1 /*! \page options Configure SimGrid
3 A number of options can be given at runtime to change the default
4 SimGrid behavior. For a complete list of all configuration options
5 accepted by the SimGrid version used in your simulator, simply pass
6 the --help configuration flag to your program. If some of the options
7 are not documented on this page, this is a bug that you should please
8 report so that we can fix it. Note that some of the options presented
9 here may not be available in your simulators, depending on the
10 @ref install_src_config "compile-time options" that you used.
14 \section options_using Passing configuration options to the simulators
16 There is several way to pass configuration options to the simulators.
17 The most common way is to use the \c --cfg command line argument. For
18 example, to set the item \c Item to the value \c Value, simply
19 type the following: \verbatim
20 my_simulator --cfg=Item:Value (other arguments)
23 Several \c `--cfg` command line arguments can naturally be used. If you
24 need to include spaces in the argument, don't forget to quote the
25 argument. You can even escape the included quotes (write \' for ' if
26 you have your argument between ').
28 Another solution is to use the \c \<config\> tag in the platform file. The
29 only restriction is that this tag must occure before the first
30 platform element (be it \c \<AS\>, \c \<cluster\>, \c \<peer\> or whatever).
31 The \c \<config\> tag takes an \c id attribute, but it is currently
32 ignored so you don't really need to pass it. The important par is that
33 within that tag, you can pass one or several \c \<prop\> tags to specify
34 the configuration to use. For example, setting \c Item to \c Value
35 can be done by adding the following to the beginning of your platform
39 <prop id="Item" value="Value"/>
43 A last solution is to pass your configuration directly using the C
44 interface. If you happen to use the MSG interface, this is very easy
45 with the MSG_config() function. If you do not use MSG, that's a bit
46 more complex, as you have to mess with the internal configuration set
47 directly as follows. Check the \ref XBT_config "relevant page" for
48 details on all the functions you can use in this context, \c
49 _sg_cfg_set being the only configuration set currently used in
53 #include <xbt/config.h>
55 extern xbt_cfg_t _sg_cfg_set;
57 int main(int argc, char *argv[]) {
60 /* Prefer MSG_config() if you use MSG!! */
61 xbt_cfg_set_parse(_sg_cfg_set,"Item:Value");
67 \section options_model Configuring the platform models
69 \anchor options_storage_model
70 \anchor options_vm_model
71 \subsection options_model_select Selecting the platform models
73 SimGrid comes with several network, CPU and storage models built in, and you
74 can change the used model at runtime by changing the passed
75 configuration. The three main configuration items are given below.
76 For each of these items, passing the special \c help value gives
77 you a short description of all possible values. Also, \c --help-models
78 should provide information about all models for all existing resources.
79 - \b network/model: specify the used network model
80 - \b cpu/model: specify the used CPU model
81 - \b host/model: specify the used host model
82 - \b storage/model: specify the used storage model (there is currently only one such model - this option is hence only useful for future releases)
83 - \b vm/model: specify the model for virtual machines (there is currently only one such model - this option is hence only useful for future releases)
85 %As of writing, the following network models are accepted. Over
86 the time new models can be added, and some experimental models can be
87 removed; check the values on your simulators for an uptodate
88 information. Note that the CM02 model is described in the research report
89 <a href="ftp://ftp.ens-lyon.fr/pub/LIP/Rapports/RR/RR2002/RR2002-40.ps.gz">A
90 Network Model for Simulation of Grid Application</a> while LV08 is
92 <a href="http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf">Accuracy Study and Improvement of Network Simulation in the SimGrid Framework</a>.
94 - \b LV08 (default one): Realistic network analytic model
95 (slow-start modeled by multiplying latency by 10.4, bandwidth by
96 .92; bottleneck sharing uses a payload of S=8775 for evaluating RTT)
97 - \anchor options_model_select_network_constant \b Constant: Simplistic network model where all communication
98 take a constant time (one second). This model provides the lowest
99 realism, but is (marginally) faster.
100 - \b SMPI: Realistic network model specifically tailored for HPC
101 settings (accurate modeling of slow start with correction factors on
102 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). See also \ref
103 options_model_network_coefs "this section" for more info.
104 - \b IB: Realistic network model specifically tailored for HPC
105 settings with InfiniBand networks (accurate modeling contention
106 behavior, based on the model explained in
107 http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf).
108 See also \ref options_model_network_coefs "this section" for more info.
109 - \b CM02: Legacy network analytic model (Very similar to LV08, but
110 without corrective factors. The timings of small messages are thus
112 - \b Reno: Model from Steven H. Low using lagrange_solve instead of
113 lmm_solve (experts only; check the code for more info).
114 - \b Reno2: Model from Steven H. Low using lagrange_solve instead of
115 lmm_solve (experts only; check the code for more info).
116 - \b Vegas: Model from Steven H. Low using lagrange_solve instead of
117 lmm_solve (experts only; check the code for more info).
119 If you compiled SimGrid accordingly, you can use packet-level network
120 simulators as network models (see \ref pls_ns3). In that case, you have
121 two extra models, described below, and some \ref options_pls "specific
122 additional configuration flags".
123 - \b NS3: Network pseudo-model using the NS3 tcp model
125 Concerning the CPU, we have only one model for now:
126 - \b Cas01: Simplistic CPU model (time=size/power)
128 The host concept is the aggregation of a CPU with a network
129 card. Three models exists, but actually, only 2 of them are
130 interesting. The "compound" one is simply due to the way our internal
131 code is organized, and can easily be ignored. So at the end, you have
132 two host models: The default one allows to aggregate an
133 existing CPU model with an existing network model, but does not allow
134 parallel tasks because these beasts need some collaboration between
135 the network and CPU model. That is why, ptask_07 is used by default
137 - \b default: Default host model. Currently, CPU:Cas01 and
138 network:LV08 (with cross traffic enabled)
139 - \b compound: Host model that is automatically chosen if
140 you change the network and CPU models
141 - \b ptask_L07: Host model somehow similar to Cas01+CM02 but
142 allowing "parallel tasks", that are intended to model the moldable
143 tasks of the grid scheduling literature.
