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 storage/max_file_descriptors: \ref option_model_storage_maxfd
135 - \c surf/precision: \ref options_model_precision
137 - \c <b>For collective operations of SMPI, please refer to Section \ref options_index_smpi_coll</b>
138 - \c smpi/async-small-thresh: \ref options_model_network_asyncsend
139 - \c smpi/bw-factor: \ref options_model_smpi_bw_factor
140 - \c smpi/coll-selector: \ref options_model_smpi_collectives
141 - \c smpi/comp-adjustment-file: \ref options_model_smpi_adj_file
142 - \c smpi/cpu-threshold: \ref options_smpi_bench
143 - \c smpi/display-timing: \ref options_smpi_timing
144 - \c smpi/grow-injected-times: \ref options_model_smpi_test
145 - \c smpi/host-speed: \ref options_smpi_bench
146 - \c smpi/IB-penalty-factors: \ref options_model_network_coefs
147 - \c smpi/iprobe: \ref options_model_smpi_iprobe
148 - \c smpi/iprobe-cpu-usage: \ref options_model_smpi_iprobe_cpu_usage
149 - \c smpi/init: \ref options_model_smpi_init
150 - \c smpi/keep-temps: \ref options_smpi_temps
151 - \c smpi/lat-factor: \ref options_model_smpi_lat_factor
152 - \c smpi/ois: \ref options_model_smpi_ois
153 - \c smpi/or: \ref options_model_smpi_or
154 - \c smpi/os: \ref options_model_smpi_os
155 - \c smpi/papi-events: \ref options_smpi_papi_events
156 - \c smpi/privatization: \ref options_smpi_privatization
157 - \c smpi/send-is-detached-thresh: \ref options_model_smpi_detached
158 - \c smpi/shared-malloc: \ref options_model_smpi_shared_malloc
159 - \c smpi/shared-malloc-hugepage: \ref options_model_smpi_shared_malloc
160 - \c smpi/simulate-computation: \ref options_smpi_bench
161 - \c smpi/test: \ref options_model_smpi_test
162 - \c smpi/wtime: \ref options_model_smpi_wtime
164 - \c <b>Tracing configuration options can be found in Section \ref tracing_tracing_options</b>.
166 - \c storage/model: \ref options_storage_model
167 - \c verbose-exit: \ref options_generic_exit
169 - \c vm/model: \ref options_vm_model
171 \subsection options_index_smpi_coll Index of SMPI collective algorithms options
173 TODO: All available collective algorithms will be made available via the ``smpirun --help-coll`` command.
175 \section options_model Configuring the platform models
177 \anchor options_storage_model
178 \anchor options_vm_model
179 \subsection options_model_select Selecting the platform models
181 SimGrid comes with several network, CPU and storage models built in, and you
182 can change the used model at runtime by changing the passed
183 configuration. The three main configuration items are given below.
184 For each of these items, passing the special \c help value gives
185 you a short description of all possible values. Also, \c --help-models
186 should provide information about all models for all existing resources.
187 - \b network/model: specify the used network model
188 - \b cpu/model: specify the used CPU model
189 - \b host/model: specify the used host model
190 - \b storage/model: specify the used storage model (there is currently only one such model - this option is hence only useful for future releases)
191 - \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)
193 As of writing, the following network models are accepted. Over
194 the time new models can be added, and some experimental models can be
195 removed; check the values on your simulators for an uptodate
196 information. Note that the CM02 model is described in the research report
197 <a href="ftp://ftp.ens-lyon.fr/pub/LIP/Rapports/RR/RR2002/RR2002-40.ps.gz">A
198 Network Model for Simulation of Grid Application</a> while LV08 is
200 <a href="http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf">Accuracy Study and Improvement of Network Simulation in the SimGrid Framework</a>.
202 - \b LV08 (default one): Realistic network analytic model
203 (slow-start modeled by multiplying latency by 10.4, bandwidth by
204 .92; bottleneck sharing uses a payload of S=8775 for evaluating RTT)
205 - \anchor options_model_select_network_constant \b Constant: Simplistic network model where all communication
206 take a constant time (one second). This model provides the lowest
207 realism, but is (marginally) faster.
208 - \b SMPI: Realistic network model specifically tailored for HPC
209 settings (accurate modeling of slow start with correction factors on
210 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). See also \ref
211 options_model_network_coefs "this section" for more info.
212 - \b IB: Realistic network model specifically tailored for HPC
213 settings with InfiniBand networks (accurate modeling contention
214 behavior, based on the model explained in
215 http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf).
216 See also \ref options_model_network_coefs "this section" for more info.
217 - \b CM02: Legacy network analytic model (Very similar to LV08, but
218 without corrective factors. The timings of small messages are thus
220 - \b Reno: Model from Steven H. Low using lagrange_solve instead of
221 lmm_solve (experts only; check the code for more info).
222 - \b Reno2: Model from Steven H. Low using lagrange_solve instead of
223 lmm_solve (experts only; check the code for more info).
224 - \b Vegas: Model from Steven H. Low using lagrange_solve instead of
225 lmm_solve (experts only; check the code for more info).
227 If you compiled SimGrid accordingly, you can use packet-level network
228 simulators as network models (see \ref pls_ns3). In that case, you have
229 two extra models, described below, and some
230 \ref options_pls "specific additional configuration flags".
231 - \b NS3: Network pseudo-model using the NS3 tcp model
233 Concerning the CPU, we have only one model for now:
234 - \b Cas01: Simplistic CPU model (time=size/power)
236 The host concept is the aggregation of a CPU with a network
237 card. Three models exists, but actually, only 2 of them are
238 interesting. The "compound" one is simply due to the way our internal
239 code is organized, and can easily be ignored. So at the end, you have
240 two host models: The default one allows to aggregate an
241 existing CPU model with an existing network model, but does not allow
242 parallel tasks because these beasts need some collaboration between
243 the network and CPU model. That is why, ptask_07 is used by default
245 - \b default: Default host model. Currently, CPU:Cas01 and
246 network:LV08 (with cross traffic enabled)
247 - \b compound: Host model that is automatically chosen if
248 you change the network and CPU models
249 - \b ptask_L07: Host model somehow similar to Cas01+CM02 but
250 allowing "parallel tasks", that are intended to model the moldable
251 tasks of the grid scheduling literature.
