1 /*! \page faq Frequently Asked Questions
3 \htmlinclude .FAQ.doc.toc
5 \section faq_installation Installing the SimGrid library
7 Many people have been asking me questions on how to use SimGrid. Quite
8 often, the questions were not really about SimGrid but on the
9 installation process. This section is intended to help people that are
10 not familiar with compiling C files under UNIX. If you follow these
11 instructions and still have some troubles, drop an e-mail to
12 <simgrid-user@lists.gforge.inria.fr>.
14 \subsection faq_compiling Compiling SimGrid
16 First of all, you need to download the latest version of SimGrid from
17 <a href="http://gforge.inria.fr/frs/?group_id=12">here</a>.
18 Suppose you have uncompressed SimGrid in some temporary location of
19 your home directory (say <tt>/home/joe/tmp/simgrid-3.0.1 </tt>). The
20 simplest way to use SimGrid is to install it in your home
21 directory. Change your directory to
22 <tt>/home/joe/tmp/simgrid-3.0.1</tt> and type
24 \verbatim./configure --prefix=$HOME
29 If at some point, something fails, check the section "\ref
30 faq_compil_trouble". If it does not help, you can report this problem to the
31 list but, please, avoid sending a laconic mail like "There is a problem. Is it
32 okay?". Send the config.log file which is automatically generated by
33 configure. Try to capture both the standard output and the error output of the
34 <tt>make</tt> command with <tt>script</tt>. There is no way for us to help you
35 without the relevant bits of information.
37 Now, the following directory should have been created :
39 \li <tt>/home/joe/doc/simgrid/html/</tt>
40 \li <tt>/home/joe/lib/</tt>
41 \li <tt>/home/joe/include/</tt>
43 SimGrid is not a binary, it is a library. Both a static and a dynamic
44 version are available. Here is what you can find if you try a <tt>ls
47 \verbatim libsimgrid.a libsimgrid.la libsimgrid.so libsimgrid.so.0 libsimgrid.so.0.0.1
50 Thus, there is two ways to link your program with SimGrid:
51 \li Either you use the static version, e.g
52 \verbatim gcc libsimgrid.a -o MainProgram MainProgram.c
54 In this case, all the SimGrid functions are directly
55 included in <tt>MainProgram</tt> (hence a bigger binary).
56 \li Either you use the dynamic version (the preferred method)
57 \verbatim gcc -lsimgrid -o MainProgram MainProgram.c
59 In this case, the SimGrid functions are not included in
60 <tt>MainProgram</tt> and you need to set your environment
61 variable in such a way that <tt>libsimgrid.so</tt> will be
62 found at runtime. This can be done by adding the following
63 line in your .bashrc (if you use bash and if you have
64 installed the SimGrid libraries in your home directory):
65 \verbatim export LD_LIBRARY_PATH=$HOME/lib/:$LD_LIBRARY_PATH
69 \subsection faq_compiling_cvs Compiling SimGrid from the CVS
71 First of all, you need to get the "simgrid" module from
72 <a href="http://gforge.inria.fr/scm/?group_id=12">here</a>.
74 You won't find any <tt>configure</tt> and a few other things
75 (<tt>Makefile.in</tt>'s, documentation, ...) will be missing as
76 well. The reason for that is that all these files have to be
77 regenerated using the latest versions of <tt>autoconf</tt>,
78 <tt>automake</tt> (1.9) and <tt>doxygen</tt>. To generate the
79 <tt>configure</tt> and the <tt>Makefile.in</tt>'s, you just have to
80 launch the <tt>bootstrap</tt> command that resides in the top of the
81 source tree. Then just follow the instructions of Section
84 We insist on the fact that you really need the latest versions of
85 autoconf and automake. Doing this step on exotic architectures/systems
86 (i.e. anything different from a recent linux distribution) may be
87 ... uncertain. If you want to use the CVS version on another
88 architecture/system, you should do the previous steps on a perfectly
89 standard box, then do a <tt>make dist</tt> that will build you a
90 perfectly portable SimGrid archive.
92 \subsection faq_setting Setting up your own code
94 Do not build your simulator by modifying the SimGrid examples. Go
95 outside the SimGrid source tree and create your own working directory
96 (say <tt>/home/joe/SimGrid/MyFirstScheduler/</tt>).
98 Suppose your simulation has the following structure (remember it is
99 just an example to illustrate a possible way to compile everything;
100 feel free to organize it as you want).
102 \li <tt>sched.h</tt>: a description of the core of the
103 scheduler (i.e. which functions are can be used by the
104 agents). For example we could find the following functions
105 (master, forwarder, slave).
107 \li <tt>sched.c</tt>: a C file including <tt>sched.h</tt> and
108 implementing the core of the scheduler. Most of these
109 functions use the MSG functions defined in section \ref
112 \li <tt>masterslave.c</tt>: a C file with the main function, i.e.
113 the MSG initialization (MSG_global_init()), the platform
114 creation (e.g. with MSG_create_environment()), the
115 deployment phase (e.g. with MSG_function_register() and
116 MSG_launch_application()) and the call to
119 To compile such a program, we suggest to use the following
120 Makefile. It is a generic Makefile that we have used many times with
121 our students when we teach the C language.
