1 /*! @page options Configure SimGrid
6 @htmlinclude graphical-toc.svg
10 document.getElementById("Config").style="opacity:0.93999999;fill:#ff0000;fill-opacity:0.1;stroke:#000000;stroke-width:0.35277778;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1";
14 A number of options can be given at runtime to change the default
15 SimGrid behavior. For a complete list of all configuration options
16 accepted by the SimGrid version used in your simulator, simply pass
17 the --help configuration flag to your program. If some of the options
18 are not documented on this page, this is a bug that you should please
19 report so that we can fix it. Note that some of the options presented
20 here may not be available in your simulators, depending on the
21 @ref install_src_config "compile-time options" that you used.
25 @section options_using Passing configuration options to the simulators
27 There is several way to pass configuration options to the simulators.
28 The most common way is to use the @c --cfg command line argument. For
29 example, to set the item @c Item to the value @c Value, simply
30 type the following: @verbatim
31 my_simulator --cfg=Item:Value (other arguments)
34 Several @c `--cfg` command line arguments can naturally be used. If you
35 need to include spaces in the argument, don't forget to quote the
36 argument. You can even escape the included quotes (write @' for ' if
37 you have your argument between ').
39 Another solution is to use the @c @<config@> tag in the platform file. The
40 only restriction is that this tag must occure before the first
41 platform element (be it @c @<AS@>, @c @<cluster@>, @c @<peer@> or whatever).
42 The @c @<config@> tag takes an @c id attribute, but it is currently
43 ignored so you don't really need to pass it. The important par is that
44 within that tag, you can pass one or several @c @<prop@> tags to specify
45 the configuration to use. For example, setting @c Item to @c Value
46 can be done by adding the following to the beginning of your platform
50 <prop id="Item" value="Value"/>
54 A last solution is to pass your configuration directly using the C
55 interface. If you happen to use the MSG interface, this is very easy
56 with the simgrid::s4u::Engine::setConfig() or MSG_config() functions. If you do not use MSG, that's a bit
57 more complex, as you have to mess with the internal configuration set
58 directly as follows. Check the @ref XBT_config "relevant page" for
59 details on all the functions you can use in this context, @c
60 _sg_cfg_set being the only configuration set currently used in
64 #include <xbt/config.h>
66 int main(int argc, char *argv[]) {
69 /* Prefer MSG_config() if you use MSG!! */
70 xbt_cfg_set_parse("Item:Value");
76 @section options_index Index of all existing configuration options
79 The full list can be retrieved by passing "--help" and
80 "--help-cfg" to an executable that uses SimGrid.
82 - @c clean-atexit: @ref options_generic_clean_atexit
84 - @c contexts/factory: @ref options_virt_factory
85 - @c contexts/guard-size: @ref options_virt_guard_size
86 - @c contexts/nthreads: @ref options_virt_parallel
87 - @c contexts/parallel-threshold: @ref options_virt_parallel
88 - @c contexts/stack-size: @ref options_virt_stacksize
89 - @c contexts/synchro: @ref options_virt_parallel
91 - @c cpu/maxmin-selective-update: @ref options_model_optim
92 - @c cpu/model: @ref options_model_select
93 - @c cpu/optim: @ref options_model_optim
95 - @c exception/cutpath: @ref options_exception_cutpath
97 - @c host/model: @ref options_model_select
99 - @c maxmin/precision: @ref options_model_precision
100 - @c maxmin/concurrency-limit: @ref options_concurrency_limit
102 - @c msg/debug-multiple-use: @ref options_msg_debug_multiple_use
104 - @c model-check: @ref options_modelchecking
105 - @c model-check/checkpoint: @ref options_modelchecking_steps
106 - @c model-check/communications-determinism: @ref options_modelchecking_comm_determinism
107 - @c model-check/dot-output: @ref options_modelchecking_dot_output
108 - @c model-check/hash: @ref options_modelchecking_hash
109 - @c model-check/property: @ref options_modelchecking_liveness
110 - @c model-check/max-depth: @ref options_modelchecking_max_depth
111 - @c model-check/record: @ref options_modelchecking_recordreplay
112 - @c model-check/reduction: @ref options_modelchecking_reduction
113 - @c model-check/replay: @ref options_modelchecking_recordreplay
114 - @c model-check/send-determinism: @ref options_modelchecking_comm_determinism
115 - @c model-check/sparse-checkpoint: @ref options_modelchecking_sparse_checkpoint
116 - @c model-check/termination: @ref options_modelchecking_termination
117 - @c model-check/timeout: @ref options_modelchecking_timeout
118 - @c model-check/visited: @ref options_modelchecking_visited
120 - @c network/bandwidth-factor: @ref options_model_network_coefs
121 - @c network/crosstraffic: @ref options_model_network_crosstraffic
122 - @c network/latency-factor: @ref options_model_network_coefs
123 - @c network/maxmin-selective-update: @ref options_model_optim
124 - @c network/model: @ref options_model_select
125 - @c network/optim: @ref options_model_optim
126 - @c network/TCP-gamma: @ref options_model_network_gamma
127 - @c network/weight-S: @ref options_model_network_coefs
129 - @c ns3/TcpModel: @ref options_pls
130 - @c path: @ref options_generic_path
131 - @c plugin: @ref options_generic_plugin
133 - @c simix/breakpoint: @ref options_generic_breakpoint
135 - @c storage/max_file_descriptors: @ref option_model_storage_maxfd
137 - @c surf/precision: @ref options_model_precision
139 - <b>For collective operations of SMPI, please refer to Section @ref options_index_smpi_coll</b>
140 - @c smpi/async-small-thresh: @ref options_model_network_asyncsend
141 - @c smpi/bw-factor: @ref options_model_smpi_bw_factor
142 - @c smpi/coll-selector: @ref options_model_smpi_collectives
143 - @c smpi/comp-adjustment-file: @ref options_model_smpi_adj_file
144 - @c smpi/cpu-threshold: @ref options_smpi_bench
145 - @c smpi/display-timing: @ref options_smpi_timing
146 - @c smpi/grow-injected-times: @ref options_model_smpi_test
147 - @c smpi/host-speed: @ref options_smpi_bench
148 - @c smpi/IB-penalty-factors: @ref options_model_network_coefs
149 - @c smpi/iprobe: @ref options_model_smpi_iprobe
150 - @c smpi/iprobe-cpu-usage: @ref options_model_smpi_iprobe_cpu_usage
151 - @c smpi/init: @ref options_model_smpi_init
152 - @c smpi/keep-temps: @ref options_smpi_temps
153 - @c smpi/lat-factor: @ref options_model_smpi_lat_factor
154 - @c smpi/ois: @ref options_model_smpi_ois
155 - @c smpi/or: @ref options_model_smpi_or
156 - @c smpi/os: @ref options_model_smpi_os
157 - @c smpi/papi-events: @ref options_smpi_papi_events
158 - @c smpi/privatization: @ref options_smpi_privatization
159 - @c smpi/privatize-libs: @ref options_smpi_privatize_libs
160 - @c smpi/send-is-detached-thresh: @ref options_model_smpi_detached
161 - @c smpi/shared-malloc: @ref options_model_smpi_shared_malloc
162 - @c smpi/shared-malloc-hugepage: @ref options_model_smpi_shared_malloc
163 - @c smpi/simulate-computation: @ref options_smpi_bench
164 - @c smpi/test: @ref options_model_smpi_test
165 - @c smpi/wtime: @ref options_model_smpi_wtime
167 - <b>Tracing configuration options can be found in Section @ref tracing_tracing_options</b>.
