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 - \c <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/send-is-detached-thresh: \ref options_model_smpi_detached
160 - \c smpi/shared-malloc: \ref options_model_smpi_shared_malloc
161 - \c smpi/shared-malloc-hugepage: \ref options_model_smpi_shared_malloc
162 - \c smpi/simulate-computation: \ref options_smpi_bench
163 - \c smpi/test: \ref options_model_smpi_test
164 - \c smpi/wtime: \ref options_model_smpi_wtime
166 - \c <b>Tracing configuration options can be found in Section \ref tracing_tracing_options</b>.
168 - \c storage/model: \ref options_storage_model
169 - \c verbose-exit: \ref options_generic_exit
171 - \c vm/model: \ref options_vm_model
173 \subsection options_index_smpi_coll Index of SMPI collective algorithms options
175 TODO: All available collective algorithms will be made available via the ``smpirun --help-coll`` command.
177 \section options_model Configuring the platform models
179 \anchor options_storage_model
180 \anchor options_vm_model
181 \subsection options_model_select Selecting the platform models
183 SimGrid comes with several network, CPU and storage models built in, and you
184 can change the used model at runtime by changing the passed
185 configuration. The three main configuration items are given below.
186 For each of these items, passing the special \c help value gives
187 you a short description of all possible values. Also, \c --help-models
188 should provide information about all models for all existing resources.
189 - \b network/model: specify the used network model
190 - \b cpu/model: specify the used CPU model
191 - \b host/model: specify the used host model
192 - \b storage/model: specify the used storage model (there is currently only one such model - this option is hence only useful for future releases)
193 - \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)
195 As of writing, the following network models are accepted. Over
196 the time new models can be added, and some experimental models can be
197 removed; check the values on your simulators for an uptodate
198 information. Note that the CM02 model is described in the research report
199 <a href="ftp://ftp.ens-lyon.fr/pub/LIP/Rapports/RR/RR2002/RR2002-40.ps.gz">A
200 Network Model for Simulation of Grid Application</a> while LV08 is
202 <a href="http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf">Accuracy Study and Improvement of Network Simulation in the SimGrid Framework</a>.
204 - \b LV08 (default one): Realistic network analytic model
205 (slow-start modeled by multiplying latency by 13.01, bandwidth by
206 .97; bottleneck sharing uses a payload of S=20537 for evaluating RTT)
207 - \anchor options_model_select_network_constant \b Constant: Simplistic network model where all communication
208 take a constant time (one second). This model provides the lowest
209 realism, but is (marginally) faster.
210 - \b SMPI: Realistic network model specifically tailored for HPC
211 settings (accurate modeling of slow start with correction factors on
212 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). See also \ref
213 options_model_network_coefs "this section" for more info.
214 - \b IB: Realistic network model specifically tailored for HPC
215 settings with InfiniBand networks (accurate modeling contention
216 behavior, based on the model explained in
217 http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf).
218 See also \ref options_model_network_coefs "this section" for more info.
219 - \b CM02: Legacy network analytic model (Very similar to LV08, but
220 without corrective factors. The timings of small messages are thus
222 - \b Reno: Model from Steven H. Low using lagrange_solve instead of
223 lmm_solve (experts only; check the code for more info).
224 - \b Reno2: Model from Steven H. Low using lagrange_solve instead of
225 lmm_solve (experts only; check the code for more info).
226 - \b Vegas: Model from Steven H. Low using lagrange_solve instead of
227 lmm_solve (experts only; check the code for more info).
229 If you compiled SimGrid accordingly, you can use packet-level network
230 simulators as network models (see \ref pls_ns3). In that case, you have
231 two extra models, described below, and some
232 \ref options_pls "specific additional configuration flags".
233 - \b NS3: Network pseudo-model using the NS3 tcp model
235 Concerning the CPU, we have only one model for now:
236 - \b Cas01: Simplistic CPU model (time=size/power)
238 The host concept is the aggregation of a CPU with a network
239 card. Three models exists, but actually, only 2 of them are
240 interesting. The "compound" one is simply due to the way our internal
241 code is organized, and can easily be ignored. So at the end, you have
242 two host models: The default one allows to aggregate an
243 existing CPU model with an existing network model, but does not allow
244 parallel tasks because these beasts need some collaboration between
245 the network and CPU model. That is why, ptask_07 is used by default
247 - \b default: Default host model. Currently, CPU:Cas01 and
248 network:LV08 (with cross traffic enabled)
249 - \b compound: Host model that is automatically chosen if
250 you change the network and CPU models
251 - \b ptask_L07: Host model somehow similar to Cas01+CM02 but
252 allowing "parallel tasks", that are intended to model the moldable
253 tasks of the grid scheduling literature.
255 \subsection options_generic_plugin Plugins
257 SimGrid supports the use of plugins; currently, no known plugins
258 can be activated but there are use-cases where you may want to write
259 your own plugin (for instance, for logging).
261 Plugins can for instance define own classes that inherit from
262 existing classes (for instance, a class "CpuEnergy" inherits from
263 "Cpu" to assess energy consumption).
265 The plugin connects to the code by registering callbacks using
266 ``signal.connect(callback)`` (see file ``src/surf/plugins/energy.cpp`` for
274 This option is case-sensitive: Energy and energy are not the same!
276 \subsection options_model_optim Optimization level of the platform models
278 The network and CPU models that are based on lmm_solve (that
279 is, all our analytical models) accept specific optimization
281 - items \b network/optim and \b cpu/optim (both default to 'Lazy'):
282 - \b Lazy: Lazy action management (partial invalidation in lmm +
283 heap in action remaining).
284 - \b TI: Trace integration. Highly optimized mode when using
285 availability traces (only available for the Cas01 CPU model for
287 - \b Full: Full update of remaining and variables. Slow but may be
288 useful when debugging.