125 masterslave: masterslave.o sched.o
127 INSTALL_PATH = $$HOME
129 PEDANTIC_PARANOID_FREAK = -O0 -Wshadow -Wcast-align \
130 -Waggregate-return -Wmissing-prototypes -Wmissing-declarations \
131 -Wstrict-prototypes -Wmissing-prototypes -Wmissing-declarations \
132 -Wmissing-noreturn -Wredundant-decls -Wnested-externs \
133 -Wpointer-arith -Wwrite-strings -finline-functions
134 REASONABLY_CAREFUL_DUDE = -Wall
135 NO_PRAYER_FOR_THE_WICKED = -w -O2
136 WARNINGS = $(REASONABLY_CAREFUL_DUDE)
137 CFLAGS = -g $(WARNINGS)
139 INCLUDES = -I$(INSTALL_PATH)/include
140 DEFS = -L$(INSTALL_PATH)/lib/
141 LDADD = -lm -lsimgrid
145 $(CC) $(INCLUDES) $(DEFS) $(CFLAGS) $^ $(LIBS) $(LDADD) -o $@
148 $(CC) $(INCLUDES) $(DEFS) $(CFLAGS) -c -o $@ $<
151 rm -f $(BIN_FILES) *.o *~
157 The first two lines indicates what should be build when typing make
158 (<tt>masterslave</tt>) and of which files it is to be made of
159 (<tt>masterslave.o</tt> and <tt>sched.o</tt>). This makefile assumes
160 that you have set up correctly your <tt>LD_LIBRARY_PATH</tt> variable
161 (look, there is a <tt>LDADD = -lm -lsimgrid</tt>). If you prefer using
162 the static version, remove the <tt>-lsimgrid</tt> and add a
163 <tt>$(INSTALL_PATH)/lib/libsimgrid.a</tt> on the next line, right
164 after the <tt>LIBS = </tt>.
166 More generally, if you have never written a Makefile by yourself, type
167 in a terminal : <tt>info make</tt> and read the introduction. The
168 previous example should be enough for a first try but you may want to
169 perform some more complex compilations...
171 \section faq_simgrid I'm new to SimGrid. I have some questions. Where should I start?
173 You are at the right place... Having a look to these
174 <a href="http://graal.ens-lyon.fr/~alegrand/articles/slides_g5k_simul.pdf">slides</a>
176 <a href="http://graal.ens-lyon.fr/~alegrand/articles/Simgrid-Introduction.pdf">"obsolete" slides</a>)
177 may give you some insights on what SimGrid can help you to do and what
178 are its limitations. Then you definitely should read the \ref
179 MSG_examples. There is also a mailing list: <simgrid-user@lists.gforge.inria.fr>.
181 \subsection faq_generic Building a generic simulator
183 Please read carefully the \ref MSG_examples. You'll find in \ref
184 MSG_ex_master_slave a very simple consisting of a master (that owns a bunch of
185 tasks and distributes them) , some slaves (that process tasks whenever
186 they receive one) and some forwarder agents (that simply pass the
187 tasks they receive to some slaves).
189 \subsection faq_visualization Visualizing the schedule
191 It is sometime convenient to "see" how the agents are behaving. If you
192 like colors, you can use <tt>tools/MSG_visualization/colorize.pl </tt>
193 as a filter to your MSG outputs. It works directly with INFO. Beware,
194 INFO() prints on stderr. Do not forget to redirect if you want to
195 filter (e.g. with bash):
197 ./msg_test small_platform.xml small_deployment.xml 2>&1 | ../../tools/MSG_visualization/colorize.pl
200 We also have a more graphical output. Have a look at MSG_paje_output(). It
201 generates an input to <a href="http://www-id.imag.fr/Logiciels/paje/">Paje</a>.
204 <a href="Paje_MSG_screenshot.jpg"><img src="Paje_MSG_screenshot_thn.jpg"></a>
208 \subsection faq_postmortem_analysis Online/postmortem analysis
210 Vizualization with Paje can be seen as a kind of postmortem
211 analysis. However, as soon as you start playing with big simulations,
212 you'll realize that processing such output is kind of tricky. There is
213 so much generic informations that it is hard to find the information
216 As a matter of fact, loging really depends on simulations (e.g. what
217 kind of events is important...). That is why we do not propose a big
218 dump of your whole simulation (it would slow everything down) but give
219 you neat tools to structure you logs. Have a look at \ref XBT_log. In
220 fact, rather than a post-mortem analysis, you may want to do it on the
221 fly. The process you are running can do whatever you want. Have you
222 thought about adding a global structure where you directly compute the
223 informations that are really important rather than writing everything
224 down and then processing huge files ?
226 \subsection faq_C Argh! Do I really have to code in C ?
228 Up until now, there is no binding for other languages. If you use C++,
229 you should be able to use the SimGrid library as a standard C library
230 and everything should work fine (simply <i>link</i> against this
231 library; recompiling SimGrid with a C++ compiler won't work and it
232 wouldn't help if you could).