145 \subsection options_generic_plugin Plugins
147 SimGrid supports the use of plugins; currently, no known plugins
148 can be activated but there are use-cases where you may want to write
149 your own plugin (for instance, for logging).
151 Plugins can for instance define own classes that inherit from
152 existing classes (for instance, a class "CpuEnergy" inherits from
153 "Cpu" to assess energy consumption).
155 The plugin connects to the code by registering callbacks using
156 ``signal.connect(callback)`` (see file ``src/surf/plugins/energy.cpp`` for
164 This option is case-sensitive: Energy and energy are not the same!
166 \subsection options_model_optim Optimization level of the platform models
168 The network and CPU models that are based on lmm_solve (that
169 is, all our analytical models) accept specific optimization
171 - items \b network/optim and \b CPU/optim (both default to 'Lazy'):
172 - \b Lazy: Lazy action management (partial invalidation in lmm +
173 heap in action remaining).
174 - \b TI: Trace integration. Highly optimized mode when using
175 availability traces (only available for the Cas01 CPU model for
177 - \b Full: Full update of remaining and variables. Slow but may be
178 useful when debugging.
179 - items \b network/maxmin-selective-update and
180 \b cpu/maxmin-selective-update: configure whether the underlying
181 should be lazily updated or not. It should have no impact on the
182 computed timings, but should speed up the computation.
184 It is still possible to disable the \c maxmin-selective-update feature
185 because it can reveal counter-productive in very specific scenarios
186 where the interaction level is high. In particular, if all your
187 communication share a given backbone link, you should disable it:
188 without \c maxmin-selective-update, every communications are updated
189 at each step through a simple loop over them. With that feature
190 enabled, every communications will still get updated in this case
191 (because of the dependency induced by the backbone), but through a
192 complicated pattern aiming at following the actual dependencies.
194 \subsection options_model_precision Numerical precision of the platform models
196 The analytical models handle a lot of floating point values. It is
197 possible to change the epsilon used to update and compare them through
198 the \b maxmin/precision item (default value: 0.00001). Changing it
199 may speedup the simulation by discarding very small actions, at the
200 price of a reduced numerical precision.
202 \subsection options_model_network Configuring the Network model
204 \subsubsection options_model_network_gamma Maximal TCP window size
206 The analytical models need to know the maximal TCP window size to take
207 the TCP congestion mechanism into account. This is set to 20000 by
208 default, but can be changed using the \b network/TCP-gamma item.
210 On linux, this value can be retrieved using the following
211 commands. Both give a set of values, and you should use the last one,
212 which is the maximal size.\verbatim
213 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
214 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
217 \subsubsection options_model_network_coefs Correcting important network parameters
219 SimGrid can take network irregularities such as a slow startup or
220 changing behavior depending on the message size into account.
221 You should not change these values unless you really know what you're doing.
223 The corresponding values were computed through data fitting one the
224 timings of packet-level simulators.
227 <a href="http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf">Accuracy Study and Improvement of Network Simulation in the SimGrid Framework</a>
228 for more information about these parameters.
230 If you are using the SMPI model, these correction coefficients are
231 themselves corrected by constant values depending on the size of the
232 exchange. Again, only hardcore experts should bother about this fact.
234 InfiniBand network behavior can be modeled through 3 parameters, as explained in
235 <a href="http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf">this PhD thesis</a>.
236 These factors can be changed through the following option:
239 smpi/IB-penalty-factors:"βe;βs;γs"
242 By default SMPI uses factors computed on the Stampede Supercomputer at TACC, with optimal
243 deployment of processes on nodes.
245 \subsubsection options_model_network_crosstraffic Simulating cross-traffic
247 %As of SimGrid v3.7, cross-traffic effects can be taken into account in
248 analytical simulations. It means that ongoing and incoming
249 communication flows are treated independently. In addition, the LV08
250 model adds 0.05 of usage on the opposite direction for each new
251 created flow. This can be useful to simulate some important TCP
252 phenomena such as ack compression.
254 For that to work, your platform must have two links for each
255 pair of interconnected hosts. An example of usable platform is
256 available in <tt>examples/platforms/crosstraffic.xml</tt>.
258 This is activated through the \b network/crosstraffic item, that
259 can be set to 0 (disable this feature) or 1 (enable it).
261 Note that with the default host model this option is activated by default.
263 \subsubsection options_model_network_coord Coordinated-based network models
265 When you want to use network coordinates, as it happens when you use
266 an \<AS\> in your platform file with \c Vivaldi as a routing (see also
267 Section \ref pf_routing_model_vivaldi "Vivaldi Routing Model"), you must
268 set the \b network/coordinates to \c yes so that all mandatory
269 initialization are done in the simulator.
271 \subsubsection options_model_network_sendergap Simulating sender gap
273 (this configuration item is experimental and may change or disapear)
275 It is possible to specify a timing gap between consecutive emission on
276 the same network card through the \b network/sender-gap item. This
277 is still under investigation as of writting, and the default value is
278 to wait 10 microseconds (1e-5 seconds) between emissions.
280 \subsubsection options_model_network_asyncsend Simulating asyncronous send
282 (this configuration item is experimental and may change or disapear)
284 It is possible to specify that messages below a certain size will be sent
285 as soon as the call to MPI_Send is issued, without waiting for the
286 correspondant receive. This threshold can be configured through the
287 \b smpi/async-small-thresh item. The default value is 0. This behavior can also be
288 manually set for MSG mailboxes, by setting the receiving mode of the mailbox
289 with a call to \ref MSG_mailbox_set_async . For MSG, all messages sent to this
290 mailbox will have this behavior, so consider using two mailboxes if needed.
292 This value needs to be smaller than or equals to the threshold set at
293 \ref options_model_smpi_detached , because asynchronous messages are
294 meant to be detached as well.
296 \subsubsection options_pls Configuring packet-level pseudo-models
298 When using the packet-level pseudo-models, several specific
299 configuration flags are provided to configure the associated tools.
300 There is by far not enough such SimGrid flags to cover every aspects
301 of the associated tools, since we only added the items that we
302 needed ourselves. Feel free to request more items (or even better:
303 provide patches adding more items).