169 - @c storage/model: @ref options_storage_model
170 - @c verbose-exit: @ref options_generic_exit
172 - @c vm/model: @ref options_vm_model
174 @subsection options_index_smpi_coll Index of SMPI collective algorithms options
176 TODO: All available collective algorithms will be made available via the ``smpirun --help-coll`` command.
178 @section options_model Configuring the platform models
180 @anchor options_storage_model
181 @anchor options_vm_model
182 @subsection options_model_select Selecting the platform models
184 SimGrid comes with several network, CPU and storage models built in, and you
185 can change the used model at runtime by changing the passed
186 configuration. The three main configuration items are given below.
187 For each of these items, passing the special @c help value gives
188 you a short description of all possible values. Also, @c --help-models
189 should provide information about all models for all existing resources.
190 - @b network/model: specify the used network model
191 - @b cpu/model: specify the used CPU model
192 - @b host/model: specify the used host model
193 - @b storage/model: specify the used storage model (there is currently only one such model - this option is hence only useful for future releases)
194 - @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)
196 As of writing, the following network models are accepted. Over
197 the time new models can be added, and some experimental models can be
198 removed; check the values on your simulators for an uptodate
199 information. Note that the CM02 model is described in the research report
200 <a href="ftp://ftp.ens-lyon.fr/pub/LIP/Rapports/RR/RR2002/RR2002-40.ps.gz">A
201 Network Model for Simulation of Grid Application</a> while LV08 is
203 <a href="http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf">Accuracy Study and Improvement of Network Simulation in the SimGrid Framework</a>.
205 - @b LV08 (default one): Realistic network analytic model
206 (slow-start modeled by multiplying latency by 13.01, bandwidth by
207 .97; bottleneck sharing uses a payload of S=20537 for evaluating RTT)
208 - @anchor options_model_select_network_constant @b Constant: Simplistic network model where all communication
209 take a constant time (one second). This model provides the lowest
210 realism, but is (marginally) faster.
211 - @b SMPI: Realistic network model specifically tailored for HPC
212 settings (accurate modeling of slow start with correction factors on
213 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). See also @ref
214 options_model_network_coefs "this section" for more info.
215 - @b IB: Realistic network model specifically tailored for HPC
216 settings with InfiniBand networks (accurate modeling contention
217 behavior, based on the model explained in
218 http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf).
219 See also @ref options_model_network_coefs "this section" for more info.
220 - @b CM02: Legacy network analytic model (Very similar to LV08, but
221 without corrective factors. The timings of small messages are thus
223 - @b Reno: Model from Steven H. Low using lagrange_solve instead of
224 lmm_solve (experts only; check the code for more info).
225 - @b Reno2: Model from Steven H. Low using lagrange_solve instead of
226 lmm_solve (experts only; check the code for more info).
227 - @b Vegas: Model from Steven H. Low using lagrange_solve instead of
228 lmm_solve (experts only; check the code for more info).
230 If you compiled SimGrid accordingly, you can use packet-level network
231 simulators as network models (see @ref pls_ns3). In that case, you have
232 two extra models, described below, and some
233 @ref options_pls "specific additional configuration flags".
234 - @b NS3: Network pseudo-model using the NS3 tcp model
236 Concerning the CPU, we have only one model for now:
237 - @b Cas01: Simplistic CPU model (time=size/power)
239 The host concept is the aggregation of a CPU with a network
240 card. Three models exists, but actually, only 2 of them are
241 interesting. The "compound" one is simply due to the way our internal
242 code is organized, and can easily be ignored. So at the end, you have
243 two host models: The default one allows to aggregate an
244 existing CPU model with an existing network model, but does not allow
245 parallel tasks because these beasts need some collaboration between
246 the network and CPU model. That is why, ptask_07 is used by default
248 - @b default: Default host model. Currently, CPU:Cas01 and
249 network:LV08 (with cross traffic enabled)
250 - @b compound: Host model that is automatically chosen if
251 you change the network and CPU models
252 - @b ptask_L07: Host model somehow similar to Cas01+CM02 but
253 allowing "parallel tasks", that are intended to model the moldable
254 tasks of the grid scheduling literature.
256 @subsection options_generic_plugin Plugins
258 SimGrid plugins allow to extend the framework without changing its
259 source code directly. Read the source code of the existing plugins to
260 learn how to do so (in ``src/plugins``), and ask your questions to the
261 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
262 that plugins usually register callbacks to some signals of interest.
263 If they need to store some information about a given object (Link, CPU
264 or Actor), they do so through the use of a dedicated object extension.
266 Some of the existing plugins can be activated from the command line,
267 meaning that you can activate them from the command line without any
268 modification to your simulation code. For example, you can activate
269 the host energy plugin by adding the following to your command line:
272 --cfg=plugin:host_energy
275 Here is the full list of plugins that can be activated this way:
277 - @b host_energy: keeps track of the energy dissipated by
278 computations. More details in @ref plugin_energy.
279 - @b link_energy: keeps track of the energy dissipated by
280 communications. More details in @ref SURF_plugin_energy.
281 - @b host_load: keeps track of the computational load.
282 More details in @ref plugin_load.
284 @subsection options_model_optim Optimization level of the platform models
286 The network and CPU models that are based on lmm_solve (that
287 is, all our analytical models) accept specific optimization
289 - items @b network/optim and @b cpu/optim (both default to 'Lazy'):
290 - @b Lazy: Lazy action management (partial invalidation in lmm +
291 heap in action remaining).
292 - @b TI: Trace integration. Highly optimized mode when using
293 availability traces (only available for the Cas01 CPU model for
295 - @b Full: Full update of remaining and variables. Slow but may be
296 useful when debugging.
297 - items @b network/maxmin-selective-update and
298 @b cpu/maxmin-selective-update: configure whether the underlying
299 should be lazily updated or not. It should have no impact on the
300 computed timings, but should speed up the computation.
