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
144 \subsection options_generic_plugin Plugins
146 SimGrid supports the use of plugins; currently, no known plugins
147 can be activated but there are use-cases where you may want to write
148 your own plugin (for instance, for logging).
150 Plugins can for instance define own classes that inherit from
151 existing classes (for instance, a class "CpuEnergy" inherits from
152 "Cpu" to assess energy consumption).
154 The plugin connects to the code by registering callbacks using
155 ``signal.connect(callback)`` (see file ``src/surf/plugins/energy.cpp`` for
163 This option is case-sensitive: Energy and energy are not the same!
165 \subsection options_model_optim Optimization level of the platform models
167 The network and CPU models that are based on lmm_solve (that
168 is, all our analytical models) accept specific optimization
170 - items \b network/optim and \b CPU/optim (both default to 'Lazy'):
171 - \b Lazy: Lazy action management (partial invalidation in lmm +
172 heap in action remaining).
173 - \b TI: Trace integration. Highly optimized mode when using
174 availability traces (only available for the Cas01 CPU model for
176 - \b Full: Full update of remaining and variables. Slow but may be
177 useful when debugging.
178 - items \b network/maxmin-selective-update and
179 \b cpu/maxmin-selective-update: configure whether the underlying
180 should be lazily updated or not. It should have no impact on the
181 computed timings, but should speed up the computation.
183 It is still possible to disable the \c maxmin-selective-update feature
184 because it can reveal counter-productive in very specific scenarios
185 where the interaction level is high. In particular, if all your
186 communication share a given backbone link, you should disable it:
187 without \c maxmin-selective-update, every communications are updated
188 at each step through a simple loop over them. With that feature
189 enabled, every communications will still get updated in this case
190 (because of the dependency induced by the backbone), but through a
191 complicated pattern aiming at following the actual dependencies.
193 \subsection options_model_precision Numerical precision of the platform models
195 The analytical models handle a lot of floating point values. It is
196 possible to change the epsilon used to update and compare them through
197 the \b maxmin/precision item (default value: 0.00001). Changing it
198 may speedup the simulation by discarding very small actions, at the
199 price of a reduced numerical precision.
201 \subsection options_model_nthreads Parallel threads for model updates
203 By default, Surf computes the analytical models sequentially to share their
204 resources and update their actions. It is possible to run them in parallel,
205 using the \b surf/nthreads item (default value: 1). If you use a
206 negative or null value, the amount of available cores is automatically
207 detected and used instead.
209 Depending on the workload of the models and their complexity, you may get a
210 speedup or a slowdown because of the synchronization costs of threads.
212 \subsection options_model_network Configuring the Network model
214 \subsubsection options_model_network_gamma Maximal TCP window size
216 The analytical models need to know the maximal TCP window size to take
217 the TCP congestion mechanism into account. This is set to 20000 by
218 default, but can be changed using the \b network/TCP-gamma item.
220 On linux, this value can be retrieved using the following
221 commands. Both give a set of values, and you should use the last one,
222 which is the maximal size.\verbatim
223 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
224 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
227 \subsubsection options_model_network_coefs Correcting important network parameters
229 SimGrid can take network irregularities such as a slow startup or
230 changing behavior depending on the message size into account.
231 You should not change these values unless you really know what you're doing.
233 The corresponding values were computed through data fitting one the
234 timings of packet-level simulators.
237 <a href="http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf">Accuracy Study and Improvement of Network Simulation in the SimGrid Framework</a>
238 for more information about these parameters.
240 If you are using the SMPI model, these correction coefficients are
241 themselves corrected by constant values depending on the size of the
242 exchange. Again, only hardcore experts should bother about this fact.
244 InfiniBand network behavior can be modeled through 3 parameters, as explained in
245 <a href="http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf">this PhD thesis</a>.
246 These factors can be changed through the following option:
249 smpi/IB-penalty-factors:"βe;βs;γs"
252 By default SMPI uses factors computed on the Stampede Supercomputer at TACC, with optimal
253 deployment of processes on nodes.
255 \subsubsection options_model_network_crosstraffic Simulating cross-traffic
257 %As of SimGrid v3.7, cross-traffic effects can be taken into account in
258 analytical simulations. It means that ongoing and incoming
259 communication flows are treated independently. In addition, the LV08
260 model adds 0.05 of usage on the opposite direction for each new
261 created flow. This can be useful to simulate some important TCP
262 phenomena such as ack compression.
264 For that to work, your platform must have two links for each
265 pair of interconnected hosts. An example of usable platform is
266 available in <tt>examples/platforms/crosstraffic.xml</tt>.
268 This is activated through the \b network/crosstraffic item, that
269 can be set to 0 (disable this feature) or 1 (enable it).
271 Note that with the default host model this option is activated by default.
273 \subsubsection options_model_network_coord Coordinated-based network models
275 When you want to use network coordinates, as it happens when you use
276 an \<AS\> in your platform file with \c Vivaldi as a routing (see also
277 Section \ref pf_routing_model_vivaldi "Vivaldi Routing Model"), you must
278 set the \b network/coordinates to \c yes so that all mandatory
279 initialization are done in the simulator.