305 When using NS3, the only existing item is \b ns3/TcpModel,
306 corresponding to the ns3::TcpL4Protocol::SocketType configuration item
307 in NS3. The only valid values (enforced on the SimGrid side) are
308 'NewReno' or 'Reno' or 'Tahoe'.
310 \section options_modelchecking Configuring the Model-Checking
312 To enable the SimGrid model-checking support the program should
313 be executed using the simgrid-mc wrapper:
315 simgrid-mc ./my_program
318 Safety properties are expressed as assertions using the function
320 void MC_assert(int prop);
323 \subsection options_modelchecking_liveness Specifying a liveness property
325 If you want to specify liveness properties (beware, that's
326 experimental), you have to pass them on the command line, specifying
327 the name of the file containing the property, as formatted by the
331 --cfg=model-check/property:<filename>
334 \subsection options_modelchecking_steps Going for stateful verification
336 By default, the system is backtracked to its initial state to explore
337 another path instead of backtracking to the exact step before the fork
338 that we want to explore (this is called stateless verification). This
339 is done this way because saving intermediate states can rapidly
340 exhaust the available memory. If you want, you can change the value of
341 the <tt>model-check/checkpoint</tt> variable. For example, the
342 following configuration will ask to take a checkpoint every step.
343 Beware, this will certainly explode your memory. Larger values are
344 probably better, make sure to experiment a bit to find the right
345 setting for your specific system.
348 --cfg=model-check/checkpoint:1
351 \subsection options_modelchecking_reduction Specifying the kind of reduction
353 The main issue when using the model-checking is the state space
354 explosion. To counter that problem, several exploration reduction
355 techniques can be used. There is unfortunately no silver bullet here,
356 and the most efficient reduction techniques cannot be applied to any
357 properties. In particular, the DPOR method cannot be applied on
358 liveness properties since it may break some cycles in the exploration
359 that are important to the property validity.
362 --cfg=model-check/reduction:<technique>
365 For now, this configuration variable can take 2 values:
366 * none: Do not apply any kind of reduction (mandatory for now for
368 * dpor: Apply Dynamic Partial Ordering Reduction. Only valid if you
369 verify local safety properties.
371 \subsection options_modelchecking_visited model-check/visited, Cycle detection
373 In order to detect cycles, the model-checker needs to check if a new explored
374 state is in fact the same state than a previous one. In order to do this,
375 the model-checker can take a snapshot of each visited state: this snapshot is
376 then used to compare it with subsequent states in the exploration graph.
378 The \b model-check/visited is the maximum number of states which are stored in
379 memory. If the maximum number of snapshotted state is reached some states will
380 be removed from the memory and some cycles might be missed.
382 By default, no state is snapshotted and cycles cannot be detected.
384 \subsection options_modelchecking_termination model-check/termination, Non termination detection
386 The \b model-check/termination configuration item can be used to report if a
387 non-termination execution path has been found. This is a path with a cycle
388 which means that the program might never terminate.
390 This only works in safety mode.
392 This options is disabled by default.
394 \subsection options_modelchecking_dot_output model-check/dot-output, Dot output
396 If set, the \b model-check/dot-output configuration item is the name of a file
397 in which to write a dot file of the path leading the found property (safety or
398 liveness violation) as well as the cycle for liveness properties. This dot file
399 can then fed to the graphviz dot tool to generate an corresponding graphical
402 \subsection options_modelchecking_max_depth model-check/max_depth, Depth limit
404 The \b model-checker/max-depth can set the maximum depth of the exploration
405 graph of the model-checker. If this limit is reached, a logging message is
406 sent and the results might not be exact.
408 By default, there is not depth limit.
410 \subsection options_modelchecking_timeout Handling of timeout
412 By default, the model-checker does not handle timeout conditions: the `wait`
413 operations never time out. With the \b model-check/timeout configuration item
414 set to \b yes, the model-checker will explore timeouts of `wait` operations.
416 \subsection options_modelchecking_comm_determinism Communication determinism
418 The \b model-check/communications-determinism and
419 \b model-check/send-determinism items can be used to select the communication
420 determinism mode of the model-checker which checks determinism properties of
421 the communications of an application.
423 \subsection options_modelchecking_sparse_checkpoint Per page checkpoints
425 When the model-checker is configured to take a snapshot of each explored state
426 (with the \b model-checker/visited item), the memory consumption can rapidly
427 reach GiB ou Tib of memory. However, for many workloads, the memory does not
428 change much between different snapshots and taking a complete copy of each
429 snapshot is a waste of memory.
431 The \b model-check/sparse-checkpoint option item can be set to \b yes in order
432 to avoid making a complete copy of each snapshot: instead, each snapshot will be
433 decomposed in blocks which will be stored separately.
434 If multiple snapshots share the same block (or if the same block
435 is used in the same snapshot), the same copy of the block will be shared leading
436 to a reduction of the memory footprint.
438 For many applications, this option considerably reduces the memory consumption.
439 In somes cases, the model-checker might be slightly slower because of the time
440 taken to manage the metadata about the blocks. In other cases however, this
441 snapshotting strategy will be much faster by reducing the cache consumption.
442 When the memory consumption is important, by avoiding to hit the swap or
443 reducing the swap usage, this option might be much faster than the basic
444 snapshotting strategy.
446 This option is currently disabled by default.
448 \subsection options_mc_perf Performance considerations for the model checker
450 The size of the stacks can have a huge impact on the memory
451 consumption when using model-checking. By default, each snapshot will
452 save a copy of the whole stacks and not only of the part which is
453 really meaningful: you should expect the contribution of the memory
454 consumption of the snapshots to be \f$ \mbox{number of processes}
455 \times \mbox{stack size} \times \mbox{number of states} \f$.
457 The \b model-check/sparse-checkpoint can be used to reduce the memory
458 consumption by trying to share memory between the different snapshots.
460 When compiled against the model checker, the stacks are not
461 protected with guards: if the stack size is too small for your
462 application, the stack will silently overflow on other parts of the
465 \subsection options_modelchecking_hash Hashing of the state (experimental)
467 Usually most of the time of the model-checker is spent comparing states. This
468 process is complicated and consumes a lot of bandwidth and cache.
