6.so)
==8414==    by 0x412D6B9: backtrace (in /lib/tls/i686/cmov/libc-2.3.6.so)
==8414==    by 0x8076446: xbt_dictelm_get_ext (dict_elm.c:714)
==8414==    by 0x80764C1: xbt_dictelm_get (dict_elm.c:732)
==8414==    by 0x8079010: xbt_cfg_register (config.c:208)
==8414==    by 0x806821B: MSG_config (msg_config.c:42)
\endverbatim

This problem is somewhere in the libc when using the backtraces and there is
very few things we can do ourselves to fix it. Instead, here is how to tell
valgrind to ignore the error. Add the following to your ~/.valgrind.supp (or
create this file on need). Make sure to change the obj line according to
your personnal mileage (change 2.3.6 to the actual version you are using,
which you can retrieve with a simple "ls /lib/ld*.so").

\verbatim {
   name: Backtrace madness
   Memcheck:Cond
   obj:/lib/ld-2.3.6.so
   fun:dl_open_worker
   fun:_dl_open
   fun:do_dlopen
   fun:dlerror_run
   fun:__libc_dlopen_mode
}\endverbatim

Then, you have to specify valgrind to use this suppression file by passing
the <tt>--suppressions=$HOME/.valgrind.supp</tt> option on the command line.
You can also add the following to your ~/.bashrc so that it gets passed
automatically. Actually, it passes a bit more options to valgrind, and this
happen to be my personnal settings. Check the valgrind documentation for
more information.

\verbatim export VALGRIND_OPTS="--leak-check=yes --leak-resolution=high --num-callers=40 --tool=memcheck --suppressions=$HOME/.valgrind.supp" \endverbatim

\subsection faq_flexml_limit I get the message "surf_parse_lex: Assertion `next&lt;limit' failed."

This is because your platform file is too big for the parser. 

Actually, the message comes directly from FleXML, the technology on top of
which the parser is built. FleXML has the bad idea of fetching the whole
document in memory before parsing it. And moreover, the memory buffer size
must be determinded at compilation time.

We use a value which seems big enough for our need without bloating the
simulators footprints. But of course your mileage may vary. In this case,
just edit src/surf/surfxml.l modify the definition of
FLEXML_BUFFERSTACKSIZE. E.g.

\verbatim
#define FLEXML_BUFFERSTACKSIZE 1000000000
\endverbatim

Then recompile and everything should be fine, provided that your version of
Flex is recent enough (>= 2.5.31). If not the compilation process should
warn you.

A while ago, we worked on FleXML to reduce a bit its memory consumtion, but
these issues remain. There is two things we should do:

  - use a dynamic buffer instead of a static one so that the only limit
    becomes your memory, not a stupid constant fixed at compilation time
    (maybe not so difficult).
  - change the parser so that it does not need to get the whole file in
    memory before parsing
    (seems quite difficult, but I'm a complete newbe wrt flex stuff).

These are changes to FleXML itself, not SimGrid. But since we kinda hijacked
the development of FleXML, I can grant you that any patches would be really
welcome and quickly integrated.

\subsection faq_gras_transport GRAS spits networking error messages

Gras, on real platforms, naturally use regular sockets to communicate. They
are deeply hiden in the gras abstraction, but when things go wrong, you may
get some weird error messages. Here are some example, with the probable
reason:

 - <b>Transport endpoint is not connected</b>: several processes try to open
   a server socket on the same port number of the same machine. This is
   naturally bad and each process should pick its own port number for this.\n
   Maybe, you just have some processes remaining from a previous experiment 
   on your machine.\n
   Killing them may help, but again if you kill -KILL them, you'll have to
   wait for a while: they didn't close there sockets properly and the system
   needs a while to notice that this port is free again.

 - <b>Socket closed by remote side</b>: if the remote process is not
   supposed to close the socket at this point, it may be dead.
   
 - <b>Connection reset by peer</b>: I found this on internet about this
   error. I think it's what's happening here, too:\n   
   <i>This basically means that a network error occurred while the client was
   receiving data from the server. But what is really happening is that the
   server actually accepts the connection, processes the request, and sends
   a reply to the client. However, when the server closes the socket, the
   client believes that the connection has been terminated abnormally
   because the socket implementation sends a TCP reset segment telling the
   client to throw away the data and report an error.\n
   Sometimes, this problem is caused by not properly closing the
   input/output streams and the socket connection. Make sure you close the
   input/output streams and socket connection properly. If everything is
   closed properly, however, and the problem persists, you can work around
   it by adding a one-second sleep before closing the streams and the
   socket. This technique, however, is not reliable and may not work on all
   systems.</i>\n
   Since GRAS sockets are closed properly (repeat after me: there is no bug
   in GRAS), it is either that you are closing your sockets on server side
   before the client get a chance to read them (use gras_os_sleep() to delay
   the server), or the server died awfully before the client got the data.