281 \subsubsection options_model_network_sendergap Simulating sender gap
283 (this configuration item is experimental and may change or disapear)
285 It is possible to specify a timing gap between consecutive emission on
286 the same network card through the \b network/sender-gap item. This
287 is still under investigation as of writting, and the default value is
288 to wait 10 microseconds (1e-5 seconds) between emissions.
290 \subsubsection options_model_network_asyncsend Simulating asyncronous send
292 (this configuration item is experimental and may change or disapear)
294 It is possible to specify that messages below a certain size will be sent
295 as soon as the call to MPI_Send is issued, without waiting for the
296 correspondant receive. This threshold can be configured through the
297 \b smpi/async-small-thresh item. The default value is 0. This behavior can also be
298 manually set for MSG mailboxes, by setting the receiving mode of the mailbox
299 with a call to \ref MSG_mailbox_set_async . For MSG, all messages sent to this
300 mailbox will have this behavior, so consider using two mailboxes if needed.
302 This value needs to be smaller than or equals to the threshold set at
303 \ref options_model_smpi_detached , because asynchronous messages are
304 meant to be detached as well.
306 \subsubsection options_pls Configuring packet-level pseudo-models
308 When using the packet-level pseudo-models, several specific
309 configuration flags are provided to configure the associated tools.
310 There is by far not enough such SimGrid flags to cover every aspects
311 of the associated tools, since we only added the items that we
312 needed ourselves. Feel free to request more items (or even better:
313 provide patches adding more items).
315 When using NS3, the only existing item is \b ns3/TcpModel,
316 corresponding to the ns3::TcpL4Protocol::SocketType configuration item
317 in NS3. The only valid values (enforced on the SimGrid side) are
318 'NewReno' or 'Reno' or 'Tahoe'.
320 \section options_modelchecking Configuring the Model-Checking
322 To enable the SimGrid model-checking support the program should
323 be executed using the simgrid-mc wrapper:
325 simgrid-mc ./my_program
328 Safety properties are expressed as assertions using the function
330 void MC_assert(int prop);
333 \subsection options_modelchecking_liveness Specifying a liveness property
335 If you want to specify liveness properties (beware, that's
336 experimental), you have to pass them on the command line, specifying
337 the name of the file containing the property, as formatted by the
341 --cfg=model-check/property:<filename>
344 \subsection options_modelchecking_steps Going for stateful verification
346 By default, the system is backtracked to its initial state to explore
347 another path instead of backtracking to the exact step before the fork
348 that we want to explore (this is called stateless verification). This
349 is done this way because saving intermediate states can rapidly
350 exhaust the available memory. If you want, you can change the value of
351 the <tt>model-check/checkpoint</tt> variable. For example, the
352 following configuration will ask to take a checkpoint every step.
353 Beware, this will certainly explode your memory. Larger values are
354 probably better, make sure to experiment a bit to find the right
355 setting for your specific system.
358 --cfg=model-check/checkpoint:1
361 \subsection options_modelchecking_reduction Specifying the kind of reduction
363 The main issue when using the model-checking is the state space
364 explosion. To counter that problem, several exploration reduction
365 techniques can be used. There is unfortunately no silver bullet here,
366 and the most efficient reduction techniques cannot be applied to any
367 properties. In particular, the DPOR method cannot be applied on
368 liveness properties since it may break some cycles in the exploration
369 that are important to the property validity.
372 --cfg=model-check/reduction:<technique>
375 For now, this configuration variable can take 2 values:
376 * none: Do not apply any kind of reduction (mandatory for now for
378 * dpor: Apply Dynamic Partial Ordering Reduction. Only valid if you
379 verify local safety properties.
381 \subsection options_modelchecking_visited model-check/visited, Cycle detection
383 In order to detect cycles, the model-checker needs to check if a new explored
384 state is in fact the same state than a previous one. In order to do this,
385 the model-checker can take a snapshot of each visited state: this snapshot is
386 then used to compare it with subsequent states in the exploration graph.
388 The \b model-check/visited is the maximum number of states which are stored in
389 memory. If the maximum number of snapshotted state is reached some states will
390 be removed from the memory and some cycles might be missed.
392 By default, no state is snapshotted and cycles cannot be detected.
394 \subsection options_modelchecking_termination model-check/termination, Non termination detection
396 The \b model-check/termination configuration item can be used to report if a
397 non-termination execution path has been found. This is a path with a cycle
398 which means that the program might never terminate.
400 This only works in safety mode.
402 This options is disabled by default.
404 \subsection options_modelchecking_dot_output model-check/dot-output, Dot output
406 If set, the \b model-check/dot-output configuration item is the name of a file
407 in which to write a dot file of the path leading the found property (safety or
408 liveness violation) as well as the cycle for liveness properties. This dot file
409 can then fed to the graphviz dot tool to generate an corresponding graphical
412 \subsection options_modelchecking_max_depth model-check/max_depth, Depth limit
414 The \b model-checker/max-depth can set the maximum depth of the exploration
415 graph of the model-checker. If this limit is reached, a logging message is
416 sent and the results might not be exact.
418 By default, there is not depth limit.
420 \subsection options_modelchecking_timeout Handling of timeout
422 By default, the model-checker does not handle timeout conditions: the `wait`
423 operations never time out. With the \b model-check/timeout configuration item
424 set to \b yes, the model-checker will explore timeouts of `wait` operations.