469 In order to speedup the state comparison, the experimental \b model-checker/hash
470 configuration item enables the computation of a hash summarizing as much
471 information of the state as possible into a single value. This hash can be used
472 to avoid most of the comparisons: the costly comparison is then only used when
473 the hashes are identical.
475 Currently most of the state is not included in the hash because the
476 implementation was found to be buggy and this options is not as useful as
477 it could be. For this reason, it is currently disabled by default.
479 \subsection options_modelchecking_recordreplay Record/replay (experimental)
481 As the model-checker keeps jumping at different places in the execution graph,
482 it is difficult to understand what happens when trying to debug an application
483 under the model-checker. Event the output of the program is difficult to
484 interpret. Moreover, the model-checker does not behave nicely with advanced
485 debugging tools such as valgrind. For those reason, to identify a trajectory
486 in the execution graph with the model-checker and replay this trajcetory and
487 without the model-checker black-magic but with more standard tools
488 (such as a debugger, valgrind, etc.). For this reason, Simgrid implements an
489 experimental record/replay functionnality in order to record a trajectory with
490 the model-checker and replay it without the model-checker.
492 When the model-checker finds an interesting path in the application execution
493 graph (where a safety or liveness property is violated), it can generate an
494 identifier for this path. In order to enable this behavious the
495 \b model-check/record must be set to \b yes. By default, this behaviour is not
498 This is an example of output:
501 [ 0.000000] (0:@) Check a safety property
502 [ 0.000000] (0:@) **************************
503 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
504 [ 0.000000] (0:@) **************************
505 [ 0.000000] (0:@) Counter-example execution trace:
506 [ 0.000000] (0:@) Path = 1/3;1/4
507 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
508 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
509 [ 0.000000] (0:@) Expanded states = 27
510 [ 0.000000] (0:@) Visited states = 68
511 [ 0.000000] (0:@) Executed transitions = 46
514 This path can then be replayed outside of the model-checker (and even in
515 non-MC build of simgrid) by setting the \b model-check/replay item to the given
516 path. The other options should be the same (but the model-checker should
519 The format and meaning of the path may change between different releases so
520 the same release of Simgrid should be used for the record phase and the replay
523 \section options_virt Configuring the User Process Virtualization
525 \subsection options_virt_factory Selecting the virtualization factory
527 In SimGrid, the user code is virtualized in a specific mechanism
528 that allows the simulation kernel to control its execution: when a user
529 process requires a blocking action (such as sending a message), it is
530 interrupted, and only gets released when the simulated clock reaches
531 the point where the blocking operation is done.
533 In SimGrid, the containers in which user processes are virtualized are
534 called contexts. Several context factory are provided, and you can
535 select the one you want to use with the \b contexts/factory
536 configuration item. Some of the following may not exist on your
537 machine because of portability issues. In any case, the default one
538 should be the most effcient one (please report bugs if the
539 auto-detection fails for you). They are sorted here from the slowest
541 - \b thread: very slow factory using full featured threads (either
542 pthreads or windows native threads)
543 - \b ucontext: fast factory using System V contexts (or a portability
544 layer of our own on top of Windows fibers)
545 - \b raw: amazingly fast factory using a context switching mechanism
546 of our own, directly implemented in assembly (only available for x86
547 and amd64 platforms for now)
548 - \b boost: This uses the [context implementation](http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html)
549 of the boost library; you must have this library installed before
550 you compile SimGrid. (On Debian GNU/Linux based systems, this is
551 provided by the libboost-contexts-dev package.)
553 The only reason to change this setting is when the debugging tools get
554 fooled by the optimized context factories. Threads are the most
555 debugging-friendly contextes, as they allow to set breakpoints anywhere with gdb
556 and visualize backtraces for all processes, in order to debug concurrency issues.
557 Valgrind is also more comfortable with threads, but it should be usable with all factories.
559 \subsection options_virt_stacksize Adapting the used stack size
561 Each virtualized used process is executed using a specific system
562 stack. The size of this stack has a huge impact on the simulation
563 scalability, but its default value is rather large. This is because
564 the error messages that you get when the stack size is too small are
565 rather disturbing: this leads to stack overflow (overwriting other
566 stacks), leading to segfaults with corrupted stack traces.
568 If you want to push the scalability limits of your code, you might
569 want to reduce the \b contexts/stack-size item. Its default value
570 is 8192 (in KiB), while our Chord simulation works with stacks as small
571 as 16 KiB, for example. For the thread factory, the default value
572 is the one of the system, if it is too large/small, it has to be set
575 The operating system should only allocate memory for the pages of the
576 stack which are actually used and you might not need to use this in
577 most cases. However, this setting is very important when using the
578 model checker (see \ref options_mc_perf).
580 In some cases, no stack guard page is used and the stack will silently
581 overflow on other parts of the memory if the stack size is too small
582 for your application. This happens :
584 - on Windows systems;
585 - when the model checker is enabled;
586 - when stack guard pages are explicitely disabled (see \ref options_perf_guard_size).
588 \subsection options_virt_parallel Running user code in parallel
590 Parallel execution of the user code is only considered stable in
591 SimGrid v3.7 and higher. It is described in
592 <a href="http://hal.inria.fr/inria-00602216/">INRIA RR-7653</a>.
594 If you are using the \c ucontext or \c raw context factories, you can
595 request to execute the user code in parallel. Several threads are
596 launched, each of them handling as much user contexts at each run. To
597 actiave this, set the \b contexts/nthreads item to the amount of
598 cores that you have in your computer (or lower than 1 to have
599 the amount of cores auto-detected).
601 Even if you asked several worker threads using the previous option,
602 you can request to start the parallel execution (and pay the
603 associated synchronization costs) only if the potential parallelism is
604 large enough. For that, set the \b contexts/parallel-threshold
605 item to the minimal amount of user contexts needed to start the
606 parallel execution. In any given simulation round, if that amount is
607 not reached, the contexts will be run sequentially directly by the
608 main thread (thus saving the synchronization costs). Note that this
609 option is mainly useful when the grain of the user code is very fine,
610 because our synchronization is now very efficient.