\subsection faq_deadlock There is a deadlock !!!

Unfortunately, we cannot debug every code written in SimGrid.  We
furthermore believe that the framework provides ways enough
information to debug such informations yourself. If the textual output
is not enough, Make sure to check the \ref faq_visualization FAQ entry to see
how to get a graphical one.

Now, if you come up with a really simple example that deadlocks and
you're absolutely convinced that it should not, you can ask on the
list. Just be aware that you'll be severely punished if the mistake is
on your side... We have plenty of FAQ entries to redact and new
features to implement for the impenitents! ;)

\subsection faq_big_fat_warning A BIG FAT WARNING is reported telling me that my platform and deployment files are too old.

We have decided to change the units in SimGrid. Now we use Bytes, Flops and
seconds instead of MBytes, MFlops and seconds... Units should be updated
accordingly and the version of platform_description should be set to a
valuer greater than 1:
\verbatim
  <platform_description version="1">
\endverbatim
You should try to use the surfxml_update.pl script that can be found
<a href="http://gforge.inria.fr/plugins/scmcvs/cvsweb.php/contrib/platform_generation/?cvsroot=cvsroot%2Fsimgrid">here</a>.

\subsection faq_surf_network_latency I get weird timings when I play with the latencies.

OK, first of all, remember that units should be Bytes, Flops and
Seconds. If you don't use such units, some SimGrid constants (e.g. the
SG_TCP_CTE_GAMMA constant used in most network models) won't have the
right unit and you'll end up with weird results.

Here is what happens with a single transfer of size L on a link
(bw,lat) when nothing else happens.

\verbatim
0-----lat--------------------------------------------------t
|-----|**** real_bw =min(bw,SG_TCP_CTE_GAMMA/(2*lat)) *****|
\endverbatim

In more complex situations, this min is the solution of a complex
max-min linear system.  Have a look 
<a href="http://lists.gforge.inria.fr/pipermail/simgrid-devel/2006-April/thread.html">here</a>
and read the two threads "Bug in SURF?" and "Surf bug not
fixed?". You'll have a few other examples of such computations. You
can also read "A Network Model for Simulation of Grid Application" by
Henri Casanova and Loris Marchal to have all the details. The fact
that the real_bw is smaller than bw is easy to understand. The fact
that real_bw is smaller than SG_TCP_CTE_GAMMA/(2*lat) is due to the
window-based congestion mechanism of TCP. With TCP, you can't exploit
your huge network capacity if you don't have a good round-trip-time
because of the acks...

Anyway, what you get is t=lat + L/min(bw,SG_TCP_CTE_GAMMA/(2*lat)).

  * if I you set (bw,lat)=(100 000 000, 0.00001), you get t =  1.00001 (you fully
use your link)
  * if I you set (bw,lat)=(100 000 000, 0.0001),  you get t =  1.0001 (you're on the
limit)
  * if I you set (bw,lat)=(100 000 000, 0.001),   you get t = 10.001  (ouch!)

This bound on the effective bandwidth of a flow is not the only thing
that may make your result be unexpected. For example, two flows
competing on a saturated link receive an amount of bandwidth inversely
proportional to their round trip time.

\subsection faq_bugrepport So I've found a bug in SimGrid. How to report it?

We do our best to make sure to hammer away any bugs of SimGrid, but this is
still an academic project so please be patient if/when you find bugs in it.
If you do, the best solution is to drop an email either on the simgrid-user
or the simgrid-devel mailing list and explain us about the issue.  You can
also decide to open a formal bug report using the
<a href="https://gforge.inria.fr/tracker/?atid=165&group_id=12&func=browse">relevant
interface</a>. You need to login on the server to get the ability to submit
bugs. 

We will do our best to solve any problem repported, but you need to help us
finding the issue. Just telling "it segfault" isn't enough. Telling "It
segfaults when running the attached simulator" doesn't really help either.
You may find the following article interesting to see how to repport
informative bug repports:
http://www.chiark.greenend.org.uk/~sgtatham/bugs.html (it is not SimGrid
specific at all, but it's full of good advices).

\author Arnaud Legrand (arnaud.legrand::imag.fr)
\author Martin Quinson (martin.quinson::loria.fr)


*/