253 \subsection options_generic_plugin Plugins
255 SimGrid supports the use of plugins; currently, no known plugins
256 can be activated but there are use-cases where you may want to write
257 your own plugin (for instance, for logging).
259 Plugins can for instance define own classes that inherit from
260 existing classes (for instance, a class "CpuEnergy" inherits from
261 "Cpu" to assess energy consumption).
263 The plugin connects to the code by registering callbacks using
264 ``signal.connect(callback)`` (see file ``src/surf/plugins/energy.cpp`` for
272 This option is case-sensitive: Energy and energy are not the same!
274 \subsection options_model_optim Optimization level of the platform models
276 The network and CPU models that are based on lmm_solve (that
277 is, all our analytical models) accept specific optimization
279 - items \b network/optim and \b cpu/optim (both default to 'Lazy'):
280 - \b Lazy: Lazy action management (partial invalidation in lmm +
281 heap in action remaining).
282 - \b TI: Trace integration. Highly optimized mode when using
283 availability traces (only available for the Cas01 CPU model for
285 - \b Full: Full update of remaining and variables. Slow but may be
286 useful when debugging.
287 - items \b network/maxmin-selective-update and
288 \b cpu/maxmin-selective-update: configure whether the underlying
289 should be lazily updated or not. It should have no impact on the
290 computed timings, but should speed up the computation.
292 It is still possible to disable the \c maxmin-selective-update feature
293 because it can reveal counter-productive in very specific scenarios
294 where the interaction level is high. In particular, if all your
295 communication share a given backbone link, you should disable it:
296 without \c maxmin-selective-update, every communications are updated
297 at each step through a simple loop over them. With that feature
298 enabled, every communications will still get updated in this case
299 (because of the dependency induced by the backbone), but through a
300 complicated pattern aiming at following the actual dependencies.
302 \subsection options_model_precision Numerical precision of the platform models
304 The analytical models handle a lot of floating point values. It is
305 possible to change the epsilon used to update and compare them through
306 the \b maxmin/precision item (default value: 0.00001). Changing it
307 may speedup the simulation by discarding very small actions, at the
308 price of a reduced numerical precision.
310 \subsection options_concurrency_limit Concurrency limit
312 The maximum number of variables per resource can be tuned through
313 the \b maxmin/concurrency-limit item. The default value is -1, meaning that
314 there is no such limitation. You can have as many simultaneous actions per
315 resources as you want. If your simulation presents a very high level of
316 concurrency, it may help to use e.g. 100 as a value here. It means that at
317 most 100 actions can consume a resource at a given time. The extraneous actions
318 are queued and wait until the amount of concurrency of the considered resource
319 lowers under the given boundary.
321 Such limitations help both to the simulation speed and simulation accuracy
322 on highly constrained scenarios, but the simulation speed suffers of this
323 setting on regular (less constrained) scenarios so it is off by default.
325 \subsection options_model_network Configuring the Network model
327 \subsubsection options_model_network_gamma Maximal TCP window size
329 The analytical models need to know the maximal TCP window size to take
330 the TCP congestion mechanism into account. This is set to 4194304 by
331 default, but can be changed using the \b network/TCP-gamma item.
333 On linux, this value can be retrieved using the following
334 commands. Both give a set of values, and you should use the last one,
335 which is the maximal size.\verbatim
336 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
337 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
340 \subsubsection options_model_network_coefs Correcting important network parameters
342 SimGrid can take network irregularities such as a slow startup or
343 changing behavior depending on the message size into account.
344 You should not change these values unless you really know what you're doing.
346 The corresponding values were computed through data fitting one the
347 timings of packet-level simulators.
350 <a href="http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf">Accuracy Study and Improvement of Network Simulation in the SimGrid Framework</a>
351 for more information about these parameters.
353 If you are using the SMPI model, these correction coefficients are
354 themselves corrected by constant values depending on the size of the
355 exchange. Again, only hardcore experts should bother about this fact.
357 InfiniBand network behavior can be modeled through 3 parameters, as explained in
358 <a href="http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf">this PhD thesis</a>.
359 These factors can be changed through the following option:
362 smpi/IB-penalty-factors:"βe;βs;γs"
365 By default SMPI uses factors computed on the Stampede Supercomputer at TACC, with optimal
366 deployment of processes on nodes.
368 \subsubsection options_model_network_crosstraffic Simulating cross-traffic
370 As of SimGrid v3.7, cross-traffic effects can be taken into account in
371 analytical simulations. It means that ongoing and incoming
372 communication flows are treated independently. In addition, the LV08
373 model adds 0.05 of usage on the opposite direction for each new
374 created flow. This can be useful to simulate some important TCP
375 phenomena such as ack compression.
377 For that to work, your platform must have two links for each
378 pair of interconnected hosts. An example of usable platform is
379 available in <tt>examples/platforms/crosstraffic.xml</tt>.
381 This is activated through the \b network/crosstraffic item, that
382 can be set to 0 (disable this feature) or 1 (enable it).
384 Note that with the default host model this option is activated by default.
386 \subsubsection options_model_network_asyncsend Simulating asyncronous send
388 (this configuration item is experimental and may change or disapear)
390 It is possible to specify that messages below a certain size will be sent
391 as soon as the call to MPI_Send is issued, without waiting for the
392 correspondant receive. This threshold can be configured through the
393 \b smpi/async-small-thresh item. The default value is 0. This behavior can also be
394 manually set for MSG mailboxes, by setting the receiving mode of the mailbox
395 with a call to \ref MSG_mailbox_set_async . For MSG, all messages sent to this
396 mailbox will have this behavior, so consider using two mailboxes if needed.