302 It is still possible to disable the @c maxmin-selective-update feature
303 because it can reveal counter-productive in very specific scenarios
304 where the interaction level is high. In particular, if all your
305 communication share a given backbone link, you should disable it:
306 without @c maxmin-selective-update, every communications are updated
307 at each step through a simple loop over them. With that feature
308 enabled, every communications will still get updated in this case
309 (because of the dependency induced by the backbone), but through a
310 complicated pattern aiming at following the actual dependencies.
312 @subsection options_model_precision Numerical precision of the platform models
314 The analytical models handle a lot of floating point values. It is
315 possible to change the epsilon used to update and compare them through
316 the @b maxmin/precision item (default value: 0.00001). Changing it
317 may speedup the simulation by discarding very small actions, at the
318 price of a reduced numerical precision.
320 @subsection options_concurrency_limit Concurrency limit
322 The maximum number of variables per resource can be tuned through
323 the @b maxmin/concurrency-limit item. The default value is -1, meaning that
324 there is no such limitation. You can have as many simultaneous actions per
325 resources as you want. If your simulation presents a very high level of
326 concurrency, it may help to use e.g. 100 as a value here. It means that at
327 most 100 actions can consume a resource at a given time. The extraneous actions
328 are queued and wait until the amount of concurrency of the considered resource
329 lowers under the given boundary.
331 Such limitations help both to the simulation speed and simulation accuracy
332 on highly constrained scenarios, but the simulation speed suffers of this
333 setting on regular (less constrained) scenarios so it is off by default.
335 @subsection options_model_network Configuring the Network model
337 @subsubsection options_model_network_gamma Maximal TCP window size
339 The analytical models need to know the maximal TCP window size to take
340 the TCP congestion mechanism into account. This is set to 4194304 by
341 default, but can be changed using the @b network/TCP-gamma item.
343 On linux, this value can be retrieved using the following
344 commands. Both give a set of values, and you should use the last one,
345 which is the maximal size.@verbatim
346 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
347 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
350 @subsubsection options_model_network_coefs Correcting important network parameters
352 SimGrid can take network irregularities such as a slow startup or
353 changing behavior depending on the message size into account.
354 You should not change these values unless you really know what you're doing.
356 The corresponding values were computed through data fitting one the
357 timings of packet-level simulators.
360 <a href="http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf">Accuracy Study and Improvement of Network Simulation in the SimGrid Framework</a>
361 for more information about these parameters.
363 If you are using the SMPI model, these correction coefficients are
364 themselves corrected by constant values depending on the size of the
365 exchange. Again, only hardcore experts should bother about this fact.
367 InfiniBand network behavior can be modeled through 3 parameters, as explained in
368 <a href="http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf">this PhD thesis</a>.
369 These factors can be changed through the following option:
372 smpi/IB-penalty-factors:"βe;βs;γs"
375 By default SMPI uses factors computed on the Stampede Supercomputer at TACC, with optimal
376 deployment of processes on nodes.
378 @subsubsection options_model_network_crosstraffic Simulating cross-traffic
380 As of SimGrid v3.7, cross-traffic effects can be taken into account in
381 analytical simulations. It means that ongoing and incoming
382 communication flows are treated independently. In addition, the LV08
383 model adds 0.05 of usage on the opposite direction for each new
384 created flow. This can be useful to simulate some important TCP
385 phenomena such as ack compression.
387 For that to work, your platform must have two links for each
388 pair of interconnected hosts. An example of usable platform is
389 available in <tt>examples/platforms/crosstraffic.xml</tt>.
391 This is activated through the @b network/crosstraffic item, that
392 can be set to 0 (disable this feature) or 1 (enable it).
394 Note that with the default host model this option is activated by default.
396 @subsubsection options_model_network_asyncsend Simulating asyncronous send
398 (this configuration item is experimental and may change or disapear)
400 It is possible to specify that messages below a certain size will be sent
401 as soon as the call to MPI_Send is issued, without waiting for the
402 correspondant receive. This threshold can be configured through the
403 @b smpi/async-small-thresh item. The default value is 0. This behavior can also be
404 manually set for MSG mailboxes, by setting the receiving mode of the mailbox
405 with a call to @ref MSG_mailbox_set_async . For MSG, all messages sent to this
406 mailbox will have this behavior, so consider using two mailboxes if needed.
408 This value needs to be smaller than or equals to the threshold set at
409 @ref options_model_smpi_detached , because asynchronous messages are
410 meant to be detached as well.
412 @subsubsection options_pls Configuring packet-level pseudo-models
414 When using the packet-level pseudo-models, several specific
415 configuration flags are provided to configure the associated tools.
416 There is by far not enough such SimGrid flags to cover every aspects
417 of the associated tools, since we only added the items that we
418 needed ourselves. Feel free to request more items (or even better:
419 provide patches adding more items).
421 When using NS3, the only existing item is @b ns3/TcpModel,
422 corresponding to the ns3::TcpL4Protocol::SocketType configuration item
423 in NS3. The only valid values (enforced on the SimGrid side) are
424 'NewReno' or 'Reno' or 'Tahoe'.
426 @subsection options_model_storage Configuring the Storage model
428 @subsubsection option_model_storage_maxfd Maximum amount of file descriptors per host
430 Each host maintains a fixed-size array of its file descriptors. You
431 can change its size (1024 by default) through the @b
432 storage/max_file_descriptors item to either enlarge it if your
433 application requires it or to reduce it to save memory space.
435 @section options_modelchecking Configuring the Model-Checking
437 To enable the SimGrid model-checking support the program should
438 be executed using the simgrid-mc wrapper:
440 simgrid-mc ./my_program
443 Safety properties are expressed as assertions using the function
445 void MC_assert(int prop);
448 @subsection options_modelchecking_liveness Specifying a liveness property
450 If you want to specify liveness properties (beware, that's
451 experimental), you have to pass them on the command line, specifying
452 the name of the file containing the property, as formatted by the
456 --cfg=model-check/property:<filename>
459 @subsection options_modelchecking_steps Going for stateful verification
461 By default, the system is backtracked to its initial state to explore
462 another path instead of backtracking to the exact step before the fork
463 that we want to explore (this is called stateless verification). This
464 is done this way because saving intermediate states can rapidly
465 exhaust the available memory. If you want, you can change the value of
466 the <tt>model-check/checkpoint</tt> variable. For example, the
467 following configuration will ask to take a checkpoint every step.
468 Beware, this will certainly explode your memory. Larger values are
469 probably better, make sure to experiment a bit to find the right
470 setting for your specific system.
473 --cfg=model-check/checkpoint:1
476 @subsection options_modelchecking_reduction Specifying the kind of reduction
478 The main issue when using the model-checking is the state space
479 explosion. To counter that problem, several exploration reduction
480 techniques can be used. There is unfortunately no silver bullet here,
481 and the most efficient reduction techniques cannot be applied to any
482 properties. In particular, the DPOR method cannot be applied on
483 liveness properties since it may break some cycles in the exploration
484 that are important to the property validity.