426 \subsection options_modelchecking_comm_determinism Communication determinism
428 The \b model-check/communications-determinism and
429 \b model-check/send-determinism items can be used to select the communication
430 determinism mode of the model-checker which checks determinism properties of
431 the communications of an application.
433 \subsection options_modelchecking_sparse_checkpoint Per page checkpoints
435 When the model-checker is configured to take a snapshot of each explored state
436 (with the \b model-checker/visited item), the memory consumption can rapidly
437 reach GiB ou Tib of memory. However, for many workloads, the memory does not
438 change much between different snapshots and taking a complete copy of each
439 snapshot is a waste of memory.
441 The \b model-check/sparse-checkpoint option item can be set to \b yes in order
442 to avoid making a complete copy of each snapshot: instead, each snapshot will be
443 decomposed in blocks which will be stored separately.
444 If multiple snapshots share the same block (or if the same block
445 is used in the same snapshot), the same copy of the block will be shared leading
446 to a reduction of the memory footprint.
448 For many applications, this option considerably reduces the memory consumption.
449 In somes cases, the model-checker might be slightly slower because of the time
450 taken to manage the metadata about the blocks. In other cases however, this
451 snapshotting strategy will be much faster by reducing the cache consumption.
452 When the memory consumption is important, by avoiding to hit the swap or
453 reducing the swap usage, this option might be much faster than the basic
454 snapshotting strategy.
456 This option is currently disabled by default.
458 \subsection options_mc_perf Performance considerations for the model checker
460 The size of the stacks can have a huge impact on the memory
461 consumption when using model-checking. By default, each snapshot will
462 save a copy of the whole stacks and not only of the part which is
463 really meaningful: you should expect the contribution of the memory
464 consumption of the snapshots to be \f$ \mbox{number of processes}
465 \times \mbox{stack size} \times \mbox{number of states} \f$.
467 The \b model-check/sparse-checkpoint can be used to reduce the memory
468 consumption by trying to share memory between the different snapshots.
470 When compiled against the model checker, the stacks are not
471 protected with guards: if the stack size is too small for your
472 application, the stack will silently overflow on other parts of the
475 \subsection options_modelchecking_hash Hashing of the state (experimental)
477 Usually most of the time of the model-checker is spent comparing states. This
478 process is complicated and consumes a lot of bandwidth and cache.
479 In order to speedup the state comparison, the experimental \b model-checker/hash
480 configuration item enables the computation of a hash summarizing as much
481 information of the state as possible into a single value. This hash can be used
482 to avoid most of the comparisons: the costly comparison is then only used when
483 the hashes are identical.
485 Currently most of the state is not included in the hash because the
486 implementation was found to be buggy and this options is not as useful as
487 it could be. For this reason, it is currently disabled by default.
489 \subsection options_modelchecking_recordreplay Record/replay (experimental)
491 As the model-checker keeps jumping at different places in the execution graph,
492 it is difficult to understand what happens when trying to debug an application
493 under the model-checker. Event the output of the program is difficult to
494 interpret. Moreover, the model-checker does not behave nicely with advanced
495 debugging tools such as valgrind. For those reason, to identify a trajectory
496 in the execution graph with the model-checker and replay this trajcetory and
497 without the model-checker black-magic but with more standard tools
498 (such as a debugger, valgrind, etc.). For this reason, Simgrid implements an
499 experimental record/replay functionnality in order to record a trajectory with
500 the model-checker and replay it without the model-checker.
502 When the model-checker finds an interesting path in the application execution
503 graph (where a safety or liveness property is violated), it can generate an
504 identifier for this path. In order to enable this behavious the
505 \b model-check/record must be set to \b yes. By default, this behaviour is not
508 This is an example of output:
511 [ 0.000000] (0:@) Check a safety property
512 [ 0.000000] (0:@) **************************
513 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
514 [ 0.000000] (0:@) **************************
515 [ 0.000000] (0:@) Counter-example execution trace:
516 [ 0.000000] (0:@) Path = 1/3;1/4
517 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
518 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
519 [ 0.000000] (0:@) Expanded states = 27
520 [ 0.000000] (0:@) Visited states = 68
521 [ 0.000000] (0:@) Executed transitions = 46
524 This path can then be replayed outside of the model-checker (and even in
525 non-MC build of simgrid) by setting the \b model-check/replay item to the given
526 path. The other options should be the same (but the model-checker should
529 The format and meaning of the path may change between different releases so
530 the same release of Simgrid should be used for the record phase and the replay
533 \section options_virt Configuring the User Process Virtualization
535 \subsection options_virt_factory Selecting the virtualization factory
537 In SimGrid, the user code is virtualized in a specific mechanism
538 that allows the simulation kernel to control its execution: when a user
539 process requires a blocking action (such as sending a message), it is
540 interrupted, and only gets released when the simulated clock reaches
541 the point where the blocking operation is done.