398 This value needs to be smaller than or equals to the threshold set at
399 \ref options_model_smpi_detached , because asynchronous messages are
400 meant to be detached as well.
402 \subsubsection options_pls Configuring packet-level pseudo-models
404 When using the packet-level pseudo-models, several specific
405 configuration flags are provided to configure the associated tools.
406 There is by far not enough such SimGrid flags to cover every aspects
407 of the associated tools, since we only added the items that we
408 needed ourselves. Feel free to request more items (or even better:
409 provide patches adding more items).
411 When using NS3, the only existing item is \b ns3/TcpModel,
412 corresponding to the ns3::TcpL4Protocol::SocketType configuration item
413 in NS3. The only valid values (enforced on the SimGrid side) are
414 'NewReno' or 'Reno' or 'Tahoe'.
416 \subsection options_model_storage Configuring the Storage model
418 \subsubsection option_model_storage_maxfd Maximum amount of file descriptors per host
420 Each host maintains a fixed-size array of its file descriptors. You
421 can change its size (1024 by default) through the \b
422 storage/max_file_descriptors item to either enlarge it if your
423 application requires it or to reduce it to save memory space.
425 \section options_modelchecking Configuring the Model-Checking
427 To enable the SimGrid model-checking support the program should
428 be executed using the simgrid-mc wrapper:
430 simgrid-mc ./my_program
433 Safety properties are expressed as assertions using the function
435 void MC_assert(int prop);
438 \subsection options_modelchecking_liveness Specifying a liveness property
440 If you want to specify liveness properties (beware, that's
441 experimental), you have to pass them on the command line, specifying
442 the name of the file containing the property, as formatted by the
446 --cfg=model-check/property:<filename>
449 \subsection options_modelchecking_steps Going for stateful verification
451 By default, the system is backtracked to its initial state to explore
452 another path instead of backtracking to the exact step before the fork
453 that we want to explore (this is called stateless verification). This
454 is done this way because saving intermediate states can rapidly
455 exhaust the available memory. If you want, you can change the value of
456 the <tt>model-check/checkpoint</tt> variable. For example, the
457 following configuration will ask to take a checkpoint every step.
458 Beware, this will certainly explode your memory. Larger values are
459 probably better, make sure to experiment a bit to find the right
460 setting for your specific system.
463 --cfg=model-check/checkpoint:1
466 \subsection options_modelchecking_reduction Specifying the kind of reduction
468 The main issue when using the model-checking is the state space
469 explosion. To counter that problem, several exploration reduction
470 techniques can be used. There is unfortunately no silver bullet here,
471 and the most efficient reduction techniques cannot be applied to any
472 properties. In particular, the DPOR method cannot be applied on
473 liveness properties since it may break some cycles in the exploration
474 that are important to the property validity.
477 --cfg=model-check/reduction:<technique>
480 For now, this configuration variable can take 2 values:
481 * none: Do not apply any kind of reduction (mandatory for now for
483 * dpor: Apply Dynamic Partial Ordering Reduction. Only valid if you
484 verify local safety properties (default value for safety checks).
486 \subsection options_modelchecking_visited model-check/visited, Cycle detection
488 In order to detect cycles, the model-checker needs to check if a new explored
489 state is in fact the same state than a previous one. For that,
490 the model-checker can take a snapshot of each visited state: this snapshot is
491 then used to compare it with subsequent states in the exploration graph.
493 The \b model-check/visited option is the maximum number of states which are stored in
494 memory. If the maximum number of snapshotted state is reached, some states will
495 be removed from the memory and some cycles might be missed. Small
496 values can lead to incorrect verifications, but large value can
497 exhaust your memory, so choose carefully.
499 By default, no state is snapshotted and cycles cannot be detected.
501 \subsection options_modelchecking_termination model-check/termination, Non termination detection
503 The \b model-check/termination configuration item can be used to report if a
504 non-termination execution path has been found. This is a path with a cycle
505 which means that the program might never terminate.
507 This only works in safety mode.
509 This options is disabled by default.
511 \subsection options_modelchecking_dot_output model-check/dot-output, Dot output
513 If set, the \b model-check/dot-output configuration item is the name of a file
514 in which to write a dot file of the path leading the found property (safety or
515 liveness violation) as well as the cycle for liveness properties. This dot file
516 can then fed to the graphviz dot tool to generate an corresponding graphical
519 \subsection options_modelchecking_max_depth model-check/max-depth, Depth limit
521 The \b model-checker/max-depth can set the maximum depth of the exploration
522 graph of the model-checker. If this limit is reached, a logging message is
523 sent and the results might not be exact.
525 By default, there is not depth limit.
527 \subsection options_modelchecking_timeout Handling of timeout
529 By default, the model-checker does not handle timeout conditions: the `wait`
530 operations never time out. With the \b model-check/timeout configuration item
531 set to \b yes, the model-checker will explore timeouts of `wait` operations.
533 \subsection options_modelchecking_comm_determinism Communication determinism
535 The \b model-check/communications-determinism and
536 \b model-check/send-determinism items can be used to select the communication
537 determinism mode of the model-checker which checks determinism properties of
538 the communications of an application.
540 \subsection options_modelchecking_sparse_checkpoint Per page checkpoints
542 When the model-checker is configured to take a snapshot of each explored state
543 (with the \b model-checker/visited item), the memory consumption can rapidly
544 reach GiB ou Tib of memory. However, for many workloads, the memory does not
545 change much between different snapshots and taking a complete copy of each
546 snapshot is a waste of memory.
548 The \b model-check/sparse-checkpoint option item can be set to \b yes in order
549 to avoid making a complete copy of each snapshot: instead, each snapshot will be
550 decomposed in blocks which will be stored separately.
551 If multiple snapshots share the same block (or if the same block
552 is used in the same snapshot), the same copy of the block will be shared leading
553 to a reduction of the memory footprint.