289 - items \b network/maxmin-selective-update and
290 \b cpu/maxmin-selective-update: configure whether the underlying
291 should be lazily updated or not. It should have no impact on the
292 computed timings, but should speed up the computation.
294 It is still possible to disable the \c maxmin-selective-update feature
295 because it can reveal counter-productive in very specific scenarios
296 where the interaction level is high. In particular, if all your
297 communication share a given backbone link, you should disable it:
298 without \c maxmin-selective-update, every communications are updated
299 at each step through a simple loop over them. With that feature
300 enabled, every communications will still get updated in this case
301 (because of the dependency induced by the backbone), but through a
302 complicated pattern aiming at following the actual dependencies.
304 \subsection options_model_precision Numerical precision of the platform models
306 The analytical models handle a lot of floating point values. It is
307 possible to change the epsilon used to update and compare them through
308 the \b maxmin/precision item (default value: 0.00001). Changing it
309 may speedup the simulation by discarding very small actions, at the
310 price of a reduced numerical precision.
312 \subsection options_concurrency_limit Concurrency limit
314 The maximum number of variables per resource can be tuned through
315 the \b maxmin/concurrency-limit item. The default value is -1, meaning that
316 there is no such limitation. You can have as many simultaneous actions per
317 resources as you want. If your simulation presents a very high level of
318 concurrency, it may help to use e.g. 100 as a value here. It means that at
319 most 100 actions can consume a resource at a given time. The extraneous actions
320 are queued and wait until the amount of concurrency of the considered resource
321 lowers under the given boundary.
323 Such limitations help both to the simulation speed and simulation accuracy
324 on highly constrained scenarios, but the simulation speed suffers of this
325 setting on regular (less constrained) scenarios so it is off by default.
327 \subsection options_model_network Configuring the Network model
329 \subsubsection options_model_network_gamma Maximal TCP window size
331 The analytical models need to know the maximal TCP window size to take
332 the TCP congestion mechanism into account. This is set to 4194304 by
333 default, but can be changed using the \b network/TCP-gamma item.
335 On linux, this value can be retrieved using the following
336 commands. Both give a set of values, and you should use the last one,
337 which is the maximal size.\verbatim
338 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
339 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
342 \subsubsection options_model_network_coefs Correcting important network parameters
344 SimGrid can take network irregularities such as a slow startup or
345 changing behavior depending on the message size into account.
346 You should not change these values unless you really know what you're doing.
348 The corresponding values were computed through data fitting one the
349 timings of packet-level simulators.
352 <a href="http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf">Accuracy Study and Improvement of Network Simulation in the SimGrid Framework</a>
353 for more information about these parameters.
355 If you are using the SMPI model, these correction coefficients are
356 themselves corrected by constant values depending on the size of the
357 exchange. Again, only hardcore experts should bother about this fact.
359 InfiniBand network behavior can be modeled through 3 parameters, as explained in
360 <a href="http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf">this PhD thesis</a>.
361 These factors can be changed through the following option:
364 smpi/IB-penalty-factors:"βe;βs;γs"
367 By default SMPI uses factors computed on the Stampede Supercomputer at TACC, with optimal
368 deployment of processes on nodes.
370 \subsubsection options_model_network_crosstraffic Simulating cross-traffic
372 As of SimGrid v3.7, cross-traffic effects can be taken into account in
373 analytical simulations. It means that ongoing and incoming
374 communication flows are treated independently. In addition, the LV08
375 model adds 0.05 of usage on the opposite direction for each new
376 created flow. This can be useful to simulate some important TCP
377 phenomena such as ack compression.
379 For that to work, your platform must have two links for each
380 pair of interconnected hosts. An example of usable platform is
381 available in <tt>examples/platforms/crosstraffic.xml</tt>.
383 This is activated through the \b network/crosstraffic item, that
384 can be set to 0 (disable this feature) or 1 (enable it).
386 Note that with the default host model this option is activated by default.
388 \subsubsection options_model_network_asyncsend Simulating asyncronous send
390 (this configuration item is experimental and may change or disapear)
392 It is possible to specify that messages below a certain size will be sent
393 as soon as the call to MPI_Send is issued, without waiting for the
394 correspondant receive. This threshold can be configured through the
395 \b smpi/async-small-thresh item. The default value is 0. This behavior can also be
396 manually set for MSG mailboxes, by setting the receiving mode of the mailbox
397 with a call to \ref MSG_mailbox_set_async . For MSG, all messages sent to this
398 mailbox will have this behavior, so consider using two mailboxes if needed.
400 This value needs to be smaller than or equals to the threshold set at
401 \ref options_model_smpi_detached , because asynchronous messages are
402 meant to be detached as well.
404 \subsubsection options_pls Configuring packet-level pseudo-models
406 When using the packet-level pseudo-models, several specific
407 configuration flags are provided to configure the associated tools.
408 There is by far not enough such SimGrid flags to cover every aspects
409 of the associated tools, since we only added the items that we
410 needed ourselves. Feel free to request more items (or even better:
411 provide patches adding more items).
413 When using NS3, the only existing item is \b ns3/TcpModel,
414 corresponding to the ns3::TcpL4Protocol::SocketType configuration item
415 in NS3. The only valid values (enforced on the SimGrid side) are
416 'NewReno' or 'Reno' or 'Tahoe'.
418 \subsection options_model_storage Configuring the Storage model
420 \subsubsection option_model_storage_maxfd Maximum amount of file descriptors per host
422 Each host maintains a fixed-size array of its file descriptors. You
423 can change its size (1024 by default) through the \b
424 storage/max_file_descriptors item to either enlarge it if your
425 application requires it or to reduce it to save memory space.