234 In fact, the bindings needed to allow one to use SimGrid from Perl,
235 Python, Java, etc. are double-layered. The first layer would allow
236 you to call for example the MSG_task_get_name(task) function while
237 what you really want is a proper object wrapping allowing you to call
238 task->name(). That's the purpose of the second layer. The first one
239 is granted with C++ but can be done with tools like
240 <a href="www.swig.org/">swig</a> for other languages like Perl, Ruby,
241 Python, CAML. None of us really need the second one (which is a bit
242 more demanding and cannot be automatically generated) yet and there is
243 no real point in doing the first one without the second. :)
245 As usual, you're welcome to participate.
247 \section faq_MIA How to ....? Is there a function in the API to simply ....?
249 Here is the deal. The whole SimGrid project (MSG, SURF, GRAS, ...) is
250 meant to be kept as simple and generic as possible. We cannot add
251 functions for everybody's need when these functions can easily be
252 built from the ones already in the API. Most of the time, it is
253 possible and when it was not possible we always have upgraded the API
254 accordingly. When somebody asks us a question like "How to do that ?
255 Is there a function in the API to simply do this ?", we're always glad
256 to answer and help. However if we don't need this code for our own
257 need, there is no chance we're going to write it... it's your job! :)
258 The counterpart to our answers is that once you come up with a neat
259 implementation of this feature (task duplication, RPC, thread
260 synchronization, ...), you should send it to us and we will be glad to
261 add it to the distribution. Thus, other people will take advantage of
262 it (and we don't have to answer this question again and again ;).
264 You'll find in this section a few "Missing In Action" features. Many
265 people have asked about it and we have given hints on how to simply do
266 it with MSG. Feel free to contribute...
268 \subsection faq_MIA_examples I want some more complex examples!
270 Many people have come to ask me a more complex example and each time,
271 they have realized afterward that the basics were in the previous three
274 Of course they have often been needing more complex functions like
275 MSG_process_suspend(), MSG_process_resume() and
276 MSG_process_isSuspended() (to perform synchronization), or
277 MSG_task_Iprobe() and MSG_process_sleep() (to avoid blocking
278 receptions), or even MSG_process_create() (to design asynchronous
279 communications or computations). But the examples are sufficient to
282 We know. We should add some more examples, but not really some more
283 complex ones... We should add some examples that illustrate some other
284 functionalities (like how to simply encode asynchronous
285 communications, RPC, process migrations, thread synchronization, ...)
286 and we will do it when we will have a little bit more time. We have
287 tried to document the examples so that they are understandable. Tell
288 us if something is not clear and once again feel free to participate!
291 \subsection faq_MIA_taskdup Missing in action: Task duplication/replication
293 There is no task duplication in MSG. When you create a task, you can
294 process it or send it somewhere else. As soon as a process has sent
295 this task, he doesn't have this task anymore. It's gone. The receiver
296 process has got the task. However, you could decide upon receiving to
297 create a "copy" of a task but you have to handle by yourself the
298 semantic associated to this "duplication".
300 As we already told, we prefer keeping the API as simple as
301 possible. This kind of feature is rather easy to implement by users
302 and the semantic you associate really depends on people. Having a
303 *generic* task duplication mechanism is not that trivial (in
304 particular because of the data field). That is why I would recommand
305 that you write it by yourself even if I can give you advice on how to
308 You have the following functions to get informations about a task:
309 MSG_task_get_name(), MSG_task_get_compute_duration(),
310 MSG_task_get_remaining_computation(), MSG_task_get_data_size(),
311 and MSG_task_get_data().
313 You could use a dictionnary (#xbt_dict_t) of dynars (#xbt_dict_t). If
314 you still don't see how to do it, please come back to us...
316 \subsection faq_MIA_asynchronous I want to do asynchronous communications.
318 Up until now, there is no asynchronous communications in MSG. However,
319 you can create as many process as you want so you should be able to do
320 whatever you want... I've written a queue module to help implementing
321 some asynchronous communications at low cost (creating thousands of
322 process only to handle communications may be problematic in term of
323 performance at some point). I'll add it in the distribution asap.
325 \subsection faq_MIA_thread_synchronization I need to synchronize my processes
327 You obviously cannot use pthread_mutexes of pthread_conds. The best
328 thing would be to propose similar structures. Unfortunately, we
329 haven't found time to do it yet. However you can try to play with
330 MSG_process_suspend() and MSG_process_resume(). You can even do some
331 synchronization with fake communications (using MSG_task_get(),
332 MSG_task_put() and MSG_task_Iprobe()).
334 \subsection faq_MIA_host_load Where is the get_host_load function hidden in MSG?
336 There is no such thing because its semantic wouldn't be really
337 clear. Of course, it is something about the amount of host throughput,
338 but there is as many definition of "host load" as people asking for
339 this function. First, you have to remember that resource availability
340 may vary over time, which make any load notion harder to define.
342 It may be instantaneous value or an average one. Moreover it may be only the
343 power of the computer, or may take the background load into account, or may
344 even take the currently running tasks into account. In some SURF models,
345 communications have an influence on computational power. Should it be taken
348 So, we decided not to include such a function into MSG and let people do it
349 thereselves so that they get the value matching exactly what they mean. One
350 possibility is to run active measurement as in next code snippet. It is very
351 close from what you would have to do out of the simulator, and thus gives
352 you information that you could also get in real settings to not hinder the
353 realism of your simulation.