487 --cfg=model-check/reduction:<technique>
490 For now, this configuration variable can take 2 values:
491 * none: Do not apply any kind of reduction (mandatory for now for
493 * dpor: Apply Dynamic Partial Ordering Reduction. Only valid if you
494 verify local safety properties (default value for safety checks).
496 @subsection options_modelchecking_visited model-check/visited, Cycle detection
498 In order to detect cycles, the model-checker needs to check if a new explored
499 state is in fact the same state than a previous one. For that,
500 the model-checker can take a snapshot of each visited state: this snapshot is
501 then used to compare it with subsequent states in the exploration graph.
503 The @b model-check/visited option is the maximum number of states which are stored in
504 memory. If the maximum number of snapshotted state is reached, some states will
505 be removed from the memory and some cycles might be missed. Small
506 values can lead to incorrect verifications, but large value can
507 exhaust your memory, so choose carefully.
509 By default, no state is snapshotted and cycles cannot be detected.
511 @subsection options_modelchecking_termination model-check/termination, Non termination detection
513 The @b model-check/termination configuration item can be used to report if a
514 non-termination execution path has been found. This is a path with a cycle
515 which means that the program might never terminate.
517 This only works in safety mode.
519 This options is disabled by default.
521 @subsection options_modelchecking_dot_output model-check/dot-output, Dot output
523 If set, the @b model-check/dot-output configuration item is the name of a file
524 in which to write a dot file of the path leading the found property (safety or
525 liveness violation) as well as the cycle for liveness properties. This dot file
526 can then fed to the graphviz dot tool to generate an corresponding graphical
529 @subsection options_modelchecking_max_depth model-check/max-depth, Depth limit
531 The @b model-checker/max-depth can set the maximum depth of the exploration
532 graph of the model-checker. If this limit is reached, a logging message is
533 sent and the results might not be exact.
535 By default, there is not depth limit.
537 @subsection options_modelchecking_timeout Handling of timeout
539 By default, the model-checker does not handle timeout conditions: the `wait`
540 operations never time out. With the @b model-check/timeout configuration item
541 set to @b yes, the model-checker will explore timeouts of `wait` operations.
543 @subsection options_modelchecking_comm_determinism Communication determinism
545 The @b model-check/communications-determinism and
546 @b model-check/send-determinism items can be used to select the communication
547 determinism mode of the model-checker which checks determinism properties of
548 the communications of an application.
550 @subsection options_modelchecking_sparse_checkpoint Per page checkpoints
552 When the model-checker is configured to take a snapshot of each explored state
553 (with the @b model-checker/visited item), the memory consumption can rapidly
554 reach GiB ou Tib of memory. However, for many workloads, the memory does not
555 change much between different snapshots and taking a complete copy of each
556 snapshot is a waste of memory.
558 The @b model-check/sparse-checkpoint option item can be set to @b yes in order
559 to avoid making a complete copy of each snapshot: instead, each snapshot will be
560 decomposed in blocks which will be stored separately.
561 If multiple snapshots share the same block (or if the same block
562 is used in the same snapshot), the same copy of the block will be shared leading
563 to a reduction of the memory footprint.
565 For many applications, this option considerably reduces the memory consumption.
566 In somes cases, the model-checker might be slightly slower because of the time
567 taken to manage the metadata about the blocks. In other cases however, this
568 snapshotting strategy will be much faster by reducing the cache consumption.
569 When the memory consumption is important, by avoiding to hit the swap or
570 reducing the swap usage, this option might be much faster than the basic
571 snapshotting strategy.
573 This option is currently disabled by default.
575 @subsection options_mc_perf Performance considerations for the model checker
577 The size of the stacks can have a huge impact on the memory
578 consumption when using model-checking. By default, each snapshot will
579 save a copy of the whole stacks and not only of the part which is
580 really meaningful: you should expect the contribution of the memory
581 consumption of the snapshots to be @f$ @mbox{number of processes}
582 @times @mbox{stack size} @times @mbox{number of states} @f$.
584 The @b model-check/sparse-checkpoint can be used to reduce the memory
585 consumption by trying to share memory between the different snapshots.
587 When compiled against the model checker, the stacks are not
588 protected with guards: if the stack size is too small for your
589 application, the stack will silently overflow on other parts of the
590 memory (see @ref options_virt_guard_size).
592 @subsection options_modelchecking_hash Hashing of the state (experimental)
594 Usually most of the time of the model-checker is spent comparing states. This
595 process is complicated and consumes a lot of bandwidth and cache.
596 In order to speedup the state comparison, the experimental @b model-checker/hash
597 configuration item enables the computation of a hash summarizing as much
598 information of the state as possible into a single value. This hash can be used
599 to avoid most of the comparisons: the costly comparison is then only used when
600 the hashes are identical.
602 Currently most of the state is not included in the hash because the
603 implementation was found to be buggy and this options is not as useful as
604 it could be. For this reason, it is currently disabled by default.
606 @subsection options_modelchecking_recordreplay Record/replay (experimental)
608 As the model-checker keeps jumping at different places in the execution graph,
609 it is difficult to understand what happens when trying to debug an application
610 under the model-checker. Event the output of the program is difficult to
611 interpret. Moreover, the model-checker does not behave nicely with advanced
612 debugging tools such as valgrind. For those reason, to identify a trajectory
613 in the execution graph with the model-checker and replay this trajcetory and
614 without the model-checker black-magic but with more standard tools
615 (such as a debugger, valgrind, etc.). For this reason, Simgrid implements an
616 experimental record/replay functionnality in order to record a trajectory with
617 the model-checker and replay it without the model-checker.
619 When the model-checker finds an interesting path in the application execution
620 graph (where a safety or liveness property is violated), it can generate an
621 identifier for this path. In order to enable this behavious the
622 @b model-check/record must be set to @b yes. By default, this behaviour is not
625 This is an example of output:
628 [ 0.000000] (0:@) Check a safety property
629 [ 0.000000] (0:@) **************************
630 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
631 [ 0.000000] (0:@) **************************
632 [ 0.000000] (0:@) Counter-example execution trace:
633 [ 0.000000] (0:@) Path = 1/3;1/4
634 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
635 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
636 [ 0.000000] (0:@) Expanded states = 27
637 [ 0.000000] (0:@) Visited states = 68
638 [ 0.000000] (0:@) Executed transitions = 46
641 This path can then be replayed outside of the model-checker (and even in
642 non-MC build of simgrid) by setting the @b model-check/replay item to the given
643 path. The other options should be the same (but the model-checker should
646 The format and meaning of the path may change between different releases so
647 the same release of Simgrid should be used for the record phase and the replay
650 @section options_virt Configuring the User Process Virtualization
652 @subsection options_virt_factory Selecting the virtualization factory
654 In SimGrid, the user code is virtualized in a specific mechanism
655 that allows the simulation kernel to control its execution: when a user
656 process requires a blocking action (such as sending a message), it is
657 interrupted, and only gets released when the simulated clock reaches
658 the point where the blocking operation is done. This is explained
659 graphically in the [relevant tutorial, available online](http://simgrid.gforge.inria.fr/tutorials/simgrid-simix-101.pdf).