612 When parallel execution is activated, you can choose the
613 synchronization schema used with the \b contexts/synchro item,
614 which value is either:
615 - \b futex: ultra optimized synchronisation schema, based on futexes
616 (fast user-mode mutexes), and thus only available on Linux systems.
617 This is the default mode when available.
618 - \b posix: slow but portable synchronisation using only POSIX
620 - \b busy_wait: not really a synchronisation: the worker threads
621 constantly request new contexts to execute. It should be the most
622 efficient synchronisation schema, but it loads all the cores of your
623 machine for no good reason. You probably prefer the other less
626 \section options_tracing Configuring the tracing subsystem
628 The \ref outcomes_vizu "tracing subsystem" can be configured in several
629 different ways depending on the nature of the simulator (MSG, SimDag,
630 SMPI) and the kind of traces that need to be obtained. See the \ref
631 tracing_tracing_options "Tracing Configuration Options subsection" to
632 get a detailed description of each configuration option.
634 We detail here a simple way to get the traces working for you, even if
635 you never used the tracing API.
638 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
640 --cfg=tracing:yes --cfg=tracing/uncategorized:yes --cfg=triva/uncategorized:uncat.plist
642 The first parameter activates the tracing subsystem, the second
643 tells it to trace host and link utilization (without any
644 categorization) and the third creates a graph configuration file
645 to configure Triva when analysing the resulting trace file.
647 - MSG or SimDag-based simulator and categorized traces (you need to declare categories and classify your tasks according to them)
649 --cfg=tracing:yes --cfg=tracing/categorized:yes --cfg=triva/categorized:cat.plist
651 The first parameter activates the tracing subsystem, the second
652 tells it to trace host and link categorized utilization and the
653 third creates a graph configuration file to configure Triva when
654 analysing the resulting trace file.
656 - SMPI simulator and traces for a space/time view:
660 The <i>-trace</i> parameter for the smpirun script runs the
661 simulation with --cfg=tracing:yes and --cfg=tracing/smpi:yes. Check the
662 smpirun's <i>-help</i> parameter for additional tracing options.
664 Sometimes you might want to put additional information on the trace to
665 correctly identify them later, or to provide data that can be used to
666 reproduce an experiment. You have two ways to do that:
668 - Add a string on top of the trace file as comment:
670 --cfg=tracing/comment:my_simulation_identifier
673 - Add the contents of a textual file on top of the trace file as comment:
675 --cfg=tracing/comment-file:my_file_with_additional_information.txt
678 Please, use these two parameters (for comments) to make reproducible
679 simulations. For additional details about this and all tracing
680 options, check See the \ref tracing_tracing_options.
682 \section options_msg Configuring MSG
684 \subsection options_msg_debug_multiple_use Debugging MSG
686 Sometimes your application may try to send a task that is still being
687 executed somewhere else, making it impossible to send this task. However,
688 for debugging purposes, one may want to know what the other host is/was
689 doing. This option shows a backtrace of the other process.
691 Enable this option by adding
694 --cfg=msg/debug-multiple-use:on
697 \section options_smpi Configuring SMPI
699 The SMPI interface provides several specific configuration items.
700 These are uneasy to see since the code is usually launched through the
701 \c smiprun script directly.
703 \subsection options_smpi_bench smpi/bench: Automatic benchmarking of SMPI code
705 In SMPI, the sequential code is automatically benchmarked, and these
706 computations are automatically reported to the simulator. That is to
707 say that if you have a large computation between a \c MPI_Recv() and a
708 \c MPI_Send(), SMPI will automatically benchmark the duration of this
709 code, and create an execution task within the simulator to take this
710 into account. For that, the actual duration is measured on the host
711 machine and then scaled to the power of the corresponding simulated
712 machine. The variable \b smpi/running-power allows to specify the
713 computational power of the host machine (in flop/s) to use when
714 scaling the execution times. It defaults to 20000, but you really want
715 to update it to get accurate simulation results.
717 When the code is constituted of numerous consecutive MPI calls, the
718 previous mechanism feeds the simulation kernel with numerous tiny
719 computations. The \b smpi/cpu-threshold item becomes handy when this
720 impacts badly the simulation performance. It specifies a threshold (in
721 seconds) below which the execution chunks are not reported to the
722 simulation kernel (default value: 1e-6).
726 The option smpi/cpu-threshold ignores any computation time spent
727 below this threshold. SMPI does not consider the \a amount of these
728 computations; there is no offset for this. Hence, by using a
729 value that is too low, you may end up with unreliable simulation
732 In some cases, however, one may wish to disable simulation of
733 application computation. This is the case when SMPI is used not to
734 simulate an MPI applications, but instead an MPI code that performs
735 "live replay" of another MPI app (e.g., ScalaTrace's replay tool,
736 various on-line simulators that run an app at scale). In this case the
737 computation of the replay/simulation logic should not be simulated by
738 SMPI. Instead, the replay tool or on-line simulator will issue
739 "computation events", which correspond to the actual MPI simulation
740 being replayed/simulated. At the moment, these computation events can
741 be simulated using SMPI by calling internal smpi_execute*() functions.
743 To disable the benchmarking/simulation of computation in the simulated
744 application, the variable \b smpi/simulate-computation should be set to no.
745 Equivalently, setting \b smpi/cpu-threshold to -1 also ignores all
749 This option just ignores the timings in your simulation; it still executes
750 the computations itself. If you want to stop SMPI from doing that,
751 you should check the SMPI_SAMPLE macros, documented in the chapter
752 \ref SMPI_adapting_speed.
754 \subsection options_model_smpi_adj_file smpi/comp-adjustment-file: Slow-down or speed-up parts of your code.
756 This option allows you to pass a file that contains two columns: The first column
757 defines the section that will be subject to a speedup; the second column is the speedup.