356 double get_host_load() {
357 m_task_t task = MSG_task_create("test", 0.001, 0, NULL);
358 double date = MSG_get_clock();
360 MSG_task_execute(task);
361 date = MSG_get_clock() - date;
362 MSG_task_destroy(task);
367 Of course, it may not match your personal definition of "host load". In this
368 case, please detail what you mean on the mailing list, and we will extend
369 this FAQ section to fit your taste if possible.
371 \subsection faq_MIA_batch_scheduler Is there a native support for batch schedulers in SimGrid ?
373 No, there is no native support for batch schedulers and none is
374 planned because this is a very specific need (and doing it in a
375 generic way is thus very hard). However some people have implemented
376 their own batch schedulers. Vincent Garonne wrote one during his PhD
377 and put his code in the contrib directory of our CVS so that other can
378 keep working on it. You may find inspinring ideas in it.
380 \subsection faq_MIA_checkpointing I need a checkpointing thing
382 Actually, it depends on whether you want to checkpoint the simulation, or to
383 simulate checkpoints.
385 The first one could help if your simulation is a long standing process you
386 want to keep running even on hardware issues. It could also help to
387 <i>rewind</i> the simulation by jumping sometimes on an old checkpoint to
388 cancel recent calculations.\n
389 Unfortunately, such thing will probably never exist in SG. One would have to
390 duplicate all data structures because doing a rewind at the simulator level
391 is very very hard (not talking about the malloc free operations that might
392 have been done in between). Instead, you may be interested in the Libckpt
393 library (http://www.cs.utk.edu/~plank/plank/www/libckpt.html). This is the
394 checkpointing solution used in the condor project, for example. It makes it
395 easy to create checkpoints (at the OS level, creating something like core
396 files), and rerunning them on need.
398 If you want to simulate checkpoints instead, it means that you want the
399 state of an executing task (in particular, the progress made towards
400 completion) to be saved somewhere. So if a host (and the task executing on
401 it) fails (cf. #MSG_HOST_FAILURE), then the task can be restarted
402 from the last checkpoint.\n
404 Actually, such a thing does not exists in SimGrid either, but it's just
405 because we don't think it is fundamental and it may be done in the user code
406 at relatively low cost. You could for example use a watcher that
407 periodically get the remaining amount of things to do (using
408 MSG_task_get_remaining_computation()), or fragment the task in smaller
411 \section faq_SG Where has SG disappeared?!?
413 OK, it's time to explain what's happening to the SimGrid project. Let's
414 start with a little bit of history.
416 * Historically, SimGrid was a low-level toolkit for scheduling with
417 classical models such as DAGs. That was SimGrid v.1.* aka SG, written
418 by Henri Casanova. I (Arnaud) had been using it in its earliest
419 versions during an internship at UCSD.
421 Then we have realized that encoding distributed algorithm in SG was a
424 * So we have built MSG on top of SG and have released SimGrid v.2.*. MSG
425 offered a very basic API to encode a distributed application easily.
426 However encoding MSG on top of SG was not really convenient and did not
427 use the DAG part since the control of the task synchronization was done
428 on top of MSG and no more in SG. We have been playing a little bit with
429 MSG. We have realized that:
431 \li 1) the platform modeling was quite flexible and could be "almost"
432 automated (e.g. using random generator and post-annotations);
434 \li 2) SG was the bottleneck because of the way we were using
435 it. We needed to simulate concurrent transfers, complex load
436 sharing mechanisms. Many optimizations (e.g. trace integration)
437 were totally inefficient when combined with MSG and made extending SG
438 to implement new sharing policies, parallel tasks models, or failures
439 (many people were asking for these kind of features) a real pain;
441 \li 3) the application modeling was not really easy. Even though the
442 application modeling depends on people's applications, we thought
443 we could improve things here. One of our target here was realistic
444 distributed applications ranging from computer sensor networks like
445 the NWS to peer-to-peer applications;
447 * So we have been planning mainly two things for SimGrid 3:
449 \li 1) I have proposed to get rid of SG and to re-implement a new kernel
450 that would be faster and more flexible. That is what I did in the
451 end of 2004: SURF. SURF is based on a fast max-min linear solver
452 using O(1) data-structures. I have quickly replaced SG by SURF in
453 MSG and the result has been that on the MSG example, the new
454 version was more than 10 times faster while we had gain a lot of
455 flexibility. I think I could still easily make MSG faster but I
456 have to work on MSG now (e.g. using some of the O(1)
457 data-structures I've been using to build SURF) since it has become
458 the bottleneck. Some MSG functions have been removed from the API
459 but they were mainly intended to build the platform by hand (they
460 had appeared in the earliest versions of MSG) and were therefore
461 not useful anymore since we are providing a complete mechanism to
462 automatically build the platform and deploy the agents on it.;
464 \li 2) GRAS is a new project Martin and I have come up with. The idea is
465 to have a programming environment that let you program real
466 distributed applications while letting you the ability to run it in
467 the simulator without having to change the slightest line of your
468 code. From the simulation point of view, GRAS performs the
469 application modeling automatically... Up until now, GRAS works on
470 top MSG for historical reasons but I'm going to make it work
471 directly on top of SURF so that it can use all the flex and the
472 speed provided by SURF.