661 In SimGrid, the containers in which user processes are virtualized are
662 called contexts. Several context factory are provided, and you can
663 select the one you want to use with the @b contexts/factory
664 configuration item. Some of the following may not exist on your
665 machine because of portability issues. In any case, the default one
666 should be the most effcient one (please report bugs if the
667 auto-detection fails for you). They are approximately sorted here from
668 the slowest to the most efficient:
670 - @b thread: very slow factory using full featured threads (either
671 pthreads or windows native threads). They are slow but very
672 standard. Some debuggers or profilers only work with this factory.
673 - @b java: Java applications are virtualized onto java threads (that
674 are regular pthreads registered to the JVM)
675 - @b ucontext: fast factory using System V contexts (Linux and FreeBSD only)
676 - @b boost: This uses the [context implementation](http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html)
677 of the boost library for a performance that is comparable to our
678 raw implementation.@n Install the relevant library (e.g. with the
679 libboost-contexts-dev package on Debian/Ubuntu) and recompile
680 SimGrid. Note that our implementation is not compatible with recent
681 implementations of the library, and it will be hard to fix this since
682 the library's author decided to hide an API that we were using.
683 - @b raw: amazingly fast factory using a context switching mechanism
684 of our own, directly implemented in assembly (only available for x86
685 and amd64 platforms for now) and without any unneeded system call.
687 The main reason to change this setting is when the debugging tools get
688 fooled by the optimized context factories. Threads are the most
689 debugging-friendly contextes, as they allow to set breakpoints
690 anywhere with gdb and visualize backtraces for all processes, in order
691 to debug concurrency issues. Valgrind is also more comfortable with
692 threads, but it should be usable with all factories (but the callgrind
693 tool that really don't like raw and ucontext factories).
695 @subsection options_virt_stacksize Adapting the used stack size
697 Each virtualized used process is executed using a specific system
698 stack. The size of this stack has a huge impact on the simulation
699 scalability, but its default value is rather large. This is because
700 the error messages that you get when the stack size is too small are
701 rather disturbing: this leads to stack overflow (overwriting other
702 stacks), leading to segfaults with corrupted stack traces.
704 If you want to push the scalability limits of your code, you might
705 want to reduce the @b contexts/stack-size item. Its default value
706 is 8192 (in KiB), while our Chord simulation works with stacks as small
707 as 16 KiB, for example. For the thread factory, the default value
708 is the one of the system but you can still change it with this parameter.
710 The operating system should only allocate memory for the pages of the
711 stack which are actually used and you might not need to use this in
712 most cases. However, this setting is very important when using the
713 model checker (see @ref options_mc_perf).
715 @subsection options_virt_guard_size Disabling stack guard pages
717 A stack guard page is usually used which prevents the stack of a given
718 actor from overflowing on another stack. But the performance impact
719 may become prohibitive when the amount of actors increases. The
720 option @b contexts:guard-size is the number of stack guard pages used.
721 By setting it to 0, no guard pages will be used: in this case, you
722 should avoid using small stacks (@b stack-size) as the stack will
723 silently overflow on other parts of the memory.
725 When no stack guard page is created, stacks may then silently overflow
726 on other parts of the memory if their size is too small for the
727 application. This happens:
729 - on Windows systems;
730 - when the model checker is enabled;
731 - and of course when guard pages are explicitely disabled (with @b contexts:guard-size=0).
733 @subsection options_virt_parallel Running user code in parallel
735 Parallel execution of the user code is only considered stable in
736 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
737 simulations may well fail in parallel mode. It is described in
738 <a href="http://hal.inria.fr/inria-00602216/">INRIA RR-7653</a>.
740 If you are using the @c ucontext or @c raw context factories, you can
741 request to execute the user code in parallel. Several threads are
742 launched, each of them handling as much user contexts at each run. To
743 actiave this, set the @b contexts/nthreads item to the amount of
744 cores that you have in your computer (or lower than 1 to have
745 the amount of cores auto-detected).
747 Even if you asked several worker threads using the previous option,
748 you can request to start the parallel execution (and pay the
749 associated synchronization costs) only if the potential parallelism is
750 large enough. For that, set the @b contexts/parallel-threshold
751 item to the minimal amount of user contexts needed to start the
752 parallel execution. In any given simulation round, if that amount is
753 not reached, the contexts will be run sequentially directly by the
754 main thread (thus saving the synchronization costs). Note that this
755 option is mainly useful when the grain of the user code is very fine,
756 because our synchronization is now very efficient.
758 When parallel execution is activated, you can choose the
759 synchronization schema used with the @b contexts/synchro item,
760 which value is either:
761 - @b futex: ultra optimized synchronisation schema, based on futexes
762 (fast user-mode mutexes), and thus only available on Linux systems.
763 This is the default mode when available.
764 - @b posix: slow but portable synchronisation using only POSIX
766 - @b busy_wait: not really a synchronisation: the worker threads
767 constantly request new contexts to execute. It should be the most
768 efficient synchronisation schema, but it loads all the cores of your
769 machine for no good reason. You probably prefer the other less
772 @section options_tracing Configuring the tracing subsystem
774 The @ref outcomes_vizu "tracing subsystem" can be configured in several
775 different ways depending on the nature of the simulator (MSG, SimDag,
776 SMPI) and the kind of traces that need to be obtained. See the @ref
777 tracing_tracing_options "Tracing Configuration Options subsection" to
778 get a detailed description of each configuration option.
780 We detail here a simple way to get the traces working for you, even if
781 you never used the tracing API.
784 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
786 --cfg=tracing:yes --cfg=tracing/uncategorized:yes --cfg=triva/uncategorized:uncat.plist
788 The first parameter activates the tracing subsystem, the second
789 tells it to trace host and link utilization (without any
790 categorization) and the third creates a graph configuration file
791 to configure Triva when analysing the resulting trace file.
793 - MSG or SimDag-based simulator and categorized traces (you need to declare categories and classify your tasks according to them)
795 --cfg=tracing:yes --cfg=tracing/categorized:yes --cfg=triva/categorized:cat.plist
797 The first parameter activates the tracing subsystem, the second
798 tells it to trace host and link categorized utilization and the
799 third creates a graph configuration file to configure Triva when
800 analysing the resulting trace file.