427 \section options_modelchecking Configuring the Model-Checking
429 To enable the SimGrid model-checking support the program should
430 be executed using the simgrid-mc wrapper:
432 simgrid-mc ./my_program
435 Safety properties are expressed as assertions using the function
437 void MC_assert(int prop);
440 \subsection options_modelchecking_liveness Specifying a liveness property
442 If you want to specify liveness properties (beware, that's
443 experimental), you have to pass them on the command line, specifying
444 the name of the file containing the property, as formatted by the
448 --cfg=model-check/property:<filename>
451 \subsection options_modelchecking_steps Going for stateful verification
453 By default, the system is backtracked to its initial state to explore
454 another path instead of backtracking to the exact step before the fork
455 that we want to explore (this is called stateless verification). This
456 is done this way because saving intermediate states can rapidly
457 exhaust the available memory. If you want, you can change the value of
458 the <tt>model-check/checkpoint</tt> variable. For example, the
459 following configuration will ask to take a checkpoint every step.
460 Beware, this will certainly explode your memory. Larger values are
461 probably better, make sure to experiment a bit to find the right
462 setting for your specific system.
465 --cfg=model-check/checkpoint:1
468 \subsection options_modelchecking_reduction Specifying the kind of reduction
470 The main issue when using the model-checking is the state space
471 explosion. To counter that problem, several exploration reduction
472 techniques can be used. There is unfortunately no silver bullet here,
473 and the most efficient reduction techniques cannot be applied to any
474 properties. In particular, the DPOR method cannot be applied on
475 liveness properties since it may break some cycles in the exploration
476 that are important to the property validity.
479 --cfg=model-check/reduction:<technique>
482 For now, this configuration variable can take 2 values:
483 * none: Do not apply any kind of reduction (mandatory for now for
485 * dpor: Apply Dynamic Partial Ordering Reduction. Only valid if you
486 verify local safety properties (default value for safety checks).
488 \subsection options_modelchecking_visited model-check/visited, Cycle detection
490 In order to detect cycles, the model-checker needs to check if a new explored
491 state is in fact the same state than a previous one. For that,
492 the model-checker can take a snapshot of each visited state: this snapshot is
493 then used to compare it with subsequent states in the exploration graph.
495 The \b model-check/visited option is the maximum number of states which are stored in
496 memory. If the maximum number of snapshotted state is reached, some states will
497 be removed from the memory and some cycles might be missed. Small
498 values can lead to incorrect verifications, but large value can
499 exhaust your memory, so choose carefully.
501 By default, no state is snapshotted and cycles cannot be detected.
503 \subsection options_modelchecking_termination model-check/termination, Non termination detection
505 The \b model-check/termination configuration item can be used to report if a
506 non-termination execution path has been found. This is a path with a cycle
507 which means that the program might never terminate.
509 This only works in safety mode.
511 This options is disabled by default.
513 \subsection options_modelchecking_dot_output model-check/dot-output, Dot output
515 If set, the \b model-check/dot-output configuration item is the name of a file
516 in which to write a dot file of the path leading the found property (safety or
517 liveness violation) as well as the cycle for liveness properties. This dot file
518 can then fed to the graphviz dot tool to generate an corresponding graphical
521 \subsection options_modelchecking_max_depth model-check/max-depth, Depth limit
523 The \b model-checker/max-depth can set the maximum depth of the exploration
524 graph of the model-checker. If this limit is reached, a logging message is
525 sent and the results might not be exact.
527 By default, there is not depth limit.
529 \subsection options_modelchecking_timeout Handling of timeout
531 By default, the model-checker does not handle timeout conditions: the `wait`
532 operations never time out. With the \b model-check/timeout configuration item
533 set to \b yes, the model-checker will explore timeouts of `wait` operations.
535 \subsection options_modelchecking_comm_determinism Communication determinism
537 The \b model-check/communications-determinism and
538 \b model-check/send-determinism items can be used to select the communication
539 determinism mode of the model-checker which checks determinism properties of
540 the communications of an application.
542 \subsection options_modelchecking_sparse_checkpoint Per page checkpoints
544 When the model-checker is configured to take a snapshot of each explored state
545 (with the \b model-checker/visited item), the memory consumption can rapidly
546 reach GiB ou Tib of memory. However, for many workloads, the memory does not
547 change much between different snapshots and taking a complete copy of each
548 snapshot is a waste of memory.
550 The \b model-check/sparse-checkpoint option item can be set to \b yes in order
551 to avoid making a complete copy of each snapshot: instead, each snapshot will be
552 decomposed in blocks which will be stored separately.
553 If multiple snapshots share the same block (or if the same block
554 is used in the same snapshot), the same copy of the block will be shared leading
555 to a reduction of the memory footprint.
557 For many applications, this option considerably reduces the memory consumption.
558 In somes cases, the model-checker might be slightly slower because of the time
559 taken to manage the metadata about the blocks. In other cases however, this
560 snapshotting strategy will be much faster by reducing the cache consumption.
561 When the memory consumption is important, by avoiding to hit the swap or
562 reducing the swap usage, this option might be much faster than the basic
563 snapshotting strategy.
565 This option is currently disabled by default.
567 \subsection options_mc_perf Performance considerations for the model checker
569 The size of the stacks can have a huge impact on the memory
570 consumption when using model-checking. By default, each snapshot will
571 save a copy of the whole stacks and not only of the part which is
572 really meaningful: you should expect the contribution of the memory
573 consumption of the snapshots to be \f$ \mbox{number of processes}
574 \times \mbox{stack size} \times \mbox{number of states} \f$.
576 The \b model-check/sparse-checkpoint can be used to reduce the memory
577 consumption by trying to share memory between the different snapshots.
579 When compiled against the model checker, the stacks are not
580 protected with guards: if the stack size is too small for your
581 application, the stack will silently overflow on other parts of the
582 memory (see \ref options_virt_guard_size).