474 Those two things are working, but we want to make everything as clean as
475 possible before releasing SimGrid v.3.
477 So what about those nice DAGs we used to have in SimGrid v.1.? They're not
478 anymore in SimGrid v.3. Let me recall you the way SimGrid 3 is organized:
492 XBT is our tool box and now, you should have an idea of what the other ones
493 are. As you can see, the primitive SG is not here anymore. However it could
494 still be brought back if people really need it. Here is how it would fit.
497 ______________________
499 |____________________|
500 | | MSG | GRAS | SG |
501 | -------------------|
503 | -------------------|
505 ----------------------
508 Re-implementing SG on top of SURF is really straightforward (it only
509 requires a little bit of time that I really don't have right now)
510 since the only thing that lacks to SURF is the DAG part. But adding it
511 to SURF would slow it down and therefore slow MSG and GRAS which is
512 not a good thing. However it is really not on the top of our TODO
513 list because we have to work on GRAS, and its MPI counterpart, and a
514 parallel task model, and ... Anyway, we finally have migrated our CVS
515 to gforge so people that are interested by helping on this part will
516 have the possibility to do it.
518 \subsection faq_SG_DAG How to implement a distributed dynamic scheduler of DAGs.
520 Distributed is somehow "contagious". If you start making distributed
521 decisions, there is no way to handle DAGs directly anymore (unless I am
522 missing something). You have to encode your DAGs in term of communicating
523 process to make the whole scheduling process distributed. Believe me, it is
524 worth the effort since you'll then be able to try your algorithms in a very
525 wide variety of conditions. Here is an example of how you could do that.
526 Assume T1 has to be done before T2.
529 int your_agent(int argc, char *argv[] {
531 T1 = MSG_task_create(...);
532 T2 = MSG_task_create(...);
536 if(cond) MSG_task_execute(T1);
538 if((MSG_task_get_remaining_computation(T1)=0.0) && (you_re_in_a_good_mood))
541 /* do something else */
547 If you decide that the distributed part is not that much important and that
548 DAG is really the level of abstraction you want to work with (but it
549 prevents you from having "realistic" platform modeling), then you should
550 keep using the 2.18.5 versions until somebody has ported SG on top of SURF.
551 Note however that SURF will be slower than the old SG to handle traces with
552 a lots of variations (there is no trace integration anymore).
554 \subsection faq_SG_future Will SG come back in the maintained branch one day?
556 Sure. In fact, we already have thought about a new and cleaner API:
558 void* SG_link_get_data(SG_link_t link);
559 void SG_link_set_data(SG_link_t link, void *data);
560 const char* SG_link_get_name(SG_link_t link);
561 double SG_link_get_capacity(SG_link_t link);
562 double SG_link_get_current_bandwidth(SG_link_t link);
563 double SG_link_get_current_latency(SG_link_t link);
565 SG_workstation_t SG_workstation_get_by_name(const char *name);
566 SG_workstation_t* SG_workstation_get_list(void);
567 int SG_workstation_get_number(void);
568 void SG_workstation_set_data(SG_workstation_t workstation, void *data);
569 void * SG_workstation_get_data(SG_workstation_t workstation);
570 const char* SG_workstation_get_name(SG_workstation_t workstation);
571 SG_link_t* SG_workstation_route_get_list(SG_workstation_t src, SG_workstation_t dst);
572 int SG_workstation_route_get_size(SG_workstation_t src, SG_workstation_t dst);
573 double SG_workstation_get_power(SG_workstation_t workstation);
574 double SG_workstation_get_available_power(SG_workstation_t workstation);
576 SG_task_t SG_task_create(const char *name, void *data, double amount);
577 int SG_task_schedule(SG_task_t task, int workstation_nb,
578 SG_workstation_t **workstation_list, double *computation_amount,
579 double *communication_amount, double rate);
581 void* SG_task_get_data(SG_task_t task);
582 void SG_task_set_data(SG_task_t task, void *data);
583 const char* SG_task_get_name(SG_task_t task);
584 double SG_task_get_amount(SG_task_t task);
585 double SG_task_get_remaining_amount(SG_task_t task);
586 void SG_task_dependency_add(const char *name, void *data, SG_task_t src, SG_task_t dst);
587 void SG_task_dependency_remove(SG_task_t src, SG_task_t dst);
588 e_SG_task_state_t SG_task_state_get(SG_task_t task); /* e_SG_task_state_t can be either SG_SCHEDULED, SG_RUNNING, SG_DONE, or SG_FAILED */
589 void SG_task_watch(SG_task_t task, e_SG_task_state_t state); /* SG_simulate will stop as soon as the state of this task is the one given in argument.