802 - SMPI simulator and traces for a space/time view:
806 The <i>-trace</i> parameter for the smpirun script runs the
807 simulation with --cfg=tracing:yes and --cfg=tracing/smpi:yes. Check the
808 smpirun's <i>-help</i> parameter for additional tracing options.
810 Sometimes you might want to put additional information on the trace to
811 correctly identify them later, or to provide data that can be used to
812 reproduce an experiment. You have two ways to do that:
814 - Add a string on top of the trace file as comment:
816 --cfg=tracing/comment:my_simulation_identifier
819 - Add the contents of a textual file on top of the trace file as comment:
821 --cfg=tracing/comment-file:my_file_with_additional_information.txt
824 Please, use these two parameters (for comments) to make reproducible
825 simulations. For additional details about this and all tracing
826 options, check See the @ref tracing_tracing_options.
828 @section options_msg Configuring MSG
830 @subsection options_msg_debug_multiple_use Debugging MSG
832 Sometimes your application may try to send a task that is still being
833 executed somewhere else, making it impossible to send this task. However,
834 for debugging purposes, one may want to know what the other host is/was
835 doing. This option shows a backtrace of the other process.
837 Enable this option by adding
840 --cfg=msg/debug-multiple-use:on
843 @section options_smpi Configuring SMPI
845 The SMPI interface provides several specific configuration items.
846 These are uneasy to see since the code is usually launched through the
847 @c smiprun script directly.
849 @subsection options_smpi_bench smpi/bench: Automatic benchmarking of SMPI code
851 In SMPI, the sequential code is automatically benchmarked, and these
852 computations are automatically reported to the simulator. That is to
853 say that if you have a large computation between a @c MPI_Recv() and a
854 @c MPI_Send(), SMPI will automatically benchmark the duration of this
855 code, and create an execution task within the simulator to take this
856 into account. For that, the actual duration is measured on the host
857 machine and then scaled to the power of the corresponding simulated
858 machine. The variable @b smpi/host-speed allows to specify the
859 computational speed of the host machine (in flop/s) to use when
860 scaling the execution times. It defaults to 20000, but you really want
861 to update it to get accurate simulation results.
863 When the code is constituted of numerous consecutive MPI calls, the
864 previous mechanism feeds the simulation kernel with numerous tiny
865 computations. The @b smpi/cpu-threshold item becomes handy when this
866 impacts badly the simulation performance. It specifies a threshold (in
867 seconds) below which the execution chunks are not reported to the
868 simulation kernel (default value: 1e-6).
871 The option smpi/cpu-threshold ignores any computation time spent
872 below this threshold. SMPI does not consider the @a amount of these
873 computations; there is no offset for this. Hence, by using a
874 value that is too low, you may end up with unreliable simulation
877 In some cases, however, one may wish to disable simulation of
878 application computation. This is the case when SMPI is used not to
879 simulate an MPI applications, but instead an MPI code that performs
880 "live replay" of another MPI app (e.g., ScalaTrace's replay tool,
881 various on-line simulators that run an app at scale). In this case the
882 computation of the replay/simulation logic should not be simulated by
883 SMPI. Instead, the replay tool or on-line simulator will issue
884 "computation events", which correspond to the actual MPI simulation
885 being replayed/simulated. At the moment, these computation events can
886 be simulated using SMPI by calling internal smpi_execute*() functions.
888 To disable the benchmarking/simulation of computation in the simulated
889 application, the variable @b smpi/simulate-computation should be set to no.
892 This option just ignores the timings in your simulation; it still executes
893 the computations itself. If you want to stop SMPI from doing that,
894 you should check the SMPI_SAMPLE macros, documented in the section
895 @ref SMPI_adapting_speed.
897 Solution | Computations actually executed? | Computations simulated ?
898 ---------------------------------- | ------------------------------- | ------------------------
899 --cfg=smpi/simulate-computation:no | Yes | No, never
900 --cfg=smpi/cpu-threshold:42 | Yes, in all cases | Only if it lasts more than 42 seconds
901 SMPI_SAMPLE() macro | Only once per loop nest (see @ref SMPI_adapting_speed "documentation") | Always
903 @subsection options_model_smpi_adj_file smpi/comp-adjustment-file: Slow-down or speed-up parts of your code.
905 This option allows you to pass a file that contains two columns: The first column
906 defines the section that will be subject to a speedup; the second column is the speedup.
912 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
915 The first line is the header - you must include it.
916 The following line means that the code between two consecutive MPI calls on
917 line 30 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
918 of 1.18244559422142. The value for the second column is therefore a speedup, if it is
919 larger than 1 and a slow-down if it is smaller than 1. Nothing will be changed if it is
922 Of course, you can set any arbitrary filenames you want (so the start and end don't have to be
923 in the same file), but be aware that this mechanism only supports @em consecutive calls!
926 Please note that you must pass the @b -trace-call-location flag to smpicc
927 or smpiff, respectively! This flag activates some macro definitions in our
928 mpi.h / mpi.f files that help with obtaining the call location.
930 @subsection options_model_smpi_bw_factor smpi/bw-factor: Bandwidth factors
932 The possible throughput of network links is often dependent on the
933 message sizes, as protocols may adapt to different message sizes. With
934 this option, a series of message sizes and factors are given, helping
935 the simulation to be more realistic. For instance, the current
939 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
942 So, messages with size 65472 and more will get a total of MAX_BANDWIDTH*0.940694,
943 messages of size 15424 to 65471 will get MAX_BANDWIDTH*0.697866 and so on.
944 Here, MAX_BANDWIDTH denotes the bandwidth of the link.
947 The SimGrid-Team has developed a script to help you determine these
948 values. You can find more information and the download here:
949 1. http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
950 2. http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
952 @subsection options_smpi_timing smpi/display-timing: Reporting simulation time
954 @b Default: 0 (false)
956 Most of the time, you run MPI code with SMPI to compute the time it
957 would take to run it on a platform. But since the
958 code is run through the @c smpirun script, you don't have any control
959 on the launcher code, making it difficult to report the simulated time
960 when the simulation ends. If you set the @b smpi/display-timing item
961 to 1, @c smpirun will display this information when the simulation ends. @verbatim
962 Simulation time: 1e3 seconds.
965 @subsection options_smpi_temps smpi/keep-temps: not cleaning up after simulation
967 @b Default: 0 (false)
969 Under some conditions, SMPI generates a lot of temporary files. They
970 usually get cleaned, but you may use this option to not erase these
971 files. This is for example useful when debugging or profiling
972 executions using the dlopen privatization schema, as missing binary
973 files tend to fool the debuggers.
975 @subsection options_model_smpi_lat_factor smpi/lat-factor: Latency factors
977 The motivation and syntax for this option is identical to the motivation/syntax
978 of smpi/bw-factor, see @ref options_model_smpi_bw_factor for details.