555 For many applications, this option considerably reduces the memory consumption.
556 In somes cases, the model-checker might be slightly slower because of the time
557 taken to manage the metadata about the blocks. In other cases however, this
558 snapshotting strategy will be much faster by reducing the cache consumption.
559 When the memory consumption is important, by avoiding to hit the swap or
560 reducing the swap usage, this option might be much faster than the basic
561 snapshotting strategy.
563 This option is currently disabled by default.
565 \subsection options_mc_perf Performance considerations for the model checker
567 The size of the stacks can have a huge impact on the memory
568 consumption when using model-checking. By default, each snapshot will
569 save a copy of the whole stacks and not only of the part which is
570 really meaningful: you should expect the contribution of the memory
571 consumption of the snapshots to be \f$ \mbox{number of processes}
572 \times \mbox{stack size} \times \mbox{number of states} \f$.
574 The \b model-check/sparse-checkpoint can be used to reduce the memory
575 consumption by trying to share memory between the different snapshots.
577 When compiled against the model checker, the stacks are not
578 protected with guards: if the stack size is too small for your
579 application, the stack will silently overflow on other parts of the
580 memory (see \ref options_virt_guard_size).
582 \subsection options_modelchecking_hash Hashing of the state (experimental)
584 Usually most of the time of the model-checker is spent comparing states. This
585 process is complicated and consumes a lot of bandwidth and cache.
586 In order to speedup the state comparison, the experimental \b model-checker/hash
587 configuration item enables the computation of a hash summarizing as much
588 information of the state as possible into a single value. This hash can be used
589 to avoid most of the comparisons: the costly comparison is then only used when
590 the hashes are identical.
592 Currently most of the state is not included in the hash because the
593 implementation was found to be buggy and this options is not as useful as
594 it could be. For this reason, it is currently disabled by default.
596 \subsection options_modelchecking_recordreplay Record/replay (experimental)
598 As the model-checker keeps jumping at different places in the execution graph,
599 it is difficult to understand what happens when trying to debug an application
600 under the model-checker. Event the output of the program is difficult to
601 interpret. Moreover, the model-checker does not behave nicely with advanced
602 debugging tools such as valgrind. For those reason, to identify a trajectory
603 in the execution graph with the model-checker and replay this trajcetory and
604 without the model-checker black-magic but with more standard tools
605 (such as a debugger, valgrind, etc.). For this reason, Simgrid implements an
606 experimental record/replay functionnality in order to record a trajectory with
607 the model-checker and replay it without the model-checker.
609 When the model-checker finds an interesting path in the application execution
610 graph (where a safety or liveness property is violated), it can generate an
611 identifier for this path. In order to enable this behavious the
612 \b model-check/record must be set to \b yes. By default, this behaviour is not
615 This is an example of output:
618 [ 0.000000] (0:@) Check a safety property
619 [ 0.000000] (0:@) **************************
620 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
621 [ 0.000000] (0:@) **************************
622 [ 0.000000] (0:@) Counter-example execution trace:
623 [ 0.000000] (0:@) Path = 1/3;1/4
624 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
625 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
626 [ 0.000000] (0:@) Expanded states = 27
627 [ 0.000000] (0:@) Visited states = 68
628 [ 0.000000] (0:@) Executed transitions = 46
631 This path can then be replayed outside of the model-checker (and even in
632 non-MC build of simgrid) by setting the \b model-check/replay item to the given
633 path. The other options should be the same (but the model-checker should
636 The format and meaning of the path may change between different releases so
637 the same release of Simgrid should be used for the record phase and the replay
640 \section options_virt Configuring the User Process Virtualization
642 \subsection options_virt_factory Selecting the virtualization factory
644 In SimGrid, the user code is virtualized in a specific mechanism
645 that allows the simulation kernel to control its execution: when a user
646 process requires a blocking action (such as sending a message), it is
647 interrupted, and only gets released when the simulated clock reaches
648 the point where the blocking operation is done. This is explained
649 graphically in the [relevant tutorial, available online](http://simgrid.gforge.inria.fr/tutorials/simgrid-simix-101.pdf).
651 In SimGrid, the containers in which user processes are virtualized are
652 called contexts. Several context factory are provided, and you can
653 select the one you want to use with the \b contexts/factory
654 configuration item. Some of the following may not exist on your
655 machine because of portability issues. In any case, the default one
656 should be the most effcient one (please report bugs if the
657 auto-detection fails for you). They are approximately sorted here from
658 the slowest to the most efficient:
660 - \b thread: very slow factory using full featured threads (either
661 pthreads or windows native threads). They are slow but very
662 standard. Some debuggers or profilers only work with this factory.
663 - \b java: Java applications are virtualized onto java threads (that
664 are regular pthreads registered to the JVM)
665 - \b ucontext: fast factory using System V contexts (Linux and FreeBSD only)
666 - \b boost: This uses the [context implementation](http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html)
667 of the boost library for a performance that is comparable to our
668 raw implementation.\n Install the relevant library (e.g. with the
669 libboost-contexts-dev package on Debian/Ubuntu) and recompile
670 SimGrid. Note that our implementation is not compatible with recent
671 implementations of the library, and it will be hard to fix this since
672 the library's author decided to hide an API that we were using.
673 - \b raw: amazingly fast factory using a context switching mechanism
674 of our own, directly implemented in assembly (only available for x86
675 and amd64 platforms for now) and without any unneeded system call.
677 The main reason to change this setting is when the debugging tools get
678 fooled by the optimized context factories. Threads are the most
679 debugging-friendly contextes, as they allow to set breakpoints
680 anywhere with gdb and visualize backtraces for all processes, in order
681 to debug concurrency issues. Valgrind is also more comfortable with
682 threads, but it should be usable with all factories (but the callgrind
683 tool that really don't like raw and ucontext factories).