543 In SimGrid, the containers in which user processes are virtualized are
544 called contexts. Several context factory are provided, and you can
545 select the one you want to use with the \b contexts/factory
546 configuration item. Some of the following may not exist on your
547 machine because of portability issues. In any case, the default one
548 should be the most effcient one (please report bugs if the
549 auto-detection fails for you). They are sorted here from the slowest
551 - \b thread: very slow factory using full featured threads (either
552 pthreads or windows native threads)
553 - \b ucontext: fast factory using System V contexts (or a portability
554 layer of our own on top of Windows fibers)
555 - \b raw: amazingly fast factory using a context switching mechanism
556 of our own, directly implemented in assembly (only available for x86
557 and amd64 platforms for now)
558 - \b boost: This uses the [context implementation](http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html)
559 of the boost library; you must have this library installed before
560 you compile SimGrid. (On Debian GNU/Linux based systems, this is
561 provided by the libboost-contexts-dev package.)
563 The only reason to change this setting is when the debugging tools get
564 fooled by the optimized context factories. Threads are the most
565 debugging-friendly contextes, as they allow to set breakpoints anywhere with gdb
566 and visualize backtraces for all processes, in order to debug concurrency issues.
567 Valgrind is also more comfortable with threads, but it should be usable with all factories.
569 \subsection options_virt_stacksize Adapting the used stack size
571 Each virtualized used process is executed using a specific system
572 stack. The size of this stack has a huge impact on the simulation
573 scalability, but its default value is rather large. This is because
574 the error messages that you get when the stack size is too small are
575 rather disturbing: this leads to stack overflow (overwriting other
576 stacks), leading to segfaults with corrupted stack traces.
578 If you want to push the scalability limits of your code, you might
579 want to reduce the \b contexts/stack-size item. Its default value
580 is 8192 (in KiB), while our Chord simulation works with stacks as small
581 as 16 KiB, for example. For the thread factory, the default value
582 is the one of the system, if it is too large/small, it has to be set
585 The operating system should only allocate memory for the pages of the
586 stack which are actually used and you might not need to use this in
587 most cases. However, this setting is very important when using the
588 model checker (see \ref options_mc_perf).
590 In some cases, no stack guard page is used and the stack will silently
591 overflow on other parts of the memory if the stack size is too small
592 for your application. This happens :
594 - on Windows systems;
595 - when the model checker is enabled;
596 - when stack guard pages are explicitely disabled (see \ref options_perf_guard_size).
598 \subsection options_virt_parallel Running user code in parallel
600 Parallel execution of the user code is only considered stable in
601 SimGrid v3.7 and higher. It is described in
602 <a href="http://hal.inria.fr/inria-00602216/">INRIA RR-7653</a>.
604 If you are using the \c ucontext or \c raw context factories, you can
605 request to execute the user code in parallel. Several threads are
606 launched, each of them handling as much user contexts at each run. To
607 actiave this, set the \b contexts/nthreads item to the amount of
608 cores that you have in your computer (or lower than 1 to have
609 the amount of cores auto-detected).
611 Even if you asked several worker threads using the previous option,
612 you can request to start the parallel execution (and pay the
613 associated synchronization costs) only if the potential parallelism is
614 large enough. For that, set the \b contexts/parallel-threshold
615 item to the minimal amount of user contexts needed to start the
616 parallel execution. In any given simulation round, if that amount is
617 not reached, the contexts will be run sequentially directly by the
618 main thread (thus saving the synchronization costs). Note that this
619 option is mainly useful when the grain of the user code is very fine,
620 because our synchronization is now very efficient.
622 When parallel execution is activated, you can choose the
623 synchronization schema used with the \b contexts/synchro item,
624 which value is either:
625 - \b futex: ultra optimized synchronisation schema, based on futexes
626 (fast user-mode mutexes), and thus only available on Linux systems.
627 This is the default mode when available.
628 - \b posix: slow but portable synchronisation using only POSIX
630 - \b busy_wait: not really a synchronisation: the worker threads
631 constantly request new contexts to execute. It should be the most
632 efficient synchronisation schema, but it loads all the cores of your
633 machine for no good reason. You probably prefer the other less
636 \section options_tracing Configuring the tracing subsystem
638 The \ref outcomes_vizu "tracing subsystem" can be configured in several
639 different ways depending on the nature of the simulator (MSG, SimDag,
640 SMPI) and the kind of traces that need to be obtained. See the \ref
641 tracing_tracing_options "Tracing Configuration Options subsection" to
642 get a detailed description of each configuration option.
644 We detail here a simple way to get the traces working for you, even if
645 you never used the tracing API.
648 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
650 --cfg=tracing:yes --cfg=tracing/uncategorized:yes --cfg=triva/uncategorized:uncat.plist
652 The first parameter activates the tracing subsystem, the second
653 tells it to trace host and link utilization (without any
654 categorization) and the third creates a graph configuration file
655 to configure Triva when analysing the resulting trace file.
657 - MSG or SimDag-based simulator and categorized traces (you need to declare categories and classify your tasks according to them)
659 --cfg=tracing:yes --cfg=tracing/categorized:yes --cfg=triva/categorized:cat.plist
661 The first parameter activates the tracing subsystem, the second
662 tells it to trace host and link categorized utilization and the
663 third creates a graph configuration file to configure Triva when
664 analysing the resulting trace file.