584 \subsection options_modelchecking_hash Hashing of the state (experimental)
586 Usually most of the time of the model-checker is spent comparing states. This
587 process is complicated and consumes a lot of bandwidth and cache.
588 In order to speedup the state comparison, the experimental \b model-checker/hash
589 configuration item enables the computation of a hash summarizing as much
590 information of the state as possible into a single value. This hash can be used
591 to avoid most of the comparisons: the costly comparison is then only used when
592 the hashes are identical.
594 Currently most of the state is not included in the hash because the
595 implementation was found to be buggy and this options is not as useful as
596 it could be. For this reason, it is currently disabled by default.
598 \subsection options_modelchecking_recordreplay Record/replay (experimental)
600 As the model-checker keeps jumping at different places in the execution graph,
601 it is difficult to understand what happens when trying to debug an application
602 under the model-checker. Event the output of the program is difficult to
603 interpret. Moreover, the model-checker does not behave nicely with advanced
604 debugging tools such as valgrind. For those reason, to identify a trajectory
605 in the execution graph with the model-checker and replay this trajcetory and
606 without the model-checker black-magic but with more standard tools
607 (such as a debugger, valgrind, etc.). For this reason, Simgrid implements an
608 experimental record/replay functionnality in order to record a trajectory with
609 the model-checker and replay it without the model-checker.
611 When the model-checker finds an interesting path in the application execution
612 graph (where a safety or liveness property is violated), it can generate an
613 identifier for this path. In order to enable this behavious the
614 \b model-check/record must be set to \b yes. By default, this behaviour is not
617 This is an example of output:
620 [ 0.000000] (0:@) Check a safety property
621 [ 0.000000] (0:@) **************************
622 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
623 [ 0.000000] (0:@) **************************
624 [ 0.000000] (0:@) Counter-example execution trace:
625 [ 0.000000] (0:@) Path = 1/3;1/4
626 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
627 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
628 [ 0.000000] (0:@) Expanded states = 27
629 [ 0.000000] (0:@) Visited states = 68
630 [ 0.000000] (0:@) Executed transitions = 46
633 This path can then be replayed outside of the model-checker (and even in
634 non-MC build of simgrid) by setting the \b model-check/replay item to the given
635 path. The other options should be the same (but the model-checker should
638 The format and meaning of the path may change between different releases so
639 the same release of Simgrid should be used for the record phase and the replay
642 \section options_virt Configuring the User Process Virtualization
644 \subsection options_virt_factory Selecting the virtualization factory
646 In SimGrid, the user code is virtualized in a specific mechanism
647 that allows the simulation kernel to control its execution: when a user
648 process requires a blocking action (such as sending a message), it is
649 interrupted, and only gets released when the simulated clock reaches
650 the point where the blocking operation is done. This is explained
651 graphically in the [relevant tutorial, available online](http://simgrid.gforge.inria.fr/tutorials/simgrid-simix-101.pdf).
653 In SimGrid, the containers in which user processes are virtualized are
654 called contexts. Several context factory are provided, and you can
655 select the one you want to use with the \b contexts/factory
656 configuration item. Some of the following may not exist on your
657 machine because of portability issues. In any case, the default one
658 should be the most effcient one (please report bugs if the
659 auto-detection fails for you). They are approximately sorted here from
660 the slowest to the most efficient:
662 - \b thread: very slow factory using full featured threads (either
663 pthreads or windows native threads). They are slow but very
664 standard. Some debuggers or profilers only work with this factory.
665 - \b java: Java applications are virtualized onto java threads (that
666 are regular pthreads registered to the JVM)
667 - \b ucontext: fast factory using System V contexts (Linux and FreeBSD only)
668 - \b boost: This uses the [context implementation](http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html)
669 of the boost library for a performance that is comparable to our
670 raw implementation.\n Install the relevant library (e.g. with the
671 libboost-contexts-dev package on Debian/Ubuntu) and recompile
672 SimGrid. Note that our implementation is not compatible with recent
673 implementations of the library, and it will be hard to fix this since
674 the library's author decided to hide an API that we were using.
675 - \b raw: amazingly fast factory using a context switching mechanism
676 of our own, directly implemented in assembly (only available for x86
677 and amd64 platforms for now) and without any unneeded system call.
679 The main reason to change this setting is when the debugging tools get
680 fooled by the optimized context factories. Threads are the most
681 debugging-friendly contextes, as they allow to set breakpoints
682 anywhere with gdb and visualize backtraces for all processes, in order
683 to debug concurrency issues. Valgrind is also more comfortable with
684 threads, but it should be usable with all factories (but the callgrind
685 tool that really don't like raw and ucontext factories).
687 \subsection options_virt_stacksize Adapting the used stack size
689 Each virtualized used process is executed using a specific system
690 stack. The size of this stack has a huge impact on the simulation
691 scalability, but its default value is rather large. This is because
692 the error messages that you get when the stack size is too small are
693 rather disturbing: this leads to stack overflow (overwriting other
694 stacks), leading to segfaults with corrupted stack traces.
696 If you want to push the scalability limits of your code, you might
697 want to reduce the \b contexts/stack-size item. Its default value
698 is 8192 (in KiB), while our Chord simulation works with stacks as small
699 as 16 KiB, for example. For the thread factory, the default value
700 is the one of the system but you can still change it with this parameter.
702 The operating system should only allocate memory for the pages of the
703 stack which are actually used and you might not need to use this in
704 most cases. However, this setting is very important when using the
705 model checker (see \ref options_mc_perf).
707 \subsection options_virt_guard_size Disabling stack guard pages
709 A stack guard page is usually used which prevents the stack of a given
710 actor from overflowing on another stack. But the performance impact
711 may become prohibitive when the amount of actors increases. The
712 option \b contexts:guard-size is the number of stack guard pages used.