685 \subsection options_virt_stacksize Adapting the used stack size
687 Each virtualized used process is executed using a specific system
688 stack. The size of this stack has a huge impact on the simulation
689 scalability, but its default value is rather large. This is because
690 the error messages that you get when the stack size is too small are
691 rather disturbing: this leads to stack overflow (overwriting other
692 stacks), leading to segfaults with corrupted stack traces.
694 If you want to push the scalability limits of your code, you might
695 want to reduce the \b contexts/stack-size item. Its default value
696 is 8192 (in KiB), while our Chord simulation works with stacks as small
697 as 16 KiB, for example. For the thread factory, the default value
698 is the one of the system but you can still change it with this parameter.
700 The operating system should only allocate memory for the pages of the
701 stack which are actually used and you might not need to use this in
702 most cases. However, this setting is very important when using the
703 model checker (see \ref options_mc_perf).
705 \subsection options_virt_guard_size Disabling stack guard pages
707 A stack guard page is usually used which prevents the stack of a given
708 actor from overflowing on another stack. But the performance impact
709 may become prohibitive when the amount of actors increases. The
710 option \b contexts:guard-size is the number of stack guard pages used.
711 By setting it to 0, no guard pages will be used: in this case, you
712 should avoid using small stacks (\b stack-size) as the stack will
713 silently overflow on other parts of the memory.
715 When no stack guard page is created, stacks may then silently overflow
716 on other parts of the memory if their size is too small for the
717 application. This happens:
719 - on Windows systems;
720 - when the model checker is enabled;
721 - and of course when guard pages are explicitely disabled (with \b contexts:guard-size=0).
723 \subsection options_virt_parallel Running user code in parallel
725 Parallel execution of the user code is only considered stable in
726 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
727 simulations may well fail in parallel mode. It is described in
728 <a href="http://hal.inria.fr/inria-00602216/">INRIA RR-7653</a>.
730 If you are using the \c ucontext or \c raw context factories, you can
731 request to execute the user code in parallel. Several threads are
732 launched, each of them handling as much user contexts at each run. To
733 actiave this, set the \b contexts/nthreads item to the amount of
734 cores that you have in your computer (or lower than 1 to have
735 the amount of cores auto-detected).
737 Even if you asked several worker threads using the previous option,
738 you can request to start the parallel execution (and pay the
739 associated synchronization costs) only if the potential parallelism is
740 large enough. For that, set the \b contexts/parallel-threshold
741 item to the minimal amount of user contexts needed to start the
742 parallel execution. In any given simulation round, if that amount is
743 not reached, the contexts will be run sequentially directly by the
744 main thread (thus saving the synchronization costs). Note that this
745 option is mainly useful when the grain of the user code is very fine,
746 because our synchronization is now very efficient.
748 When parallel execution is activated, you can choose the
749 synchronization schema used with the \b contexts/synchro item,
750 which value is either:
751 - \b futex: ultra optimized synchronisation schema, based on futexes
752 (fast user-mode mutexes), and thus only available on Linux systems.
753 This is the default mode when available.
754 - \b posix: slow but portable synchronisation using only POSIX
756 - \b busy_wait: not really a synchronisation: the worker threads
757 constantly request new contexts to execute. It should be the most
758 efficient synchronisation schema, but it loads all the cores of your
759 machine for no good reason. You probably prefer the other less
762 \section options_tracing Configuring the tracing subsystem
764 The \ref outcomes_vizu "tracing subsystem" can be configured in several
765 different ways depending on the nature of the simulator (MSG, SimDag,
766 SMPI) and the kind of traces that need to be obtained. See the \ref
767 tracing_tracing_options "Tracing Configuration Options subsection" to
768 get a detailed description of each configuration option.
770 We detail here a simple way to get the traces working for you, even if
771 you never used the tracing API.
774 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
776 --cfg=tracing:yes --cfg=tracing/uncategorized:yes --cfg=triva/uncategorized:uncat.plist
778 The first parameter activates the tracing subsystem, the second
779 tells it to trace host and link utilization (without any
780 categorization) and the third creates a graph configuration file
781 to configure Triva when analysing the resulting trace file.
783 - MSG or SimDag-based simulator and categorized traces (you need to declare categories and classify your tasks according to them)
785 --cfg=tracing:yes --cfg=tracing/categorized:yes --cfg=triva/categorized:cat.plist
787 The first parameter activates the tracing subsystem, the second
788 tells it to trace host and link categorized utilization and the
789 third creates a graph configuration file to configure Triva when
790 analysing the resulting trace file.
792 - SMPI simulator and traces for a space/time view:
796 The <i>-trace</i> parameter for the smpirun script runs the
797 simulation with --cfg=tracing:yes and --cfg=tracing/smpi:yes. Check the
798 smpirun's <i>-help</i> parameter for additional tracing options.
800 Sometimes you might want to put additional information on the trace to
801 correctly identify them later, or to provide data that can be used to
802 reproduce an experiment. You have two ways to do that:
804 - Add a string on top of the trace file as comment:
806 --cfg=tracing/comment:my_simulation_identifier
809 - Add the contents of a textual file on top of the trace file as comment:
811 --cfg=tracing/comment-file:my_file_with_additional_information.txt
814 Please, use these two parameters (for comments) to make reproducible
815 simulations. For additional details about this and all tracing
816 options, check See the \ref tracing_tracing_options.
818 \section options_msg Configuring MSG
820 \subsection options_msg_debug_multiple_use Debugging MSG
822 Sometimes your application may try to send a task that is still being
823 executed somewhere else, making it impossible to send this task. However,
824 for debugging purposes, one may want to know what the other host is/was
825 doing. This option shows a backtrace of the other process.
827 Enable this option by adding
830 --cfg=msg/debug-multiple-use:on
833 \section options_smpi Configuring SMPI
835 The SMPI interface provides several specific configuration items.
836 These are uneasy to see since the code is usually launched through the
837 \c smiprun script directly.