666 - SMPI simulator and traces for a space/time view:
670 The <i>-trace</i> parameter for the smpirun script runs the
671 simulation with --cfg=tracing:yes and --cfg=tracing/smpi:yes. Check the
672 smpirun's <i>-help</i> parameter for additional tracing options.
674 Sometimes you might want to put additional information on the trace to
675 correctly identify them later, or to provide data that can be used to
676 reproduce an experiment. You have two ways to do that:
678 - Add a string on top of the trace file as comment:
680 --cfg=tracing/comment:my_simulation_identifier
683 - Add the contents of a textual file on top of the trace file as comment:
685 --cfg=tracing/comment-file:my_file_with_additional_information.txt
688 Please, use these two parameters (for comments) to make reproducible
689 simulations. For additional details about this and all tracing
690 options, check See the \ref tracing_tracing_options.
692 \section options_msg Configuring MSG
694 \subsection options_msg_debug_multiple_use Debugging MSG
696 Sometimes your application may try to send a task that is still being
697 executed somewhere else, making it impossible to send this task. However,
698 for debugging purposes, one may want to know what the other host is/was
699 doing. This option shows a backtrace of the other process.
701 Enable this option by adding
704 --cfg=msg/debug-multiple-use:on
707 \section options_smpi Configuring SMPI
709 The SMPI interface provides several specific configuration items.
710 These are uneasy to see since the code is usually launched through the
711 \c smiprun script directly.
713 \subsection options_smpi_bench smpi/bench: Automatic benchmarking of SMPI code
715 In SMPI, the sequential code is automatically benchmarked, and these
716 computations are automatically reported to the simulator. That is to
717 say that if you have a large computation between a \c MPI_Recv() and a
718 \c MPI_Send(), SMPI will automatically benchmark the duration of this
719 code, and create an execution task within the simulator to take this
720 into account. For that, the actual duration is measured on the host
721 machine and then scaled to the power of the corresponding simulated
722 machine. The variable \b smpi/running-power allows to specify the
723 computational power of the host machine (in flop/s) to use when
724 scaling the execution times. It defaults to 20000, but you really want
725 to update it to get accurate simulation results.
727 When the code is constituted of numerous consecutive MPI calls, the
728 previous mechanism feeds the simulation kernel with numerous tiny
729 computations. The \b smpi/cpu-threshold item becomes handy when this
730 impacts badly the simulation performance. It specifies a threshold (in
731 seconds) below which the execution chunks are not reported to the
732 simulation kernel (default value: 1e-6).
736 The option smpi/cpu-threshold ignores any computation time spent
737 below this threshold. SMPI does not consider the \a amount of these
738 computations; there is no offset for this. Hence, by using a
739 value that is too low, you may end up with unreliable simulation
742 In some cases, however, one may wish to disable simulation of
743 application computation. This is the case when SMPI is used not to
744 simulate an MPI applications, but instead an MPI code that performs
745 "live replay" of another MPI app (e.g., ScalaTrace's replay tool,
746 various on-line simulators that run an app at scale). In this case the
747 computation of the replay/simulation logic should not be simulated by
748 SMPI. Instead, the replay tool or on-line simulator will issue
749 "computation events", which correspond to the actual MPI simulation
750 being replayed/simulated. At the moment, these computation events can
751 be simulated using SMPI by calling internal smpi_execute*() functions.
753 To disable the benchmarking/simulation of computation in the simulated
754 application, the variable \b smpi/simulate-computation should be set to no.
755 Equivalently, setting \b smpi/cpu-threshold to -1 also ignores all
759 This option just ignores the timings in your simulation; it still executes
760 the computations itself. If you want to stop SMPI from doing that,
761 you should check the SMPI_SAMPLE macros, documented in the chapter
762 \ref SMPI_adapting_speed.
764 \subsection options_model_smpi_adj_file smpi/comp-adjustment-file: Slow-down or speed-up parts of your code.
766 This option allows you to pass a file that contains two columns: The first column
767 defines the section that will be subject to a speedup; the second column is the speedup.
773 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
776 The first line is the header - you must include it.
777 The following line means that the code between two consecutive MPI calls on
778 line 30 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
779 of 1.18244559422142. The value for the second column is therefore a speedup, if it is
780 larger than 1 and a slow-down if it is smaller than 1. Nothing will be changed if it is
783 Of course, you can set any arbitrary filenames you want (so the start and end don't have to be
784 in the same file), but be aware that this mechanism only supports @em consecutive calls!
787 Please note that you must pass the \b -trace-call-location flag to smpicc
788 or smpiff, respectively! This flag activates some macro definitions in our
789 mpi.h / mpi.f files that help with obtaining the call location.
791 \subsection options_model_smpi_bw_factor smpi/bw-factor: Bandwidth factors
793 The possible throughput of network links is often dependent on the
794 message sizes, as protocols may adapt to different message sizes. With
795 this option, a series of message sizes and factors are given, helping
796 the simulation to be more realistic. For instance, the current
800 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
803 So, messages with size 65472 and more will get a total of MAX_BANDWIDTH*0.940694,
804 messages of size 15424 to 65471 will get MAX_BANDWIDTH*0.697866 and so on.