980 There is an important difference, though: While smpi/bw-factor @a reduces the
981 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
982 increase the latency, i.e., values larger than or equal to 1 are valid here.
984 This is the default value:
987 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
991 The SimGrid-Team has developed a script to help you determine these
992 values. You can find more information and the download here:
993 1. http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
994 2. http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
996 @subsection options_smpi_papi_events smpi/papi-events: Trace hardware counters with PAPI
999 This option is experimental and will be subject to change.
1000 This feature currently requires superuser privileges, as registers are queried.
1001 Only use this feature with code you trust! Call smpirun for instance via
1002 smpirun -wrapper "sudo " <your-parameters>
1003 or run sudo sh -c "echo 0 > /proc/sys/kernel/perf_event_paranoid"
1004 In the later case, sudo will not be required.
1007 This option is only available when SimGrid was compiled with PAPI support.
1009 This option takes the names of PAPI counters and adds their respective values
1010 to the trace files. (See Section @ref tracing_tracing_options.)
1012 It is planned to make this feature available on a per-process (or per-thread?) basis.
1013 The first draft, however, just implements a "global" (i.e., for all processes) set
1014 of counters, the "default" set.
1017 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1020 @subsection options_smpi_privatization smpi/privatization: Automatic privatization of global variables
1022 MPI executables are usually meant to be executed in separated
1023 processes, but SMPI is executed in only one process. Global variables
1024 from executables will be placed in the same memory zone and shared
1025 between processes, causing intricate bugs. Several options are
1026 possible to avoid this, as described in the main
1027 <a href="https://hal.inria.fr/hal-01415484">SMPI publication</a> and in
1028 the @ref SMPI_what_globals "SMPI documentation". SimGrid provides two
1029 ways of automatically privatizing the globals, and this option allows
1030 to choose between them.
1032 - <b>no</b> (default when not using smpirun): Do not automatically privatize variables.
1033 Pass @c -no-privatize to smpirun to disable this feature.
1034 - <b>dlopen</b> or <b>yes</b> (default when using smpirun): Link multiple times against the binary.
1035 - <b>mmap</b> (slower, but maybe somewhat more stable):
1036 Runtime automatic switching of the data segments.
1039 This configuration option cannot be set in your platform file. You can only
1040 pass it as an argument to smpirun.
1042 @subsection options_smpi_privatize_libs smpi/privatize-libs: Automatic privatization of
1043 global variables inside external libraries
1045 Linux/BSD only: When using dlopen (default) privatization, privatize specific
1046 shared libraries with internal global variables, if they can't be linked statically.
1047 For example libgfortran is usually used for Fortran I/O and indexes in files
1051 This configuration option can only use either full paths to libraries, or full names.
1052 Check with ldd the name of the library you want to use.
1055 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1056 Then you can use --cfg=smpi/privatize-libs:"libgfortran.so.3" or --cfg=smpi/privatize-libs:"/usr/lib/x86_64-linux-gnu/libgfortran.so.3", but not "libgfortran" or "libgfortran.so".
1057 Multiple libraries can be given, semicolon separated.
1060 @subsection options_model_smpi_detached Simulating MPI detached send
1062 This threshold specifies the size in bytes under which the send will return
1063 immediately. This is different from the threshold detailed in @ref options_model_network_asyncsend
1064 because the message is not effectively sent when the send is posted. SMPI still waits for the
1065 correspondant receive to be posted to perform the communication operation. This threshold can be set
1066 by changing the @b smpi/send-is-detached-thresh item. The default value is 65536.
1068 @subsection options_model_smpi_collectives Simulating MPI collective algorithms
1070 SMPI implements more than 100 different algorithms for MPI collective communication, to accurately
1071 simulate the behavior of most of the existing MPI libraries. The @b smpi/coll-selector item can be used
1072 to use the decision logic of either OpenMPI or MPICH libraries (values: ompi or mpich, by default SMPI
1073 uses naive version of collective operations). Each collective operation can be manually selected with a
1074 @b smpi/collective_name:algo_name. Available algorithms are listed in @ref SMPI_use_colls .
1076 @subsection options_model_smpi_iprobe smpi/iprobe: Inject constant times for calls to MPI_Iprobe
1078 @b Default value: 0.0001
1080 The behavior and motivation for this configuration option is identical with @a smpi/test, see
1081 Section @ref options_model_smpi_test for details.
1083 @subsection options_model_smpi_iprobe_cpu_usage smpi/iprobe-cpu-usage: Reduce speed for iprobe calls
1085 @b Default value: 1 (no change from default behavior)
1087 MPI_Iprobe calls can be heavily used in applications. To account correctly for the energy
1088 cores spend probing, it is necessary to reduce the load that these calls cause inside
1091 For instance, we measured a max power consumption of 220 W for a particular application but
1092 only 180 W while this application was probing. Hence, the correct factor that should
1093 be passed to this option would be 180/220 = 0.81.
1095 @subsection options_model_smpi_init smpi/init: Inject constant times for calls to MPI_Init
1099 The behavior for this configuration option is identical with @a smpi/test, see
1100 Section @ref options_model_smpi_test for details.
1102 @subsection options_model_smpi_ois smpi/ois: Inject constant times for asynchronous send operations
1104 This configuration option works exactly as @a smpi/os, see Section @ref options_model_smpi_os.
1105 Of course, @a smpi/ois is used to account for MPI_Isend instead of MPI_Send.
1107 @subsection options_model_smpi_os smpi/os: Inject constant times for send operations
1109 In several network models such as LogP, send (MPI_Send, MPI_Isend) and receive (MPI_Recv)
1110 operations incur costs (i.e., they consume CPU time). SMPI can factor these costs in as well, but the
1111 user has to configure SMPI accordingly as these values may vary by machine.
1112 This can be done by using smpi/os for MPI_Send operations; for MPI_Isend and
1113 MPI_Recv, use @a smpi/ois and @a smpi/or, respectively. These work exactly as
1116 @a smpi/os can consist of multiple sections; each section takes three values, for example:
1122 Here, the sections are divided by ";" (that is, this example contains two sections).
1123 Furthermore, each section consists of three values.
1125 1. The first value denotes the minimum size for this section to take effect;
1126 read it as "if message size is greater than this value (and other section has a larger
1127 first value that is also smaller than the message size), use this".
1128 In the first section above, this value is "1".
1130 2. The second value is the startup time; this is a constant value that will always
1131 be charged, no matter what the size of the message. In the first section above,
1134 3. The third value is the @a per-byte cost. That is, it is charged for every
1135 byte of the message (incurring cost messageSize*cost_per_byte)
1136 and hence accounts also for larger messages. In the first
1137 section of the example above, this value is "2".