763 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
766 The first line is the header - you must include it.
767 The following line means that the code between two consecutive MPI calls on
768 line 30 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
769 of 1.18244559422142. The value for the second column is therefore a speedup, if it is
770 larger than 1 and a slow-down if it is smaller than 1. Nothing will be changed if it is
773 Of course, you can set any arbitrary filenames you want (so the start and end don't have to be
774 in the same file), but be aware that this mechanism only supports @em consecutive calls!
777 Please note that you must pass the \b -trace-call-location flag to smpicc
778 or smpiff, respectively! This flag activates some macro definitions in our
779 mpi.h / mpi.f files that help with obtaining the call location.
781 \subsection options_model_smpi_bw_factor smpi/bw-factor: Bandwidth factors
783 The possible throughput of network links is often dependent on the
784 message sizes, as protocols may adapt to different message sizes. With
785 this option, a series of message sizes and factors are given, helping
786 the simulation to be more realistic. For instance, the current
790 65472:0.940694;15424:0.697866;9376:0.58729;5776:1.08739;3484:0.77493;1426:0.608902;732:0.341987;257:0.338112;0:0.812084
793 So, messages with size 65472 and more will get a total of MAX_BANDWIDTH*0.940694,
794 messages of size 15424 to 65471 will get MAX_BANDWIDTH*0.697866 and so on.
795 Here, MAX_BANDWIDTH denotes the bandwidth of the link.
798 The SimGrid-Team has developed a script to help you determine these
799 values. You can find more information and the download here:
800 1. http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
801 2. http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
803 \subsection options_smpi_timing smpi/display-timing: Reporting simulation time
805 \b Default: 0 (false)
807 Most of the time, you run MPI code with SMPI to compute the time it
808 would take to run it on a platform. But since the
809 code is run through the \c smpirun script, you don't have any control
810 on the launcher code, making it difficult to report the simulated time
811 when the simulation ends. If you set the \b smpi/display-timing item
812 to 1, \c smpirun will display this information when the simulation ends. \verbatim
813 Simulation time: 1e3 seconds.
816 \subsection options_model_smpi_lat_factor smpi/lat-factor: Latency factors
818 The motivation and syntax for this option is identical to the motivation/syntax
819 of smpi/bw-factor, see \ref options_model_smpi_bw_factor for details.
821 There is an important difference, though: While smpi/bw-factor \a reduces the
822 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
823 increase the latency, i.e., values larger than or equal to 1 are valid here.
825 This is the default value:
828 65472:11.6436;15424:3.48845;9376:2.59299;5776:2.18796;3484:1.88101;1426:1.61075;732:1.9503;257:1.95341;0:2.01467
832 The SimGrid-Team has developed a script to help you determine these
833 values. You can find more information and the download here:
834 1. http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
835 2. http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
837 \subsection options_smpi_papi_events smpi/papi-events: Trace hardware counters with PAPI
840 This option is experimental and will be subject to change.
841 This feature currently requires superuser privileges, as registers are queried.
842 Only use this feature with code you trust! Call smpirun for instance via
843 smpirun -wrapper "sudo " <your-parameters>
844 or run sudo sh -c "echo 0 > /proc/sys/kernel/perf_event_paranoid"
845 In the later case, sudo will not be required.
848 This option is only available when SimGrid was compiled with PAPI support.
850 This option takes the names of PAPI counters and adds their respective values
851 to the trace files. (See Section \ref tracing_tracing_options.)
853 It is planned to make this feature available on a per-process (or per-thread?) basis.
854 The first draft, however, just implements a "global" (i.e., for all processes) set
855 of counters, the "default" set.
858 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
861 \subsection options_smpi_global smpi/privatize-global-variables: Automatic privatization of global variables
863 MPI executables are meant to be executed in separated processes, but SMPI is
864 executed in only one process. Global variables from executables will be placed
865 in the same memory zone and shared between processes, causing hard to find bugs.
866 To avoid this, several options are possible :
867 - Manual edition of the code, for example to add __thread keyword before data
868 declaration, which allows the resulting code to work with SMPI, but only
869 if the thread factory (see \ref options_virt_factory) is used, as global
870 variables are then placed in the TLS (thread local storage) segment.
871 - Source-to-source transformation, to add a level of indirection
872 to the global variables. SMPI does this for F77 codes compiled with smpiff,
873 and used to provide coccinelle scripts for C codes, which are not functional anymore.
874 - Compilation pass, to have the compiler automatically put the data in
876 - Runtime automatic switching of the data segments. SMPI stores a copy of
877 each global data segment for each process, and at each context switch replaces
878 the actual data with its copy from the right process. This mechanism uses mmap,
879 and is for now limited to systems supporting this functionnality (all Linux
880 and some BSD should be compatible).
881 Another limitation is that SMPI only accounts for global variables defined in
882 the executable. If the processes use external global variables from dynamic
883 libraries, they won't be switched correctly. To avoid this, using static
884 linking is advised (but not with the simgrid library, to avoid replicating
885 its own global variables).
887 To use this runtime automatic switching, the variable \b smpi/privatize-global-variables
891 This configuration option cannot be set in your platform file. You can only
892 pass it as an argument to smpirun.
895 \subsection options_model_smpi_detached Simulating MPI detached send
897 This threshold specifies the size in bytes under which the send will return
898 immediately. This is different from the threshold detailed in \ref options_model_network_asyncsend
899 because the message is not effectively sent when the send is posted. SMPI still waits for the
900 correspondant receive to be posted to perform the communication operation. This threshold can be set
901 by changing the \b smpi/send-is-detached-thresh item. The default value is 65536.
903 \subsection options_model_smpi_collectives Simulating MPI collective algorithms
905 SMPI implements more than 100 different algorithms for MPI collective communication, to accurately
906 simulate the behavior of most of the existing MPI libraries. The \b smpi/coll-selector item can be used
907 to use the decision logic of either OpenMPI or MPICH libraries (values: ompi or mpich, by default SMPI
908 uses naive version of collective operations). Each collective operation can be manually selected with a
909 \b smpi/collective_name:algo_name. Available algorithms are listed in \ref SMPI_collective_algorithms .