805 Here, MAX_BANDWIDTH denotes the bandwidth of the link.
808 The SimGrid-Team has developed a script to help you determine these
809 values. You can find more information and the download here:
810 1. http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
811 2. http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
813 \subsection options_smpi_timing smpi/display-timing: Reporting simulation time
815 \b Default: 0 (false)
817 Most of the time, you run MPI code with SMPI to compute the time it
818 would take to run it on a platform. But since the
819 code is run through the \c smpirun script, you don't have any control
820 on the launcher code, making it difficult to report the simulated time
821 when the simulation ends. If you set the \b smpi/display-timing item
822 to 1, \c smpirun will display this information when the simulation ends. \verbatim
823 Simulation time: 1e3 seconds.
826 \subsection options_model_smpi_lat_factor smpi/lat-factor: Latency factors
828 The motivation and syntax for this option is identical to the motivation/syntax
829 of smpi/bw-factor, see \ref options_model_smpi_bw_factor for details.
831 There is an important difference, though: While smpi/bw-factor \a reduces the
832 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
833 increase the latency, i.e., values larger than or equal to 1 are valid here.
835 This is the default value:
838 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
842 The SimGrid-Team has developed a script to help you determine these
843 values. You can find more information and the download here:
844 1. http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
845 2. http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
847 \subsection options_smpi_global smpi/privatize-global-variables: Automatic privatization of global variables
849 MPI executables are meant to be executed in separated processes, but SMPI is
850 executed in only one process. Global variables from executables will be placed
851 in the same memory zone and shared between processes, causing hard to find bugs.
852 To avoid this, several options are possible :
853 - Manual edition of the code, for example to add __thread keyword before data
854 declaration, which allows the resulting code to work with SMPI, but only
855 if the thread factory (see \ref options_virt_factory) is used, as global
856 variables are then placed in the TLS (thread local storage) segment.
857 - Source-to-source transformation, to add a level of indirection
858 to the global variables. SMPI does this for F77 codes compiled with smpiff,
859 and used to provide coccinelle scripts for C codes, which are not functional anymore.
860 - Compilation pass, to have the compiler automatically put the data in
862 - Runtime automatic switching of the data segments. SMPI stores a copy of
863 each global data segment for each process, and at each context switch replaces
864 the actual data with its copy from the right process. This mechanism uses mmap,
865 and is for now limited to systems supporting this functionnality (all Linux
866 and some BSD should be compatible).
867 Another limitation is that SMPI only accounts for global variables defined in
868 the executable. If the processes use external global variables from dynamic
869 libraries, they won't be switched correctly. To avoid this, using static
870 linking is advised (but not with the simgrid library, to avoid replicating
871 its own global variables).
873 To use this runtime automatic switching, the variable \b smpi/privatize-global-variables
878 \subsection options_model_smpi_detached Simulating MPI detached send
880 This threshold specifies the size in bytes under which the send will return
881 immediately. This is different from the threshold detailed in \ref options_model_network_asyncsend
882 because the message is not effectively sent when the send is posted. SMPI still waits for the
883 correspondant receive to be posted to perform the communication operation. This threshold can be set
884 by changing the \b smpi/send-is-detached item. The default value is 65536.
886 \subsection options_model_smpi_collectives Simulating MPI collective algorithms
888 SMPI implements more than 100 different algorithms for MPI collective communication, to accurately
889 simulate the behavior of most of the existing MPI libraries. The \b smpi/coll-selector item can be used
890 to use the decision logic of either OpenMPI or MPICH libraries (values: ompi or mpich, by default SMPI
891 uses naive version of collective operations). Each collective operation can be manually selected with a
892 \b smpi/collective_name:algo_name. Available algorithms are listed in \ref SMPI_collective_algorithms .
894 \subsection options_model_smpi_iprobe smpi/iprobe: Inject constant times for calls to MPI_Iprobe
896 \b Default value: 0.0001
898 The behavior and motivation for this configuration option is identical with \a smpi/test, see
899 Section \ref options_model_smpi_test for details.
901 \subsection options_model_smpi_ois smpi/ois: Inject constant times for asynchronous send operations
903 This configuration option works exactly as \a smpi/os, see Section \ref options_model_smpi_os.
904 Of course, \a smpi/ois is used to account for MPI_Isend instead of MPI_Send.
906 \subsection options_model_smpi_os smpi/os: Inject constant times for send operations
908 In several network models such as LogP, send (MPI_Send, MPI_Isend) and receive (MPI_Recv)
909 operations incur costs (i.e., they consume CPU time). SMPI can factor these costs in as well, but the
910 user has to configure SMPI accordingly as these values may vary by machine.
911 This can be done by using smpi/os for MPI_Send operations; for MPI_Isend and
912 MPI_Recv, use \a smpi/ois and \a smpi/or, respectively. These work exactly as
915 \a smpi/os can consist of multiple sections; each section takes three values, for example:
921 Here, the sections are divided by ";" (that is, this example contains two sections).
922 Furthermore, each section consists of three values.
924 1. The first value denotes the minimum size for this section to take effect;
925 read it as "if message size is greater than this value (and other section has a larger
926 first value that is also smaller than the message size), use this".