590 Watch-point is then automatically removed */
591 void SG_task_unwatch(SG_task_t task, e_SG_task_state_t state);
593 void SG_task_unschedule(SG_task_t task); /* change state and rerun.. */
595 SG_task_t *SG_simulate(double how_long); /* returns a NULL-terminated array of SG_task_t whose state has changed */
598 We're just looking for somebody to implement it... :)
600 \section faq_dynamic Dynamic resources and platform building
602 \subsection faq_platform Building a realistic platform
604 We can speak more than an hour on this subject and we still do not have
605 the right answer, just some ideas. You can read the following
606 <a href="http://graal.ens-lyon.fr/~alegrand/articles/Simgrid-Introduction.pdf">slides</a>.
607 It may give you some hints. You can also have a look at the
608 <tt>tools/platform_generation/</tt> directory. There is a perl-script
609 we use to annotate a Tiers generated platform.
611 \subsection faq_SURF_dynamic How can I have variable resource availability?
613 A nice feature of SimGrid is that it enables you to seamlessly have
614 resources whose availability change over time. When you build a
615 platform, you generally declare CPUs like that:
618 <cpu name="Cpu A" power="100.00"/>
621 If you want the availability of "CPU A" to change over time, the only
622 thing you have to do is change this definition like that:
625 <cpu name="Cpu A" power="100.00" availability_file="trace_A.txt" state_file="trace_A_failure.txt"/>
628 For CPUs, availability files are expressed in fraction of available
629 power. Let's have a look at what "trace_A.txt" may look like:
638 At time 0, our CPU will deliver 100 Mflop/s. At time 11.0, it will
639 deliver only 50 Mflop/s until time 20.0 where it will will start
640 delivering 90 Mflop/s. Last at time 21.0 (20.0 plus the periodicity
641 1.0), we'll be back to the beginning and it will deliver 100Mflop/s.
643 Now let's look at the state file:
650 A negative value means "off" while a positive one means "on". At time
651 1.0, the CPU is on. At time 1.0, it is turned off and at time 2.0, it
652 is turned on again until time 12 (2.0 plus the periodicity 10.0). It
653 will be turned on again at time 13.0 until time 23.0, and so on.
655 Now, let's look how the same kind of thing can be done for network
656 links. A usual declaration looks like:
659 <network_link name="LinkA" bandwidth="10.0" latency="0.2"/>
662 You have at your disposal the following options: bandwidth_file,
663 latency_file and state_file. The only difference with CPUs is that
664 bandwidth_file and latency_file do not express fraction of available
665 power but are expressed directly in Mb/s and seconds.
667 \subsection faq_flexml_bypassing How can I have some C functions do what the platform file does?
669 So you want to bypass the XML files parser, uh? Maybe doin some parameter
670 sweep experiments on your simulations or so? This is possible, but it's not
671 really easy. Here is how it goes.
673 For this, you have to first remember that the XML parsing in SimGrid is done
674 using a tool called FleXML. Given a DTD, this gives a flex-based parser. If
675 you want to bypass the parser, you need to provide some code mimicking what
676 it does and replacing it in its interactions with the SURF code. So, let's
677 have a look at these interactions.
679 FleXML parser are close to classical SAX parsers. It means that a
680 well-formed SimGrid platform XML file might result in the following
683 - start "platform_description"
684 - start "cpu" with attributes name="host1" power="1.0"
686 - start "cpu" with attributes name="host2" power="2.0"
688 - start "network_link" with ...
690 - start "route" with ...
692 - start "route" with ...
694 - end "platform_description"
696 The communication from the parser to the SURF code uses two means:
697 Attributes get copied into some global variables, and a surf-provided
698 function gets called by the parser for each event. For example, the event
699 - start "cpu" with attributes name="host1" power="1.0"
701 let the parser do the equivalent of:
703 strcpy("host1",A_cpu_name);
708 In SURF, we attach callbacks to the different events by initializing the
709 pointer functions to some the right surf functions. Example in
710 workstation_KCCFLN05.c (surf_parse_open() ends up calling surf_parse()):
712 // Building the routes
713 surf_parse_reset_parser();
714 STag_route_fun=parse_route_set_endpoints;
715 ETag_route_element_fun=parse_route_elem;
716 ETag_route_fun=parse_route_set_route;
717 surf_parse_open(file);
718 xbt_assert1((!surf_parse()),"Parse error in %s",file);
722 So, to bypass the FleXML parser, you need to write your own version of the
723 surf_parse function, which should do the following:
724 - Call the corresponding STag_<tag>_fun function to simulate tag start
725 - Fill the A_<tag>_<attribute> variables with the wanted values
726 - Call the corresponding ETag_<tag>_fun function to simulate tag end
727 - (do the same for the next set of values, and loop)
729 Then, tell SimGrid that you want to use your own "parser" instead of the stock one:
731 surf_parse = surf_parse_bypass;
732 MSG_create_environment(NULL);
735 An example of this trick is distributed in the file examples/msg/msg_test_surfxml_bypassed.c
737 \section faq_troubleshooting Troubleshooting
739 \subsection faq_compil_trouble ./configure fails!
741 We now only one reason for the configure to fail:
743 - <b>You are using a borken build environment</b>\n
744 If symptom is that configure complains about gcc not being able to build
745 executables, you are probably missing the libc6-dev package. Damn Ubuntu.
747 If you experience other kind of issue, please get in touch with us. We are
748 always interested in improving our portability to new systems.