713 By setting it to 0, no guard pages will be used: in this case, you
714 should avoid using small stacks (\b stack-size) as the stack will
715 silently overflow on other parts of the memory.
717 When no stack guard page is created, stacks may then silently overflow
718 on other parts of the memory if their size is too small for the
719 application. This happens:
721 - on Windows systems;
722 - when the model checker is enabled;
723 - and of course when guard pages are explicitely disabled (with \b contexts:guard-size=0).
725 \subsection options_virt_parallel Running user code in parallel
727 Parallel execution of the user code is only considered stable in
728 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
729 simulations may well fail in parallel mode. It is described in
730 <a href="http://hal.inria.fr/inria-00602216/">INRIA RR-7653</a>.
732 If you are using the \c ucontext or \c raw context factories, you can
733 request to execute the user code in parallel. Several threads are
734 launched, each of them handling as much user contexts at each run. To
735 actiave this, set the \b contexts/nthreads item to the amount of
736 cores that you have in your computer (or lower than 1 to have
737 the amount of cores auto-detected).
739 Even if you asked several worker threads using the previous option,
740 you can request to start the parallel execution (and pay the
741 associated synchronization costs) only if the potential parallelism is
742 large enough. For that, set the \b contexts/parallel-threshold
743 item to the minimal amount of user contexts needed to start the
744 parallel execution. In any given simulation round, if that amount is
745 not reached, the contexts will be run sequentially directly by the
746 main thread (thus saving the synchronization costs). Note that this
747 option is mainly useful when the grain of the user code is very fine,
748 because our synchronization is now very efficient.
750 When parallel execution is activated, you can choose the
751 synchronization schema used with the \b contexts/synchro item,
752 which value is either:
753 - \b futex: ultra optimized synchronisation schema, based on futexes
754 (fast user-mode mutexes), and thus only available on Linux systems.
755 This is the default mode when available.
756 - \b posix: slow but portable synchronisation using only POSIX
758 - \b busy_wait: not really a synchronisation: the worker threads
759 constantly request new contexts to execute. It should be the most
760 efficient synchronisation schema, but it loads all the cores of your
761 machine for no good reason. You probably prefer the other less
764 \section options_tracing Configuring the tracing subsystem
766 The \ref outcomes_vizu "tracing subsystem" can be configured in several
767 different ways depending on the nature of the simulator (MSG, SimDag,
768 SMPI) and the kind of traces that need to be obtained. See the \ref
769 tracing_tracing_options "Tracing Configuration Options subsection" to
770 get a detailed description of each configuration option.
772 We detail here a simple way to get the traces working for you, even if
773 you never used the tracing API.
776 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
778 --cfg=tracing:yes --cfg=tracing/uncategorized:yes --cfg=triva/uncategorized:uncat.plist
780 The first parameter activates the tracing subsystem, the second
781 tells it to trace host and link utilization (without any
782 categorization) and the third creates a graph configuration file
783 to configure Triva when analysing the resulting trace file.
785 - MSG or SimDag-based simulator and categorized traces (you need to declare categories and classify your tasks according to them)
787 --cfg=tracing:yes --cfg=tracing/categorized:yes --cfg=triva/categorized:cat.plist
789 The first parameter activates the tracing subsystem, the second
790 tells it to trace host and link categorized utilization and the
791 third creates a graph configuration file to configure Triva when
792 analysing the resulting trace file.
794 - SMPI simulator and traces for a space/time view:
798 The <i>-trace</i> parameter for the smpirun script runs the
799 simulation with --cfg=tracing:yes and --cfg=tracing/smpi:yes. Check the
800 smpirun's <i>-help</i> parameter for additional tracing options.
802 Sometimes you might want to put additional information on the trace to
803 correctly identify them later, or to provide data that can be used to
804 reproduce an experiment. You have two ways to do that:
806 - Add a string on top of the trace file as comment:
808 --cfg=tracing/comment:my_simulation_identifier
811 - Add the contents of a textual file on top of the trace file as comment:
813 --cfg=tracing/comment-file:my_file_with_additional_information.txt
816 Please, use these two parameters (for comments) to make reproducible
817 simulations. For additional details about this and all tracing
818 options, check See the \ref tracing_tracing_options.
820 \section options_msg Configuring MSG
822 \subsection options_msg_debug_multiple_use Debugging MSG
824 Sometimes your application may try to send a task that is still being
825 executed somewhere else, making it impossible to send this task. However,
826 for debugging purposes, one may want to know what the other host is/was
827 doing. This option shows a backtrace of the other process.
829 Enable this option by adding
832 --cfg=msg/debug-multiple-use:on
835 \section options_smpi Configuring SMPI
837 The SMPI interface provides several specific configuration items.
838 These are uneasy to see since the code is usually launched through the
839 \c smiprun script directly.
841 \subsection options_smpi_bench smpi/bench: Automatic benchmarking of SMPI code
843 In SMPI, the sequential code is automatically benchmarked, and these
844 computations are automatically reported to the simulator. That is to
845 say that if you have a large computation between a \c MPI_Recv() and a
846 \c MPI_Send(), SMPI will automatically benchmark the duration of this
847 code, and create an execution task within the simulator to take this
848 into account. For that, the actual duration is measured on the host
849 machine and then scaled to the power of the corresponding simulated
850 machine. The variable \b smpi/host-speed allows to specify the
851 computational speed of the host machine (in flop/s) to use when
852 scaling the execution times. It defaults to 20000, but you really want
853 to update it to get accurate simulation results.
855 When the code is constituted of numerous consecutive MPI calls, the
856 previous mechanism feeds the simulation kernel with numerous tiny
857 computations. The \b smpi/cpu-threshold item becomes handy when this
858 impacts badly the simulation performance. It specifies a threshold (in
859 seconds) below which the execution chunks are not reported to the
860 simulation kernel (default value: 1e-6).