927 In the first section above, this value is "1".
929 2. The second value is the startup time; this is a constant value that will always
930 be charged, no matter what the size of the message. In the first section above,
933 3. The third value is the \a per-byte cost. That is, it is charged for every
934 byte of the message (incurring cost messageSize*cost_per_byte)
935 and hence accounts also for larger messages. In the first
936 section of the example above, this value is "2".
938 Now, SMPI always checks which section it should take for a given message; that is,
939 if a message of size 11 is sent with the configuration of the example above, only
940 the second section will be used, not the first, as the first value of the second
941 section is closer to the message size. Hence, a message of size 11 incurs the
942 following cost inside MPI_Send:
948 %As 5 is the startup cost and 1 is the cost per byte.
951 The order of sections can be arbitrary; they will be ordered internally.
953 \subsection options_model_smpi_or smpi/or: Inject constant times for receive operations
955 This configuration option works exactly as \a smpi/os, see Section \ref options_model_smpi_os.
956 Of course, \a smpi/or is used to account for MPI_Recv instead of MPI_Send.
958 \subsection options_model_smpi_test smpi/test: Inject constant times for calls to MPI_Test
960 \b Default value: 0.0001
962 By setting this option, you can control the amount of time a process sleeps
963 when MPI_Test() is called; this is important, because SimGrid normally only
964 advances the time while communication is happening and thus,
965 MPI_Test will not add to the time, resulting in a deadlock if used as a
972 MPI_Test(request, flag, status);
978 Internally, in order to speed up execution, we use a counter to keep track
979 on how often we already checked if the handle is now valid or not. Hence, we
980 actually use counter*SLEEP_TIME, that is, the time MPI_Test() causes the process
981 to sleep increases linearly with the number of previously failed testk.
984 \subsection options_model_smpi_use_shared_malloc smpi/use-shared-malloc: Factorize malloc()s
988 SMPI can use shared memory by calling shm_* functions; this might speed up the simulation.
989 This opens or creates a new POSIX shared memory object, kept in RAM, in /dev/shm.
991 If you want to disable this behavior, set the value to 0.
993 \subsection options_model_smpi_wtime smpi/wtime: Inject constant times for calls to MPI_Wtime
997 By setting this option, you can control the amount of time a process sleeps
998 when MPI_Wtime() is called; this is important, because SimGrid normally only
999 advances the time while communication is happening and thus,
1000 MPI_Wtime will not add to the time, resulting in a deadlock if used as a
1006 while(MPI_Wtime() < some_time_bound) {
1011 If the time is never advanced, this loop will clearly never end as MPI_Wtime()
1012 always returns the same value. Hence, pass a (small) value to the smpi/wtime
1013 option to force a call to MPI_Wtime to advance the time as well.
1016 \section options_generic Configuring other aspects of SimGrid
1018 \subsection options_generic_clean_atexit Cleanup before termination
1020 The C / C++ standard contains a function called \b [atexit](http://www.cplusplus.com/reference/cstdlib/atexit/).
1021 atexit registers callbacks, which are called just before the program terminates.
1023 By setting the configuration option clean-atexit to 1 (true), a callback
1024 is registered and will clean up some variables and terminate/cleanup the tracing.
1026 TODO: Add when this should be used.
1028 \subsection options_generic_path XML file inclusion path
1030 It is possible to specify a list of directories to search into for the
1031 \<include\> tag in XML files by using the \b path configuration
1032 item. To add several directory to the path, set the configuration
1033 item several times, as in \verbatim
1034 --cfg=path:toto --cfg=path:tutu
1037 \subsection options_generic_exit Behavior on Ctrl-C
1039 By default, when Ctrl-C is pressed, the status of all existing
1040 simulated processes is displayed before exiting the simulation. This is very useful to debug your
1041 code, but it can reveal troublesome in some cases (such as when the
1042 amount of processes becomes really big). This behavior is disabled
1043 when \b verbose-exit is set to 0 (it is to 1 by default).
1045 \subsection options_exception_cutpath Truncate local path from exception backtrace
1047 <b>This configuration option is an internal option and should normally not be used
1048 by the user.</b> It is used to remove the path from the backtrace
1049 shown when an exception is thrown; if we didn't remove this part, the tests
1050 testing the exception parts of simgrid would fail on most machines, as we are
1051 currently comparing output. Clearly, the path used on different machines are almost
1052 guaranteed to be different and hence, the output would
1053 mismatch, causing the test to fail.
1055 \section options_log Logging Configuration
1057 It can be done by using XBT. Go to \ref XBT_log for more details.
1059 \section options_perf Performance optimizations
1061 \subsection options_perf_context Context factory
1063 In order to achieve higher performance, you might want to use the raw
1064 context factory which avoids any system call when switching between
1065 tasks. If it is not possible you might use ucontext instead.
1067 \subsection options_perf_guard_size Disabling stack guard pages
1069 A stack guard page is usually used which prevents the stack from
1070 overflowing on other parts of the memory. However this might have a
1071 performance impact if a huge number of processes is created. The
1072 option \b contexts:guard-size is the number of stack guard pages
1073 used. By setting it to 0, no guard pages will be used: in this case,
1074 you should avoid using small stacks (\b stack-size) as the stack will
1075 silently overflow on other parts of the memory.
1077 \section options_index Index of all existing configuration options
1080 Almost all options are defined in <i>src/simgrid/sg_config.c</i>. You may
1081 want to check this file, too, but this index should be somewhat complete
1082 for the moment (May 2015).