839 \subsection options_smpi_bench smpi/bench: Automatic benchmarking of SMPI code
841 In SMPI, the sequential code is automatically benchmarked, and these
842 computations are automatically reported to the simulator. That is to
843 say that if you have a large computation between a \c MPI_Recv() and a
844 \c MPI_Send(), SMPI will automatically benchmark the duration of this
845 code, and create an execution task within the simulator to take this
846 into account. For that, the actual duration is measured on the host
847 machine and then scaled to the power of the corresponding simulated
848 machine. The variable \b smpi/host-speed allows to specify the
849 computational speed of the host machine (in flop/s) to use when
850 scaling the execution times. It defaults to 20000, but you really want
851 to update it to get accurate simulation results.
853 When the code is constituted of numerous consecutive MPI calls, the
854 previous mechanism feeds the simulation kernel with numerous tiny
855 computations. The \b smpi/cpu-threshold item becomes handy when this
856 impacts badly the simulation performance. It specifies a threshold (in
857 seconds) below which the execution chunks are not reported to the
858 simulation kernel (default value: 1e-6).
861 The option smpi/cpu-threshold ignores any computation time spent
862 below this threshold. SMPI does not consider the \a amount of these
863 computations; there is no offset for this. Hence, by using a
864 value that is too low, you may end up with unreliable simulation
867 In some cases, however, one may wish to disable simulation of
868 application computation. This is the case when SMPI is used not to
869 simulate an MPI applications, but instead an MPI code that performs
870 "live replay" of another MPI app (e.g., ScalaTrace's replay tool,
871 various on-line simulators that run an app at scale). In this case the
872 computation of the replay/simulation logic should not be simulated by
873 SMPI. Instead, the replay tool or on-line simulator will issue
874 "computation events", which correspond to the actual MPI simulation
875 being replayed/simulated. At the moment, these computation events can
876 be simulated using SMPI by calling internal smpi_execute*() functions.
878 To disable the benchmarking/simulation of computation in the simulated
879 application, the variable \b smpi/simulate-computation should be set to no.
882 This option just ignores the timings in your simulation; it still executes
883 the computations itself. If you want to stop SMPI from doing that,
884 you should check the SMPI_SAMPLE macros, documented in the section
885 \ref SMPI_adapting_speed.
887 Solution | Computations actually executed? | Computations simulated ?
888 ---------------------------------- | ------------------------------- | ------------------------
889 --cfg=smpi/simulate-computation:no | Yes | No, never
890 --cfg=smpi/cpu-threshold:42 | Yes, in all cases | Only if it lasts more than 42 seconds
891 SMPI_SAMPLE() macro | Only once per loop nest (see @ref SMPI_adapting_speed "documentation") | Always
893 \subsection options_model_smpi_adj_file smpi/comp-adjustment-file: Slow-down or speed-up parts of your code.
895 This option allows you to pass a file that contains two columns: The first column
896 defines the section that will be subject to a speedup; the second column is the speedup.
902 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
905 The first line is the header - you must include it.
906 The following line means that the code between two consecutive MPI calls on
907 line 30 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
908 of 1.18244559422142. The value for the second column is therefore a speedup, if it is
909 larger than 1 and a slow-down if it is smaller than 1. Nothing will be changed if it is
912 Of course, you can set any arbitrary filenames you want (so the start and end don't have to be
913 in the same file), but be aware that this mechanism only supports @em consecutive calls!
916 Please note that you must pass the \b -trace-call-location flag to smpicc
917 or smpiff, respectively! This flag activates some macro definitions in our
918 mpi.h / mpi.f files that help with obtaining the call location.
920 \subsection options_model_smpi_bw_factor smpi/bw-factor: Bandwidth factors
922 The possible throughput of network links is often dependent on the
923 message sizes, as protocols may adapt to different message sizes. With
924 this option, a series of message sizes and factors are given, helping
925 the simulation to be more realistic. For instance, the current
929 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
932 So, messages with size 65472 and more will get a total of MAX_BANDWIDTH*0.940694,
933 messages of size 15424 to 65471 will get MAX_BANDWIDTH*0.697866 and so on.
934 Here, MAX_BANDWIDTH denotes the bandwidth of the link.
937 The SimGrid-Team has developed a script to help you determine these
938 values. You can find more information and the download here:
939 1. http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
940 2. http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
942 \subsection options_smpi_timing smpi/display-timing: Reporting simulation time
944 \b Default: 0 (false)
946 Most of the time, you run MPI code with SMPI to compute the time it
947 would take to run it on a platform. But since the
948 code is run through the \c smpirun script, you don't have any control
949 on the launcher code, making it difficult to report the simulated time
950 when the simulation ends. If you set the \b smpi/display-timing item
951 to 1, \c smpirun will display this information when the simulation ends. \verbatim
952 Simulation time: 1e3 seconds.
955 \subsection options_smpi_temps smpi/keep-temps: not cleaning up after simulation
957 \b Default: 0 (false)
959 Under some conditions, SMPI generates a lot of temporary files. They
960 usually get cleaned, but you may use this option to not erase these
961 files. This is for example useful when debugging or profiling
962 executions using the dlopen privatization schema, as missing binary
963 files tend to fool the debuggers.
965 \subsection options_model_smpi_lat_factor smpi/lat-factor: Latency factors
967 The motivation and syntax for this option is identical to the motivation/syntax
968 of smpi/bw-factor, see \ref options_model_smpi_bw_factor for details.
970 There is an important difference, though: While smpi/bw-factor \a reduces the
971 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
972 increase the latency, i.e., values larger than or equal to 1 are valid here.