863 The option smpi/cpu-threshold ignores any computation time spent
864 below this threshold. SMPI does not consider the \a amount of these
865 computations; there is no offset for this. Hence, by using a
866 value that is too low, you may end up with unreliable simulation
869 In some cases, however, one may wish to disable simulation of
870 application computation. This is the case when SMPI is used not to
871 simulate an MPI applications, but instead an MPI code that performs
872 "live replay" of another MPI app (e.g., ScalaTrace's replay tool,
873 various on-line simulators that run an app at scale). In this case the
874 computation of the replay/simulation logic should not be simulated by
875 SMPI. Instead, the replay tool or on-line simulator will issue
876 "computation events", which correspond to the actual MPI simulation
877 being replayed/simulated. At the moment, these computation events can
878 be simulated using SMPI by calling internal smpi_execute*() functions.
880 To disable the benchmarking/simulation of computation in the simulated
881 application, the variable \b smpi/simulate-computation should be set to no.
884 This option just ignores the timings in your simulation; it still executes
885 the computations itself. If you want to stop SMPI from doing that,
886 you should check the SMPI_SAMPLE macros, documented in the section
887 \ref SMPI_adapting_speed.
889 Solution | Computations actually executed? | Computations simulated ?
890 ---------------------------------- | ------------------------------- | ------------------------
891 --cfg=smpi/simulate-computation:no | Yes | No, never
892 --cfg=smpi/cpu-threshold:42 | Yes, in all cases | Only if it lasts more than 42 seconds
893 SMPI_SAMPLE() macro | Only once per loop nest (see @ref SMPI_adapting_speed "documentation") | Always
895 \subsection options_model_smpi_adj_file smpi/comp-adjustment-file: Slow-down or speed-up parts of your code.
897 This option allows you to pass a file that contains two columns: The first column
898 defines the section that will be subject to a speedup; the second column is the speedup.
904 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
907 The first line is the header - you must include it.
908 The following line means that the code between two consecutive MPI calls on
909 line 30 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
910 of 1.18244559422142. The value for the second column is therefore a speedup, if it is
911 larger than 1 and a slow-down if it is smaller than 1. Nothing will be changed if it is
914 Of course, you can set any arbitrary filenames you want (so the start and end don't have to be
915 in the same file), but be aware that this mechanism only supports @em consecutive calls!
918 Please note that you must pass the \b -trace-call-location flag to smpicc
919 or smpiff, respectively! This flag activates some macro definitions in our
920 mpi.h / mpi.f files that help with obtaining the call location.
922 \subsection options_model_smpi_bw_factor smpi/bw-factor: Bandwidth factors
924 The possible throughput of network links is often dependent on the
925 message sizes, as protocols may adapt to different message sizes. With
926 this option, a series of message sizes and factors are given, helping
927 the simulation to be more realistic. For instance, the current
931 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
934 So, messages with size 65472 and more will get a total of MAX_BANDWIDTH*0.940694,
935 messages of size 15424 to 65471 will get MAX_BANDWIDTH*0.697866 and so on.
936 Here, MAX_BANDWIDTH denotes the bandwidth of the link.
939 The SimGrid-Team has developed a script to help you determine these
940 values. You can find more information and the download here:
941 1. http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
942 2. http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
944 \subsection options_smpi_timing smpi/display-timing: Reporting simulation time
946 \b Default: 0 (false)
948 Most of the time, you run MPI code with SMPI to compute the time it
949 would take to run it on a platform. But since the
950 code is run through the \c smpirun script, you don't have any control
951 on the launcher code, making it difficult to report the simulated time
952 when the simulation ends. If you set the \b smpi/display-timing item
953 to 1, \c smpirun will display this information when the simulation ends. \verbatim
954 Simulation time: 1e3 seconds.
957 \subsection options_smpi_temps smpi/keep-temps: not cleaning up after simulation
959 \b Default: 0 (false)
961 Under some conditions, SMPI generates a lot of temporary files. They
962 usually get cleaned, but you may use this option to not erase these
963 files. This is for example useful when debugging or profiling
964 executions using the dlopen privatization schema, as missing binary
965 files tend to fool the debuggers.
967 \subsection options_model_smpi_lat_factor smpi/lat-factor: Latency factors
969 The motivation and syntax for this option is identical to the motivation/syntax
970 of smpi/bw-factor, see \ref options_model_smpi_bw_factor for details.
972 There is an important difference, though: While smpi/bw-factor \a reduces the
973 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
974 increase the latency, i.e., values larger than or equal to 1 are valid here.
976 This is the default value:
979 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
983 The SimGrid-Team has developed a script to help you determine these
984 values. You can find more information and the download here:
985 1. http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
986 2. http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
988 \subsection options_smpi_papi_events smpi/papi-events: Trace hardware counters with PAPI
991 This option is experimental and will be subject to change.
992 This feature currently requires superuser privileges, as registers are queried.
993 Only use this feature with code you trust! Call smpirun for instance via
994 smpirun -wrapper "sudo " <your-parameters>
995 or run sudo sh -c "echo 0 > /proc/sys/kernel/perf_event_paranoid"
996 In the later case, sudo will not be required.
999 This option is only available when SimGrid was compiled with PAPI support.
1001 This option takes the names of PAPI counters and adds their respective values
1002 to the trace files. (See Section \ref tracing_tracing_options.)
1004 It is planned to make this feature available on a per-process (or per-thread?) basis.
1005 The first draft, however, just implements a "global" (i.e., for all processes) set
1006 of counters, the "default" set.