911 \subsection options_model_smpi_iprobe smpi/iprobe: Inject constant times for calls to MPI_Iprobe
913 \b Default value: 0.0001
915 The behavior and motivation for this configuration option is identical with \a smpi/test, see
916 Section \ref options_model_smpi_test for details.
918 \subsection options_model_smpi_init smpi/init: Inject constant times for calls to MPI_Init
922 The behavior for this configuration option is identical with \a smpi/test, see
923 Section \ref options_model_smpi_test for details.
925 \subsection options_model_smpi_ois smpi/ois: Inject constant times for asynchronous send operations
927 This configuration option works exactly as \a smpi/os, see Section \ref options_model_smpi_os.
928 Of course, \a smpi/ois is used to account for MPI_Isend instead of MPI_Send.
930 \subsection options_model_smpi_os smpi/os: Inject constant times for send operations
932 In several network models such as LogP, send (MPI_Send, MPI_Isend) and receive (MPI_Recv)
933 operations incur costs (i.e., they consume CPU time). SMPI can factor these costs in as well, but the
934 user has to configure SMPI accordingly as these values may vary by machine.
935 This can be done by using smpi/os for MPI_Send operations; for MPI_Isend and
936 MPI_Recv, use \a smpi/ois and \a smpi/or, respectively. These work exactly as
939 \a smpi/os can consist of multiple sections; each section takes three values, for example:
945 Here, the sections are divided by ";" (that is, this example contains two sections).
946 Furthermore, each section consists of three values.
948 1. The first value denotes the minimum size for this section to take effect;
949 read it as "if message size is greater than this value (and other section has a larger
950 first value that is also smaller than the message size), use this".
951 In the first section above, this value is "1".
953 2. The second value is the startup time; this is a constant value that will always
954 be charged, no matter what the size of the message. In the first section above,
957 3. The third value is the \a per-byte cost. That is, it is charged for every
958 byte of the message (incurring cost messageSize*cost_per_byte)
959 and hence accounts also for larger messages. In the first
960 section of the example above, this value is "2".
962 Now, SMPI always checks which section it should take for a given message; that is,
963 if a message of size 11 is sent with the configuration of the example above, only
964 the second section will be used, not the first, as the first value of the second
965 section is closer to the message size. Hence, a message of size 11 incurs the
966 following cost inside MPI_Send:
972 %As 5 is the startup cost and 1 is the cost per byte.
975 The order of sections can be arbitrary; they will be ordered internally.
977 \subsection options_model_smpi_or smpi/or: Inject constant times for receive operations
979 This configuration option works exactly as \a smpi/os, see Section \ref options_model_smpi_os.
980 Of course, \a smpi/or is used to account for MPI_Recv instead of MPI_Send.
982 \subsection options_model_smpi_test smpi/test: Inject constant times for calls to MPI_Test
984 \b Default value: 0.0001
986 By setting this option, you can control the amount of time a process sleeps
987 when MPI_Test() is called; this is important, because SimGrid normally only
988 advances the time while communication is happening and thus,
989 MPI_Test will not add to the time, resulting in a deadlock if used as a
996 MPI_Test(request, flag, status);
1002 Internally, in order to speed up execution, we use a counter to keep track
1003 on how often we already checked if the handle is now valid or not. Hence, we
1004 actually use counter*SLEEP_TIME, that is, the time MPI_Test() causes the process
1005 to sleep increases linearly with the number of previously failed testk.
1008 \subsection options_model_smpi_use_shared_malloc smpi/use-shared-malloc: Factorize malloc()s
1012 SMPI can use shared memory by calling shm_* functions; this might speed up the simulation.
1013 This opens or creates a new POSIX shared memory object, kept in RAM, in /dev/shm.
1015 If you want to disable this behavior, set the value to 0.
1017 \subsection options_model_smpi_wtime smpi/wtime: Inject constant times for calls to MPI_Wtime
1021 By setting this option, you can control the amount of time a process sleeps
1022 when MPI_Wtime() is called; this is important, because SimGrid normally only
1023 advances the time while communication is happening and thus,
1024 MPI_Wtime will not add to the time, resulting in a deadlock if used as a
1030 while(MPI_Wtime() < some_time_bound) {
1035 If the time is never advanced, this loop will clearly never end as MPI_Wtime()
1036 always returns the same value. Hence, pass a (small) value to the smpi/wtime
1037 option to force a call to MPI_Wtime to advance the time as well.
1040 \section options_generic Configuring other aspects of SimGrid
1042 \subsection options_generic_clean_atexit Cleanup before termination
1044 The C / C++ standard contains a function called \b [atexit](http://www.cplusplus.com/reference/cstdlib/atexit/).
1045 atexit registers callbacks, which are called just before the program terminates.
1047 By setting the configuration option clean-atexit to 1 (true), a callback
1048 is registered and will clean up some variables and terminate/cleanup the tracing.
1050 TODO: Add when this should be used.
1052 \subsection options_generic_path XML file inclusion path
1054 It is possible to specify a list of directories to search into for the
1055 \<include\> tag in XML files by using the \b path configuration
1056 item. To add several directory to the path, set the configuration
1057 item several times, as in \verbatim
1058 --cfg=path:toto --cfg=path:tutu
1061 \subsection options_generic_exit Behavior on Ctrl-C
1063 By default, when Ctrl-C is pressed, the status of all existing
1064 simulated processes is displayed before exiting the simulation. This is very useful to debug your
1065 code, but it can reveal troublesome in some cases (such as when the
1066 amount of processes becomes really big). This behavior is disabled
1067 when \b verbose-exit is set to 0 (it is to 1 by default).
1069 \subsection options_exception_cutpath Truncate local path from exception backtrace
1072 --cfg=exceptions/cutpath:1
1075 This configuration option is used to remove the path from the
1076 backtrace shown when an exception is thrown. This is mainly useful for
1077 the tests: the full file path makes the tests not reproducible, and
1078 thus failing as we are currently comparing output. Clearly, the path
1079 used on different machines are almost guaranteed to be different and
1080 hence, the output would mismatch, causing the test to fail.