1085 \b Please \b note: You can also pass the command-line option "--help" and
1086 "--help-cfg" to an executable that uses simgrid.
1088 - \c clean-atexit: \ref options_generic_clean_atexit
1090 - \c contexts/factory: \ref options_virt_factory
1091 - \c contexts/guard-size: \ref options_virt_parallel
1092 - \c contexts/nthreads: \ref options_virt_parallel
1093 - \c contexts/parallel_threshold: \ref options_virt_parallel
1094 - \c contexts/stack-size: \ref options_virt_stacksize
1095 - \c contexts/synchro: \ref options_virt_parallel
1097 - \c cpu/maxmin-selective-update: \ref options_model_optim
1098 - \c cpu/model: \ref options_model_select
1099 - \c cpu/optim: \ref options_model_optim
1101 - \c exception/cutpath: \ref options_exception_cutpath
1103 - \c host/model: \ref options_model_select
1105 - \c maxmin/precision: \ref options_model_precision
1107 - \c msg/debug-multiple-use: \ref options_msg_debug_multiple_use
1109 - \c model-check: \ref options_modelchecking
1110 - \c model-check/checkpoint: \ref options_modelchecking_steps
1111 - \c model-check/communications-determinism: \ref options_modelchecking_comm_determinism
1112 - \c model-check/dot-output: \ref options_modelchecking_dot_output
1113 - \c model-check/hash: \ref options_modelchecking_hash
1114 - \c model-check/property: \ref options_modelchecking_liveness
1115 - \c model-check/max-depth: \ref options_modelchecking_max_depth
1116 - \c model-check/record: \ref options_modelchecking_recordreplay
1117 - \c model-check/reduction: \ref options_modelchecking_reduction
1118 - \c model-check/replay: \ref options_modelchecking_recordreplay
1119 - \c model-check/send-determinism: \ref options_modelchecking_comm_determinism
1120 - \c model-check/sparse-checkpoint: \ref options_modelchecking_sparse_checkpoint
1121 - \c model-check/termination: \ref options_modelchecking_termination
1122 - \c model-check/timeout: \ref options_modelchecking_timeout
1123 - \c model-check/visited: \ref options_modelchecking_visited
1125 - \c network/bandwidth-factor: \ref options_model_network_coefs
1126 - \c network/coordinates: \ref options_model_network_coord
1127 - \c network/crosstraffic: \ref options_model_network_crosstraffic
1128 - \c network/latency-factor: \ref options_model_network_coefs
1129 - \c network/maxmin-selective-update: \ref options_model_optim
1130 - \c network/model: \ref options_model_select
1131 - \c network/optim: \ref options_model_optim
1132 - \c network/sender_gap: \ref options_model_network_sendergap
1133 - \c network/TCP-gamma: \ref options_model_network_gamma
1134 - \c network/weight-S: \ref options_model_network_coefs
1136 - \c ns3/TcpModel: \ref options_pls
1137 - \c path: \ref options_generic_path
1138 - \c plugin: \ref options_generic_plugin
1140 - \c surf/nthreads: \ref options_model_nthreads
1141 - \c surf/precision: \ref options_model_precision
1143 - \c <b>For collective operations of SMPI, please refer to Section \ref options_index_smpi_coll</b>
1144 - \c smpi/async-small-thresh: \ref options_model_network_asyncsend
1145 - \c smpi/bw-factor: \ref options_model_smpi_bw_factor
1146 - \c smpi/coll-selector: \ref options_model_smpi_collectives
1147 - \c smpi/comp-adjustment-file: \ref options_model_smpi_adj_file
1148 - \c smpi/cpu-threshold: \ref options_smpi_bench
1149 - \c smpi/display-timing: \ref options_smpi_timing
1150 - \c smpi/lat-factor: \ref options_model_smpi_lat_factor
1151 - \c smpi/IB-penalty-factors: \ref options_model_network_coefs
1152 - \c smpi/iprobe: \ref options_model_smpi_iprobe
1153 - \c smpi/ois: \ref options_model_smpi_ois
1154 - \c smpi/or: \ref options_model_smpi_or
1155 - \c smpi/os: \ref options_model_smpi_os
1156 - \c smpi/privatize-global-variables: \ref options_smpi_global
1157 - \c smpi/running-power: \ref options_smpi_bench
1158 - \c smpi/send-is-detached-thresh: \ref options_model_smpi_detached
1159 - \c smpi/simulate-computation: \ref options_smpi_bench
1160 - \c smpi/test: \ref options_model_smpi_test
1161 - \c smpi/use-shared-malloc: \ref options_model_smpi_use_shared_malloc
1162 - \c smpi/wtime: \ref options_model_smpi_wtime
1164 - \c <b>Tracing configuration options can be found in Section \ref tracing_tracing_options</b>.
1166 - \c storage/model: \ref options_storage_model
1167 - \c verbose-exit: \ref options_generic_exit
1169 - \c vm/model: \ref options_vm_model
1171 \subsection options_index_smpi_coll Index of SMPI collective algorithms options
1173 TODO: All available collective algorithms will be made available via the ``smpirun --help-coll`` command.