1009 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1012 \subsection options_smpi_privatization smpi/privatization: Automatic privatization of global variables
1014 MPI executables are usually meant to be executed in separated
1015 processes, but SMPI is executed in only one process. Global variables
1016 from executables will be placed in the same memory zone and shared
1017 between processes, causing intricate bugs. Several options are
1018 possible to avoid this, as described in the main
1019 <a href="https://hal.inria.fr/hal-01415484">SMPI publication</a> and in
1020 the @ref SMPI_what_globals "SMPI documentation". SimGrid provides two
1021 ways of automatically privatizing the globals, and this option allows
1022 to choose between them.
1024 - <b>no</b> (default when not using smpirun): Do not automatically privatize variables.
1025 Pass \c -no-privatize to smpirun to disable this feature.
1026 - <b>dlopen</b> or <b>yes</b> (default when using smpirun): Link multiple times against the binary.
1027 - <b>mmap</b> (slower, but maybe somewhat more stable):
1028 Runtime automatic switching of the data segments.
1031 This configuration option cannot be set in your platform file. You can only
1032 pass it as an argument to smpirun.
1034 \subsection options_model_smpi_detached Simulating MPI detached send
1036 This threshold specifies the size in bytes under which the send will return
1037 immediately. This is different from the threshold detailed in \ref options_model_network_asyncsend
1038 because the message is not effectively sent when the send is posted. SMPI still waits for the
1039 correspondant receive to be posted to perform the communication operation. This threshold can be set
1040 by changing the \b smpi/send-is-detached-thresh item. The default value is 65536.
1042 \subsection options_model_smpi_collectives Simulating MPI collective algorithms
1044 SMPI implements more than 100 different algorithms for MPI collective communication, to accurately
1045 simulate the behavior of most of the existing MPI libraries. The \b smpi/coll-selector item can be used
1046 to use the decision logic of either OpenMPI or MPICH libraries (values: ompi or mpich, by default SMPI
1047 uses naive version of collective operations). Each collective operation can be manually selected with a
1048 \b smpi/collective_name:algo_name. Available algorithms are listed in \ref SMPI_use_colls .
1050 \subsection options_model_smpi_iprobe smpi/iprobe: Inject constant times for calls to MPI_Iprobe
1052 \b Default value: 0.0001
1054 The behavior and motivation for this configuration option is identical with \a smpi/test, see
1055 Section \ref options_model_smpi_test for details.
1057 \subsection options_model_smpi_iprobe_cpu_usage smpi/iprobe-cpu-usage: Reduce speed for iprobe calls
1059 \b Default value: 1 (no change from default behavior)
1061 MPI_Iprobe calls can be heavily used in applications. To account correctly for the energy
1062 cores spend probing, it is necessary to reduce the load that these calls cause inside
1065 For instance, we measured a max power consumption of 220 W for a particular application but
1066 only 180 W while this application was probing. Hence, the correct factor that should
1067 be passed to this option would be 180/220 = 0.81.
1069 \subsection options_model_smpi_init smpi/init: Inject constant times for calls to MPI_Init
1073 The behavior for this configuration option is identical with \a smpi/test, see
1074 Section \ref options_model_smpi_test for details.
1076 \subsection options_model_smpi_ois smpi/ois: Inject constant times for asynchronous send operations
1078 This configuration option works exactly as \a smpi/os, see Section \ref options_model_smpi_os.
1079 Of course, \a smpi/ois is used to account for MPI_Isend instead of MPI_Send.
1081 \subsection options_model_smpi_os smpi/os: Inject constant times for send operations
1083 In several network models such as LogP, send (MPI_Send, MPI_Isend) and receive (MPI_Recv)
1084 operations incur costs (i.e., they consume CPU time). SMPI can factor these costs in as well, but the
1085 user has to configure SMPI accordingly as these values may vary by machine.
1086 This can be done by using smpi/os for MPI_Send operations; for MPI_Isend and
1087 MPI_Recv, use \a smpi/ois and \a smpi/or, respectively. These work exactly as
1090 \a smpi/os can consist of multiple sections; each section takes three values, for example:
1096 Here, the sections are divided by ";" (that is, this example contains two sections).
1097 Furthermore, each section consists of three values.
1099 1. The first value denotes the minimum size for this section to take effect;
1100 read it as "if message size is greater than this value (and other section has a larger
1101 first value that is also smaller than the message size), use this".
1102 In the first section above, this value is "1".
1104 2. The second value is the startup time; this is a constant value that will always
1105 be charged, no matter what the size of the message. In the first section above,
1108 3. The third value is the \a per-byte cost. That is, it is charged for every
1109 byte of the message (incurring cost messageSize*cost_per_byte)
1110 and hence accounts also for larger messages. In the first
1111 section of the example above, this value is "2".
1113 Now, SMPI always checks which section it should take for a given message; that is,
1114 if a message of size 11 is sent with the configuration of the example above, only
1115 the second section will be used, not the first, as the first value of the second
1116 section is closer to the message size. Hence, a message of size 11 incurs the
1117 following cost inside MPI_Send:
1123 As 5 is the startup cost and 1 is the cost per byte.
1126 The order of sections can be arbitrary; they will be ordered internally.
1128 \subsection options_model_smpi_or smpi/or: Inject constant times for receive operations
1130 This configuration option works exactly as \a smpi/os, see Section \ref options_model_smpi_os.
1131 Of course, \a smpi/or is used to account for MPI_Recv instead of MPI_Send.
1133 \subsection options_model_smpi_test smpi/test: Inject constant times for calls to MPI_Test
1135 \b Default value: 0.0001
1137 By setting this option, you can control the amount of time a process sleeps
1138 when MPI_Test() is called; this is important, because SimGrid normally only
1139 advances the time while communication is happening and thus,
1140 MPI_Test will not add to the time, resulting in a deadlock if used as a
1147 MPI_Test(request, flag, status);
1153 Internally, in order to speed up execution, we use a counter to keep track
1154 on how often we already checked if the handle is now valid or not. Hence, we
1155 actually use counter*SLEEP_TIME, that is, the time MPI_Test() causes the process
1156 to sleep increases linearly with the number of previously failed tests. This
1157 behavior can be disabled by setting smpi/grow-injected-times to no. This will
1158 also disable this behavior for MPI_Iprobe.