1082 \section options_log Logging Configuration
1084 It can be done by using XBT. Go to \ref XBT_log for more details.
1086 \section options_perf Performance optimizations
1088 \subsection options_perf_context Context factory
1090 In order to achieve higher performance, you might want to use the raw
1091 context factory which avoids any system call when switching between
1092 tasks. If it is not possible you might use ucontext instead.
1094 \subsection options_perf_guard_size Disabling stack guard pages
1096 A stack guard page is usually used which prevents the stack from
1097 overflowing on other parts of the memory. However this might have a
1098 performance impact if a huge number of processes is created. The
1099 option \b contexts:guard-size is the number of stack guard pages
1100 used. By setting it to 0, no guard pages will be used: in this case,
1101 you should avoid using small stacks (\b stack-size) as the stack will
1102 silently overflow on other parts of the memory.
1104 \section options_index Index of all existing configuration options
1107 Almost all options are defined in <i>src/simgrid/sg_config.c</i>. You may
1108 want to check this file, too, but this index should be somewhat complete
1109 for the moment (May 2015).
1112 \b Please \b note: You can also pass the command-line option "--help" and
1113 "--help-cfg" to an executable that uses simgrid.
1115 - \c clean-atexit: \ref options_generic_clean_atexit
1117 - \c contexts/factory: \ref options_virt_factory
1118 - \c contexts/guard-size: \ref options_virt_parallel
1119 - \c contexts/nthreads: \ref options_virt_parallel
1120 - \c contexts/parallel_threshold: \ref options_virt_parallel
1121 - \c contexts/stack-size: \ref options_virt_stacksize
1122 - \c contexts/synchro: \ref options_virt_parallel
1124 - \c cpu/maxmin-selective-update: \ref options_model_optim
1125 - \c cpu/model: \ref options_model_select
1126 - \c cpu/optim: \ref options_model_optim
1128 - \c exception/cutpath: \ref options_exception_cutpath
1130 - \c host/model: \ref options_model_select
1132 - \c maxmin/precision: \ref options_model_precision
1134 - \c msg/debug-multiple-use: \ref options_msg_debug_multiple_use
1136 - \c model-check: \ref options_modelchecking
1137 - \c model-check/checkpoint: \ref options_modelchecking_steps
1138 - \c model-check/communications-determinism: \ref options_modelchecking_comm_determinism
1139 - \c model-check/dot-output: \ref options_modelchecking_dot_output
1140 - \c model-check/hash: \ref options_modelchecking_hash
1141 - \c model-check/property: \ref options_modelchecking_liveness
1142 - \c model-check/max-depth: \ref options_modelchecking_max_depth
1143 - \c model-check/record: \ref options_modelchecking_recordreplay
1144 - \c model-check/reduction: \ref options_modelchecking_reduction
1145 - \c model-check/replay: \ref options_modelchecking_recordreplay
1146 - \c model-check/send-determinism: \ref options_modelchecking_comm_determinism
1147 - \c model-check/sparse-checkpoint: \ref options_modelchecking_sparse_checkpoint
1148 - \c model-check/termination: \ref options_modelchecking_termination
1149 - \c model-check/timeout: \ref options_modelchecking_timeout
1150 - \c model-check/visited: \ref options_modelchecking_visited
1152 - \c network/bandwidth-factor: \ref options_model_network_coefs
1153 - \c network/coordinates: \ref options_model_network_coord
1154 - \c network/crosstraffic: \ref options_model_network_crosstraffic
1155 - \c network/latency-factor: \ref options_model_network_coefs
1156 - \c network/maxmin-selective-update: \ref options_model_optim
1157 - \c network/model: \ref options_model_select
1158 - \c network/optim: \ref options_model_optim
1159 - \c network/sender_gap: \ref options_model_network_sendergap
1160 - \c network/TCP-gamma: \ref options_model_network_gamma
1161 - \c network/weight-S: \ref options_model_network_coefs
1163 - \c ns3/TcpModel: \ref options_pls
1164 - \c path: \ref options_generic_path
1165 - \c plugin: \ref options_generic_plugin
1167 - \c surf/nthreads: \ref options_model_nthreads
1168 - \c surf/precision: \ref options_model_precision
1170 - \c <b>For collective operations of SMPI, please refer to Section \ref options_index_smpi_coll</b>
1171 - \c smpi/async-small-thresh: \ref options_model_network_asyncsend
1172 - \c smpi/bw-factor: \ref options_model_smpi_bw_factor
1173 - \c smpi/coll-selector: \ref options_model_smpi_collectives
1174 - \c smpi/comp-adjustment-file: \ref options_model_smpi_adj_file
1175 - \c smpi/cpu-threshold: \ref options_smpi_bench
1176 - \c smpi/display-timing: \ref options_smpi_timing
1177 - \c smpi/lat-factor: \ref options_model_smpi_lat_factor
1178 - \c smpi/IB-penalty-factors: \ref options_model_network_coefs
1179 - \c smpi/iprobe: \ref options_model_smpi_iprobe
1180 - \c smpi/init: \ref options_model_smpi_init
1181 - \c smpi/ois: \ref options_model_smpi_ois
1182 - \c smpi/or: \ref options_model_smpi_or
1183 - \c smpi/os: \ref options_model_smpi_os
1184 - \c smpi/papi-events: \ref options_smpi_papi_events
1185 - \c smpi/privatize-global-variables: \ref options_smpi_global
1186 - \c smpi/running-power: \ref options_smpi_bench
1187 - \c smpi/send-is-detached-thresh: \ref options_model_smpi_detached
1188 - \c smpi/simulate-computation: \ref options_smpi_bench
1189 - \c smpi/test: \ref options_model_smpi_test
1190 - \c smpi/use-shared-malloc: \ref options_model_smpi_use_shared_malloc
1191 - \c smpi/wtime: \ref options_model_smpi_wtime
1193 - \c <b>Tracing configuration options can be found in Section \ref tracing_tracing_options</b>.
1195 - \c storage/model: \ref options_storage_model
1196 - \c verbose-exit: \ref options_generic_exit
1198 - \c vm/model: \ref options_vm_model
1200 \subsection options_index_smpi_coll Index of SMPI collective algorithms options
1202 TODO: All available collective algorithms will be made available via the ``smpirun --help-coll`` command.