974 This is the default value:
977 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
981 The SimGrid-Team has developed a script to help you determine these
982 values. You can find more information and the download here:
983 1. http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
984 2. http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
986 \subsection options_smpi_papi_events smpi/papi-events: Trace hardware counters with PAPI
989 This option is experimental and will be subject to change.
990 This feature currently requires superuser privileges, as registers are queried.
991 Only use this feature with code you trust! Call smpirun for instance via
992 smpirun -wrapper "sudo " <your-parameters>
993 or run sudo sh -c "echo 0 > /proc/sys/kernel/perf_event_paranoid"
994 In the later case, sudo will not be required.
997 This option is only available when SimGrid was compiled with PAPI support.
999 This option takes the names of PAPI counters and adds their respective values
1000 to the trace files. (See Section \ref tracing_tracing_options.)
1002 It is planned to make this feature available on a per-process (or per-thread?) basis.
1003 The first draft, however, just implements a "global" (i.e., for all processes) set
1004 of counters, the "default" set.
1007 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1010 \subsection options_smpi_privatization smpi/privatization: Automatic privatization of global variables
1012 MPI executables are usually meant to be executed in separated
1013 processes, but SMPI is executed in only one process. Global variables
1014 from executables will be placed in the same memory zone and shared
1015 between processes, causing intricate bugs. Several options are
1016 possible to avoid this, as described in the main
1017 <a href="https://hal.inria.fr/hal-01415484">SMPI publication</a> and in
1018 the @ref SMPI_what_globals "SMPI documentation". SimGrid provides two
1019 ways of automatically privatizing the globals, and this option allows
1020 to choose between them.
1022 - <b>no</b> (default when not using smpirun): Do not automatically privatize variables.
1023 Pass \c -no-privatize to smpirun to disable this feature.
1024 - <b>dlopen</b> or <b>yes</b> (default when using smpirun): Link multiple times against the binary.
1025 - <b>mmap</b> (slower, but maybe somewhat more stable):
1026 Runtime automatic switching of the data segments.
1029 This configuration option cannot be set in your platform file. You can only
1030 pass it as an argument to smpirun.
1032 \subsection options_model_smpi_detached Simulating MPI detached send
1034 This threshold specifies the size in bytes under which the send will return
1035 immediately. This is different from the threshold detailed in \ref options_model_network_asyncsend
1036 because the message is not effectively sent when the send is posted. SMPI still waits for the
1037 correspondant receive to be posted to perform the communication operation. This threshold can be set
1038 by changing the \b smpi/send-is-detached-thresh item. The default value is 65536.
1040 \subsection options_model_smpi_collectives Simulating MPI collective algorithms
1042 SMPI implements more than 100 different algorithms for MPI collective communication, to accurately
1043 simulate the behavior of most of the existing MPI libraries. The \b smpi/coll-selector item can be used
1044 to use the decision logic of either OpenMPI or MPICH libraries (values: ompi or mpich, by default SMPI
1045 uses naive version of collective operations). Each collective operation can be manually selected with a
1046 \b smpi/collective_name:algo_name. Available algorithms are listed in \ref SMPI_use_colls .
1048 \subsection options_model_smpi_iprobe smpi/iprobe: Inject constant times for calls to MPI_Iprobe
1050 \b Default value: 0.0001
1052 The behavior and motivation for this configuration option is identical with \a smpi/test, see
1053 Section \ref options_model_smpi_test for details.
1055 \subsection options_model_smpi_iprobe_cpu_usage smpi/iprobe-cpu-usage: Reduce speed for iprobe calls
1057 \b Default value: 1 (no change from default behavior)
1059 MPI_Iprobe calls can be heavily used in applications. To account correctly for the energy
1060 cores spend probing, it is necessary to reduce the load that these calls cause inside
1063 For instance, we measured a max power consumption of 220 W for a particular application but
1064 only 180 W while this application was probing. Hence, the correct factor that should
1065 be passed to this option would be 180/220 = 0.81.
1067 \subsection options_model_smpi_init smpi/init: Inject constant times for calls to MPI_Init
1071 The behavior for this configuration option is identical with \a smpi/test, see
1072 Section \ref options_model_smpi_test for details.
1074 \subsection options_model_smpi_ois smpi/ois: Inject constant times for asynchronous send operations
1076 This configuration option works exactly as \a smpi/os, see Section \ref options_model_smpi_os.
1077 Of course, \a smpi/ois is used to account for MPI_Isend instead of MPI_Send.
1079 \subsection options_model_smpi_os smpi/os: Inject constant times for send operations
1081 In several network models such as LogP, send (MPI_Send, MPI_Isend) and receive (MPI_Recv)
1082 operations incur costs (i.e., they consume CPU time). SMPI can factor these costs in as well, but the
1083 user has to configure SMPI accordingly as these values may vary by machine.
1084 This can be done by using smpi/os for MPI_Send operations; for MPI_Isend and
1085 MPI_Recv, use \a smpi/ois and \a smpi/or, respectively. These work exactly as
1088 \a smpi/os can consist of multiple sections; each section takes three values, for example:
1094 Here, the sections are divided by ";" (that is, this example contains two sections).
1095 Furthermore, each section consists of three values.
1097 1. The first value denotes the minimum size for this section to take effect;
1098 read it as "if message size is greater than this value (and other section has a larger
1099 first value that is also smaller than the message size), use this".
1100 In the first section above, this value is "1".
1102 2. The second value is the startup time; this is a constant value that will always
1103 be charged, no matter what the size of the message. In the first section above,
1106 3. The third value is the \a per-byte cost. That is, it is charged for every
1107 byte of the message (incurring cost messageSize*cost_per_byte)
1108 and hence accounts also for larger messages. In the first
1109 section of the example above, this value is "2".