1161 \subsection options_model_smpi_shared_malloc smpi/shared-malloc: Factorize malloc()s
1165 If your simulation consumes too much memory, you may want to modify
1166 your code so that the working areas are shared by all MPI ranks. For
1167 example, in a bloc-cyclic matrix multiplication, you will only
1168 allocate one set of blocs, and every processes will share them.
1169 Naturally, this will lead to very wrong results, but this will save a
1170 lot of memory so this is still desirable for some studies. For more on
1171 the motivation for that feature, please refer to the
1172 <a href="https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication/">relevant
1173 section</a> of the SMPI CourseWare (see Activity #2.2 of the pointed
1174 assignment). In practice, change the call to malloc() and free() into
1175 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1177 SMPI provides 2 algorithms for this feature. The first one, called \c
1178 local, allocates one bloc per call to SMPI_SHARED_MALLOC() in your
1179 code (each call location gets its own bloc) and this bloc is shared
1180 amongst all MPI ranks. This is implemented with the shm_* functions
1181 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1182 for each shared bloc.
1184 With the \c global algorithm, each call to SMPI_SHARED_MALLOC()
1185 returns a new adress, but it only points to a shadow bloc: its memory
1186 area is mapped on a 1MiB file on disk. If the returned bloc is of size
1187 N MiB, then the same file is mapped N times to cover the whole bloc.
1188 At the end, no matter how many SMPI_SHARED_MALLOC you do, this will
1189 only consume 1 MiB in memory.
1191 You can disable this behavior and come back to regular mallocs (for
1192 example for debugging purposes) using \c "no" as a value.
1194 If you want to keep private some parts of the buffer, for instance if these
1195 parts are used by the application logic and should not be corrupted, you
1196 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count).
1201 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1204 will allocate 500 bytes to mem, such that mem[27..41] and mem[100..199]
1205 are shared and other area remain private.
1207 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1209 When smpi/shared-malloc:global is used, the memory consumption problem
1210 is solved, but it may induce too much load on the kernel's pages table.
1211 In this case, you should use huge pages so that we create only one
1212 entry per Mb of malloced data instead of one entry per 4k.
1213 To activate this, you must mount a hugetlbfs on your system and allocate
1214 at least one huge page:
1218 sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1219 sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1222 Then, you can pass the option --cfg=smpi/shared-malloc-hugepage:/home/huge
1223 to smpirun to actually activate the huge page support in shared mallocs.
1225 \subsection options_model_smpi_wtime smpi/wtime: Inject constant times for calls to MPI_Wtime
1229 By setting this option, you can control the amount of time a process sleeps
1230 when MPI_Wtime() is called; this is important, because SimGrid normally only
1231 advances the time while communication is happening and thus,
1232 MPI_Wtime will not add to the time, resulting in a deadlock if used as a
1238 while(MPI_Wtime() < some_time_bound) {
1243 If the time is never advanced, this loop will clearly never end as MPI_Wtime()
1244 always returns the same value. Hence, pass a (small) value to the smpi/wtime
1245 option to force a call to MPI_Wtime to advance the time as well.
1248 \section options_generic Configuring other aspects of SimGrid
1250 \subsection options_generic_clean_atexit Cleanup before termination
1252 The C / C++ standard contains a function called \b [atexit](http://www.cplusplus.com/reference/cstdlib/atexit/).
1253 atexit registers callbacks, which are called just before the program terminates.
1255 By setting the configuration option clean-atexit to 1 (true), a callback
1256 is registered and will clean up some variables and terminate/cleanup the tracing.
1258 TODO: Add when this should be used.
1260 \subsection options_generic_path Profile files' search path
1262 It is possible to specify a list of directories to search into for the
1263 trace files (see @ref pf_trace) by using the \b path configuration
1264 item. To add several directory to the path, set the configuration
1265 item several times, as in \verbatim
1266 --cfg=path:toto --cfg=path:tutu
1269 \subsection options_generic_breakpoint Set a breakpoint
1272 --cfg=simix/breakpoint:3.1416
1275 This configuration option sets a breakpoint: when the simulated clock reaches
1276 the given time, a SIGTRAP is raised. This can be used to stop the execution and
1277 get a backtrace with a debugger.
1279 It is also possible to set the breakpoint from inside the debugger, by writing
1280 in global variable simgrid::simix::breakpoint. For example, with gdb:
1283 set variable simgrid::simix::breakpoint = 3.1416
1286 \subsection options_generic_exit Behavior on Ctrl-C
1288 By default, when Ctrl-C is pressed, the status of all existing
1289 simulated processes is displayed before exiting the simulation. This is very useful to debug your
1290 code, but it can reveal troublesome in some cases (such as when the
1291 amount of processes becomes really big). This behavior is disabled
1292 when \b verbose-exit is set to 0 (it is to 1 by default).
1294 \subsection options_exception_cutpath Truncate local path from exception backtrace
1297 --cfg=exception/cutpath:1
1300 This configuration option is used to remove the path from the
1301 backtrace shown when an exception is thrown. This is mainly useful for
1302 the tests: the full file path makes the tests not reproducible, and
1303 thus failing as we are currently comparing output. Clearly, the path
1304 used on different machines are almost guaranteed to be different and
1305 hence, the output would mismatch, causing the test to fail.
1307 \section options_log Logging Configuration
1309 It can be done by using XBT. Go to \ref XBT_log for more details.