750 \subsection faq_distcheck_fails Dude! "make check" fails on my machine!
752 Don't assume we never run this target, because we do. Really. Promise!
754 There is several reasons which may cause the make check to fail on your
757 - <b>You are using a borken compiler</b>.\n
758 The symptom may be that the "make check" fails within testsuite/gras
760 For example, the breezy release of Ubuntu comes with a prerelease of the
761 4.0 gcc compiler. This version happens to be completely unusable, and you
762 should install a gcc-3.4 compiler and change the /usr/bin/gcc link to let
763 it point on /usr/bin/gcc-3.4.
764 - <b>You are using a borken libc (probably concerning the contextes)</b>.\n
765 The symptom is that the "make check" fails within the examples/msg directory.\n
766 By default, SimGrid uses something called ucontexts. This is part of the
767 libc, but it's quite undertested. For example, some (old) versions of the
768 glibc on alpha do not implement these functions, but provide the stubs
769 (which return ENOSYS: not implemented). It fools our detection mecanism
770 and leads to segfaults.\n
771 On some x86_64, the pointer to function is stored into a integer, but int
772 are 32bits only on this arch while pointers are 64bits. Our detection
773 mecanism also fails to detect the problem, which leads to segfaults.\n
774 In both cases, there is not much we can do to fix the bug. We are working
775 on a workaround for x86_64 machines, but in the meanwhile, you can
776 compile with --with-context=pthread to avoid ucontext completely. You'll
777 be a bit more limitated in the number of simulated processes you can start
778 concurently, but 5000 processes is still enough for most purposes, isn't
780 This limitation is the reason why we insist on using this piece of ...
781 software even if it's so troublesome.
782 - <b>There is a bug in SimGrid we aren't aware of</b>.\n
783 If none of the above apply, please drop us a mail on the mailing list so
784 that we can check it out.
786 \subsection faq_context_1000 I want thousands of simulated processes
788 SimGrid can use either pthreads library or the UNIX98 contextes. On most
789 systems, the number of pthreads is limited and then your simulation may be
790 limited for a stupid reason. This is especially true with the current linux
791 pthreads, and I cannot get more than 2000 simulated processes with pthreads
792 on my box. The UNIX98 contexts allow me to raise the limit to 25,000
793 simulated processes on my laptop.
795 The <tt>--with-context</tt> option of the <tt>./configure</tt> script allows
796 you to choose between UNIX98 contextes (<tt>--with-context=ucontext</tt>)
797 and the pthread version ( (<tt>--with-context=pthread</tt>). The default
798 value is ucontext when the script detect a working UNIX98 context
799 implementation. On Windows boxes, the provided value is discarded and an
800 adapted version is picked up.
802 We experienced some issues with contextes on some rare systems (solaris 8
803 and lower or old alpha linuxes comes to mind). The main problem is that the
804 configure script detect the contextes as being functional when it's not
805 true. If you happen to use such a system, switch manually to the pthread
806 version, and provide us with a good patch for the configure script so that
807 it is done automatically ;)
809 \subsection faq_context_10000 I want hundred thousands of simulated processes
811 As explained above, SimGrid can use UNIX98 contextes to represent and handle
812 the simulated processes. Thanks to this, the main limitation to the number
813 of simulated processes becomes the available memory.
815 Here are some tricks I had to use in order to run a token ring between
816 25,000 processes on my laptop (1Gb memory, 1.5Gb swap).
818 - First of all, make sure your code runs for a few hundreds processes
819 before trying to push the limit. Make sure it's valgrind-clean, ie that
820 valgrind does not report neither memory error nor memory leaks. Indeed,
821 numerous simulated processes result in *fat* simulation hindering debugging.
823 - It was really boring to write 25,000 entries in the deployment file, so I wrote
824 a little script <tt>examples/gras/tokenS/make_deployment.pl</tt>, which you may
825 want to adapt to your case.
827 - The deployment file became quite big, so I had to do what is in the FAQ
828 entry \ref faq_flexml_limit
830 - Each UNIX98 context has its own stack entry. As debugging this is quite
831 hairly, the default value is a bit overestimated so that user don't get
832 into trouble about this. You want to tune this size to increse the number
833 of processes. This is the <tt>STACK_SIZE</tt> define in
834 <tt>src/xbt/context_private.h</tt>, which is 128kb by default.
835 Reduce this as much as you can, but be warned that if this value is too
836 low, you'll get a segfault. The token ring example, which is quite simple,
837 runs with 40kb stacks.
839 \subsection faq_longjmp longjmp madness
841 This is when valgrind starts complaining about longjmp things, just like:
843 \verbatim ==21434== Conditional jump or move depends on uninitialised value(s)
844 ==21434== at 0x420DBE5: longjmp (longjmp.c:33)
846 ==21434== Use of uninitialised value of size 4
847 ==21434== at 0x420DC3A: __longjmp (__longjmp.S:48)
850 or even when it reports scary things like:
852 \verbatim ==24023== Warning: client switching stacks? SP change: 0xBE3FF618 --> 0xBE7FF710
853 x86->IR: unhandled instruction bytes: 0xF4 0xC7 0x83 0xD0
854 ==24023== to suppress, use: --max-stackframe=4194552 or greater
855 ==24023== Your program just tried to execute an instruction that Valgrind
856 ==24023== did not recognise. There are two possible reasons for this.