1139 Now, SMPI always checks which section it should take for a given message; that is,
1140 if a message of size 11 is sent with the configuration of the example above, only
1141 the second section will be used, not the first, as the first value of the second
1142 section is closer to the message size. Hence, a message of size 11 incurs the
1143 following cost inside MPI_Send:
1149 As 5 is the startup cost and 1 is the cost per byte.
1152 The order of sections can be arbitrary; they will be ordered internally.
1154 @subsection options_model_smpi_or smpi/or: Inject constant times for receive operations
1156 This configuration option works exactly as @a smpi/os, see Section @ref options_model_smpi_os.
1157 Of course, @a smpi/or is used to account for MPI_Recv instead of MPI_Send.
1159 @subsection options_model_smpi_test smpi/test: Inject constant times for calls to MPI_Test
1161 @b Default value: 0.0001
1163 By setting this option, you can control the amount of time a process sleeps
1164 when MPI_Test() is called; this is important, because SimGrid normally only
1165 advances the time while communication is happening and thus,
1166 MPI_Test will not add to the time, resulting in a deadlock if used as a
1173 MPI_Test(request, flag, status);
1179 Internally, in order to speed up execution, we use a counter to keep track
1180 on how often we already checked if the handle is now valid or not. Hence, we
1181 actually use counter*SLEEP_TIME, that is, the time MPI_Test() causes the process
1182 to sleep increases linearly with the number of previously failed tests. This
1183 behavior can be disabled by setting smpi/grow-injected-times to no. This will
1184 also disable this behavior for MPI_Iprobe.
1187 @subsection options_model_smpi_shared_malloc smpi/shared-malloc: Factorize malloc()s
1191 If your simulation consumes too much memory, you may want to modify
1192 your code so that the working areas are shared by all MPI ranks. For
1193 example, in a bloc-cyclic matrix multiplication, you will only
1194 allocate one set of blocs, and every processes will share them.
1195 Naturally, this will lead to very wrong results, but this will save a
1196 lot of memory so this is still desirable for some studies. For more on
1197 the motivation for that feature, please refer to the
1198 <a href="https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication/">relevant
1199 section</a> of the SMPI CourseWare (see Activity #2.2 of the pointed
1200 assignment). In practice, change the call to malloc() and free() into
1201 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1203 SMPI provides 2 algorithms for this feature. The first one, called @c
1204 local, allocates one bloc per call to SMPI_SHARED_MALLOC() in your
1205 code (each call location gets its own bloc) and this bloc is shared
1206 amongst all MPI ranks. This is implemented with the shm_* functions
1207 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1208 for each shared bloc.
1210 With the @c global algorithm, each call to SMPI_SHARED_MALLOC()
1211 returns a new adress, but it only points to a shadow bloc: its memory
1212 area is mapped on a 1MiB file on disk. If the returned bloc is of size
1213 N MiB, then the same file is mapped N times to cover the whole bloc.
1214 At the end, no matter how many SMPI_SHARED_MALLOC you do, this will
1215 only consume 1 MiB in memory.
1217 You can disable this behavior and come back to regular mallocs (for
1218 example for debugging purposes) using @c "no" as a value.
1220 If you want to keep private some parts of the buffer, for instance if these
1221 parts are used by the application logic and should not be corrupted, you
1222 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count).
1227 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1230 will allocate 500 bytes to mem, such that mem[27..41] and mem[100..199]
1231 are shared and other area remain private.
1233 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1235 When smpi/shared-malloc:global is used, the memory consumption problem
1236 is solved, but it may induce too much load on the kernel's pages table.
1237 In this case, you should use huge pages so that we create only one
1238 entry per Mb of malloced data instead of one entry per 4k.
1239 To activate this, you must mount a hugetlbfs on your system and allocate
1240 at least one huge page:
1244 sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1245 sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1248 Then, you can pass the option --cfg=smpi/shared-malloc-hugepage:/home/huge
1249 to smpirun to actually activate the huge page support in shared mallocs.
1251 @subsection options_model_smpi_wtime smpi/wtime: Inject constant times for calls to MPI_Wtime
1255 By setting this option, you can control the amount of time a process sleeps
1256 when MPI_Wtime() is called; this is important, because SimGrid normally only
1257 advances the time while communication is happening and thus,
1258 MPI_Wtime will not add to the time, resulting in a deadlock if used as a
1264 while(MPI_Wtime() < some_time_bound) {
1269 If the time is never advanced, this loop will clearly never end as MPI_Wtime()
1270 always returns the same value. Hence, pass a (small) value to the smpi/wtime
1271 option to force a call to MPI_Wtime to advance the time as well.
1274 @section options_generic Configuring other aspects of SimGrid
1276 @subsection options_generic_clean_atexit Cleanup before termination
1278 The C / C++ standard contains a function called @b [atexit](http://www.cplusplus.com/reference/cstdlib/atexit/).
1279 atexit registers callbacks, which are called just before the program terminates.
1281 By setting the configuration option clean-atexit to 1 (true), a callback
1282 is registered and will clean up some variables and terminate/cleanup the tracing.
1284 TODO: Add when this should be used.
1286 @subsection options_generic_path Profile files' search path
1288 It is possible to specify a list of directories to search into for the
1289 trace files (see @ref pf_trace) by using the @b path configuration
1290 item. To add several directory to the path, set the configuration
1291 item several times, as in @verbatim
1292 --cfg=path:toto --cfg=path:tutu
1295 @subsection options_generic_breakpoint Set a breakpoint
1298 --cfg=simix/breakpoint:3.1416
1301 This configuration option sets a breakpoint: when the simulated clock reaches
1302 the given time, a SIGTRAP is raised. This can be used to stop the execution and
1303 get a backtrace with a debugger.
1305 It is also possible to set the breakpoint from inside the debugger, by writing
1306 in global variable simgrid::simix::breakpoint. For example, with gdb:
1309 set variable simgrid::simix::breakpoint = 3.1416
1312 @subsection options_generic_exit Behavior on Ctrl-C
1314 By default, when Ctrl-C is pressed, the status of all existing
1315 simulated processes is displayed before exiting the simulation. This is very useful to debug your
1316 code, but it can reveal troublesome in some cases (such as when the
1317 amount of processes becomes really big). This behavior is disabled
1318 when @b verbose-exit is set to 0 (it is to 1 by default).
1320 @subsection options_exception_cutpath Truncate local path from exception backtrace
1323 --cfg=exception/cutpath:1
1326 This configuration option is used to remove the path from the
1327 backtrace shown when an exception is thrown. This is mainly useful for
1328 the tests: the full file path makes the tests not reproducible, and
1329 thus failing as we are currently comparing output. Clearly, the path
1330 used on different machines are almost guaranteed to be different and
1331 hence, the output would mismatch, causing the test to fail.
1333 @section options_log Logging Configuration
1335 It can be done by using XBT. Go to @ref XBT_log for more details.