1111 Now, SMPI always checks which section it should take for a given message; that is,
1112 if a message of size 11 is sent with the configuration of the example above, only
1113 the second section will be used, not the first, as the first value of the second
1114 section is closer to the message size. Hence, a message of size 11 incurs the
1115 following cost inside MPI_Send:
1121 As 5 is the startup cost and 1 is the cost per byte.
1124 The order of sections can be arbitrary; they will be ordered internally.
1126 \subsection options_model_smpi_or smpi/or: Inject constant times for receive operations
1128 This configuration option works exactly as \a smpi/os, see Section \ref options_model_smpi_os.
1129 Of course, \a smpi/or is used to account for MPI_Recv instead of MPI_Send.
1131 \subsection options_model_smpi_test smpi/test: Inject constant times for calls to MPI_Test
1133 \b Default value: 0.0001
1135 By setting this option, you can control the amount of time a process sleeps
1136 when MPI_Test() is called; this is important, because SimGrid normally only
1137 advances the time while communication is happening and thus,
1138 MPI_Test will not add to the time, resulting in a deadlock if used as a
1145 MPI_Test(request, flag, status);
1151 Internally, in order to speed up execution, we use a counter to keep track
1152 on how often we already checked if the handle is now valid or not. Hence, we
1153 actually use counter*SLEEP_TIME, that is, the time MPI_Test() causes the process
1154 to sleep increases linearly with the number of previously failed tests. This
1155 behavior can be disabled by setting smpi/grow-injected-times to no. This will
1156 also disable this behavior for MPI_Iprobe.
1159 \subsection options_model_smpi_shared_malloc smpi/shared-malloc: Factorize malloc()s
1163 If your simulation consumes too much memory, you may want to modify
1164 your code so that the working areas are shared by all MPI ranks. For
1165 example, in a bloc-cyclic matrix multiplication, you will only
1166 allocate one set of blocs, and every processes will share them.
1167 Naturally, this will lead to very wrong results, but this will save a
1168 lot of memory so this is still desirable for some studies. For more on
1169 the motivation for that feature, please refer to the
1170 <a href="https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication/">relevant
1171 section</a> of the SMPI CourseWare (see Activity #2.2 of the pointed
1172 assignment). In practice, change the call to malloc() and free() into
1173 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1175 SMPI provides 2 algorithms for this feature. The first one, called \c
1176 local, allocates one bloc per call to SMPI_SHARED_MALLOC() in your
1177 code (each call location gets its own bloc) and this bloc is shared
1178 amongst all MPI ranks. This is implemented with the shm_* functions
1179 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1180 for each shared bloc.
1182 With the \c global algorithm, each call to SMPI_SHARED_MALLOC()
1183 returns a new adress, but it only points to a shadow bloc: its memory
1184 area is mapped on a 1MiB file on disk. If the returned bloc is of size
1185 N MiB, then the same file is mapped N times to cover the whole bloc.
1186 At the end, no matter how many SMPI_SHARED_MALLOC you do, this will
1187 only consume 1 MiB in memory.
1189 You can disable this behavior and come back to regular mallocs (for
1190 example for debugging purposes) using \c "no" as a value.
1192 If you want to keep private some parts of the buffer, for instance if these
1193 parts are used by the application logic and should not be corrupted, you
1194 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count).
1199 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1202 will allocate 500 bytes to mem, such that mem[27..41] and mem[100..199]
1203 are shared and other area remain private.
1205 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1207 When smpi/shared-malloc:global is used, the memory consumption problem
1208 is solved, but it may induce too much load on the kernel's pages table.
1209 In this case, you should use huge pages so that we create only one
1210 entry per Mb of malloced data instead of one entry per 4k.
1211 To activate this, you must mount a hugetlbfs on your system and allocate
1212 at least one huge page:
1216 sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1217 sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1220 Then, you can pass the option --cfg=smpi/shared-malloc-hugepage:/home/huge
1221 to smpirun to actually activate the huge page support in shared mallocs.
1223 \subsection options_model_smpi_wtime smpi/wtime: Inject constant times for calls to MPI_Wtime
1227 By setting this option, you can control the amount of time a process sleeps
1228 when MPI_Wtime() is called; this is important, because SimGrid normally only
1229 advances the time while communication is happening and thus,
1230 MPI_Wtime will not add to the time, resulting in a deadlock if used as a
1236 while(MPI_Wtime() < some_time_bound) {
1241 If the time is never advanced, this loop will clearly never end as MPI_Wtime()
1242 always returns the same value. Hence, pass a (small) value to the smpi/wtime
1243 option to force a call to MPI_Wtime to advance the time as well.
1246 \section options_generic Configuring other aspects of SimGrid
1248 \subsection options_generic_clean_atexit Cleanup before termination
1250 The C / C++ standard contains a function called \b [atexit](http://www.cplusplus.com/reference/cstdlib/atexit/).
1251 atexit registers callbacks, which are called just before the program terminates.
1253 By setting the configuration option clean-atexit to 1 (true), a callback
1254 is registered and will clean up some variables and terminate/cleanup the tracing.
1256 TODO: Add when this should be used.
1258 \subsection options_generic_path Profile files' search path
1260 It is possible to specify a list of directories to search into for the
1261 trace files (see @ref pf_trace) by using the \b path configuration
1262 item. To add several directory to the path, set the configuration
1263 item several times, as in \verbatim
1264 --cfg=path:toto --cfg=path:tutu
1267 \subsection options_generic_exit Behavior on Ctrl-C
1269 By default, when Ctrl-C is pressed, the status of all existing
1270 simulated processes is displayed before exiting the simulation. This is very useful to debug your
1271 code, but it can reveal troublesome in some cases (such as when the
1272 amount of processes becomes really big). This behavior is disabled
1273 when \b verbose-exit is set to 0 (it is to 1 by default).
1275 \subsection options_exception_cutpath Truncate local path from exception backtrace
1278 --cfg=exception/cutpath:1
1281 This configuration option is used to remove the path from the
1282 backtrace shown when an exception is thrown. This is mainly useful for
1283 the tests: the full file path makes the tests not reproducible, and
1284 thus failing as we are currently comparing output. Clearly, the path
1285 used on different machines are almost guaranteed to be different and
1286 hence, the output would mismatch, causing the test to fail.
1288 \section options_log Logging Configuration
1290 It can be done by using XBT. Go to \ref XBT_log for more details.