857 ==24023== 1. Your program has a bug and erroneously jumped to a non-code
858 ==24023== location. If you are running Memcheck and you just saw a
859 ==24023== warning about a bad jump, it's probably your program's fault.
860 ==24023== 2. The instruction is legitimate but Valgrind doesn't handle it,
861 ==24023== i.e. it's Valgrind's fault. If you think this is the case or
862 ==24023== you are not sure, please let us know.
863 ==24023== Either way, Valgrind will now raise a SIGILL signal which will
864 ==24023== probably kill your program.
866 ==24023== Process terminating with default action of signal 4 (SIGILL)
867 ==24023== Illegal opcode at address 0x420D234
868 ==24023== at 0x420D234: abort (abort.c:124)
871 This is the sign that you didn't used the exception mecanism well. Most
872 probably, you have a <tt>return;</tt> somewhere within a <tt>TRY{}</tt>
873 block. This is <b>evil</b>, and you must not do this. Did you read the section
876 \subsection faq_flexml_limit I get the message "surf_parse_lex: Assertion `next<limit' failed."
878 This is because your platform file is too big for the parser.
880 Actually, the message comes directly from FleXML, the technology on top of
881 which the parser is built. FleXML has the bad idea of fetching the whole
882 document in memory before parsing it. And moreover, the memory buffer size
883 must be determinded at compilation time.
885 We use a value which seems big enough for our need without bloating the
886 simulators footprints. But of course your mileage may vary. In this case,
887 just edit src/surf/surfxml.l modify the definition of
888 FLEXML_BUFFERSTACKSIZE. E.g.
891 #define FLEXML_BUFFERSTACKSIZE 1000000000
894 Then recompile and everything should be fine, provided that your version of
895 Flex is recent enough (>= 2.5.31). If not the compilation process should
898 A while ago, we worked on FleXML to reduce a bit its memory consumtion, but
899 these issues remain. There is two things we should do:
901 - use a dynamic buffer instead of a static one so that the only limit
902 becomes your memory, not a stupid constant fixed at compilation time
903 (maybe not so difficult).
904 - change the parser so that it does not need to get the whole file in
905 memory before parsing
906 (seems quite difficult, but I'm a complete newbe wrt flex stuff).
908 These are changes to FleXML itself, not SimGrid. But since we kinda hijacked
909 the development of FleXML, I can grant you that any patches would be really
910 welcome and quickly integrated.
912 \subsection faq_gras_transport GRAS spits networking error messages
914 Gras, on real platforms, naturally use regular sockets to communicate. They
915 are deeply hiden in the gras abstraction, but when things go wrong, you may
916 get some weird error messages. Here are some example, with the probable
919 - <b>Transport endpoint is not connected</b>: several processes try to open
920 a server socket on the same port number of the same machine. This is
921 naturally bad and each process should pick its own port number for this.\n
922 Maybe, you just have some processes remaining from a previous experiment
924 Killing them may help, but again if you kill -KILL them, you'll have to
925 wait for a while: they didn't close there sockets properly and the system
926 needs a while to notice that this port is free again.
928 - <b>Socket closed by remote side</b>: if the remote process is not
929 supposed to close the socket at this point, it may be dead.
931 - <b>Connection reset by peer</b>: I found this on internet about this
932 error. I think it's what's happening here, too:\n
933 <i>This basically means that a network error occurred while the client was
934 receiving data from the server. But what is really happening is that the
935 server actually accepts the connection, processes the request, and sends
936 a reply to the client. However, when the server closes the socket, the
937 client believes that the connection has been terminated abnormally
938 because the socket implementation sends a TCP reset segment telling the
939 client to throw away the data and report an error.\n
940 Sometimes, this problem is caused by not properly closing the
941 input/output streams and the socket connection. Make sure you close the
942 input/output streams and socket connection properly. If everything is
943 closed properly, however, and the problem persists, you can work around
944 it by adding a one-second sleep before closing the streams and the
945 socket. This technique, however, is not reliable and may not work on all
947 Since GRAS sockets are closed properly (repeat after me: there is no bug
948 in GRAS), it is either that you are closing your sockets on server side
949 before the client get a chance to read them (use gras_os_sleep() to delay
950 the server), or the server died awfully before the client got the data.
953 \subsection faq_deadlock There is a deadlock !!!
955 Unfortunately, we cannot debug every code written in SimGrid. We
956 furthermore believe that the framework provides ways enough
957 information to debug such informations yourself. If the textual output
958 is not enough, Make sure to check the \ref faq_visualization FAQ entry to see
959 how to get a graphical one.
961 Now, if you come up with a really simple example that deadlocks and
962 you're absolutely convinced that it should not, you can ask on the
963 list. Just be aware that you'll be severely punished if the mistake is
964 on your side... We have plenty of FAQ entries to redact and new
965 features to implement for the impenitents! ;)
967 \author Arnaud Legrand (arnaud.legrand::imag.fr)
968 \author Martin Quinson (martin.quinson::loria.fr)