8 <object id="TOC" data="graphical-toc.svg" type="image/svg+xml"></object>
10 window.onload=function() { // Wait for the SVG to be loaded before changing it
11 var elem=document.querySelector("#TOC").contentDocument.getElementById("ConfigBox")
12 elem.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";
18 A number of options can be given at runtime to change the default
19 SimGrid behavior. For a complete list of all configuration options
20 accepted by the SimGrid version used in your simulator, simply pass
21 the --help configuration flag to your program. If some of the options
22 are not documented on this page, this is a bug that you should please
23 report so that we can fix it. Note that some of the options presented
24 here may not be available in your simulators, depending on the
25 :ref:`compile-time options <install_src_config>` that you used.
27 Setting Configuration Items
28 ---------------------------
30 There is several way to pass configuration options to the simulators.
31 The most common way is to use the ``--cfg`` command line argument. For
32 example, to set the item ``Item`` to the value ``Value``, simply
33 type the following on the command-line:
37 my_simulator --cfg=Item:Value (other arguments)
39 Several ``--cfg`` command line arguments can naturally be used. If you
40 need to include spaces in the argument, don't forget to quote the
41 argument. You can even escape the included quotes (write ``@'`` for ``'`` if
42 you have your argument between simple quotes).
44 Another solution is to use the ``<config>`` tag in the platform file. The
45 only restriction is that this tag must occur before the first
46 platform element (be it ``<zone>``, ``<cluster>``, ``<peer>`` or whatever).
47 The ``<config>`` tag takes an ``id`` attribute, but it is currently
48 ignored so you don't really need to pass it. The important part is that
49 within that tag, you can pass one or several ``<prop>`` tags to specify
50 the configuration to use. For example, setting ``Item`` to ``Value``
51 can be done by adding the following to the beginning of your platform
57 <prop id="Item" value="Value"/>
60 A last solution is to pass your configuration directly in your program
61 with :cpp:func:`simgrid::s4u::Engine::set_config` or :cpp:func:`MSG_config`.
65 #include <simgrid/s4u.hpp>
67 int main(int argc, char *argv[]) {
68 simgrid::s4u::Engine e(&argc, argv);
70 e.set_config("Item:Value");
77 Existing Configuration Items
78 ----------------------------
81 The full list can be retrieved by passing ``--help`` and
82 ``--help-cfg`` to an executable that uses SimGrid. Try passing
83 ``help`` as a value to get the list of values accepted by a given
84 option. For example, ``--cfg=plugin:help`` will give you the list
85 of plugins available in your installation of SimGrid.
87 - **contexts/factory:** :ref:`cfg=contexts/factory`
88 - **contexts/guard-size:** :ref:`cfg=contexts/guard-size`
89 - **contexts/nthreads:** :ref:`cfg=contexts/nthreads`
90 - **contexts/parallel-threshold:** :ref:`cfg=contexts/parallel-threshold`
91 - **contexts/stack-size:** :ref:`cfg=contexts/stack-size`
92 - **contexts/synchro:** :ref:`cfg=contexts/synchro`
94 - **cpu/maxmin-selective-update:** :ref:`Cpu Optimization Level <options_model_optim>`
95 - **cpu/model:** :ref:`options_model_select`
96 - **cpu/optim:** :ref:`Cpu Optimization Level <options_model_optim>`
98 - **debug/breakpoint:** :ref:`cfg=debug/breakpoint`
99 - **debug/clean-atexit:** :ref:`cfg=debug/clean-atexit`
100 - **debug/verbose-exit:** :ref:`cfg=debug/verbose-exit`
102 - **exception/cutpath:** :ref:`cfg=exception/cutpath`
104 - **host/model:** :ref:`options_model_select`
106 - **maxmin/precision:** :ref:`cfg=maxmin/precision`
107 - **maxmin/concurrency-limit:** :ref:`cfg=maxmin/concurrency-limit`
109 - **msg/debug-multiple-use:** :ref:`cfg=msg/debug-multiple-use`
111 - **model-check:** :ref:`options_modelchecking`
112 - **model-check/checkpoint:** :ref:`cfg=model-check/checkpoint`
113 - **model-check/communications-determinism:** :ref:`cfg=model-check/communications-determinism`
114 - **model-check/dot-output:** :ref:`cfg=model-check/dot-output`
115 - **model-check/max-depth:** :ref:`cfg=model-check/max-depth`
116 - **model-check/property:** :ref:`cfg=model-check/property`
117 - **model-check/reduction:** :ref:`cfg=model-check/reduction`
118 - **model-check/replay:** :ref:`cfg=model-check/replay`
119 - **model-check/send-determinism:** :ref:`cfg=model-check/send-determinism`
120 - **model-check/termination:** :ref:`cfg=model-check/termination`
121 - **model-check/timeout:** :ref:`cfg=model-check/timeout`
122 - **model-check/visited:** :ref:`cfg=model-check/visited`
124 - **network/bandwidth-factor:** :ref:`cfg=network/bandwidth-factor`
125 - **network/crosstraffic:** :ref:`cfg=network/crosstraffic`
126 - **network/latency-factor:** :ref:`cfg=network/latency-factor`
127 - **network/loopback-lat:** :ref:`cfg=network/loopback`
128 - **network/loopback-bw:** :ref:`cfg=network/loopback`
129 - **network/maxmin-selective-update:** :ref:`Network Optimization Level <options_model_optim>`
130 - **network/model:** :ref:`options_model_select`
131 - **network/optim:** :ref:`Network Optimization Level <options_model_optim>`
132 - **network/TCP-gamma:** :ref:`cfg=network/TCP-gamma`
133 - **network/weight-S:** :ref:`cfg=network/weight-S`
135 - **ns3/TcpModel:** :ref:`options_pls`
136 - **path:** :ref:`cfg=path`
137 - **plugin:** :ref:`cfg=plugin`
139 - **storage/max_file_descriptors:** :ref:`cfg=storage/max_file_descriptors`
141 - **surf/precision:** :ref:`cfg=surf/precision`
143 - **For collective operations of SMPI,** please refer to Section :ref:`cfg=smpi/coll-selector`
144 - **smpi/async-small-thresh:** :ref:`cfg=smpi/async-small-thresh`
145 - **smpi/buffering:** :ref:`cfg=smpi/buffering`
146 - **smpi/bw-factor:** :ref:`cfg=smpi/bw-factor`
147 - **smpi/coll-selector:** :ref:`cfg=smpi/coll-selector`
148 - **smpi/comp-adjustment-file:** :ref:`cfg=smpi/comp-adjustment-file`
149 - **smpi/cpu-threshold:** :ref:`cfg=smpi/cpu-threshold`
150 - **smpi/display-timing:** :ref:`cfg=smpi/display-timing`
151 - **smpi/grow-injected-times:** :ref:`cfg=smpi/grow-injected-times`
152 - **smpi/host-speed:** :ref:`cfg=smpi/host-speed`
153 - **smpi/IB-penalty-factors:** :ref:`cfg=smpi/IB-penalty-factors`
154 - **smpi/iprobe:** :ref:`cfg=smpi/iprobe`
155 - **smpi/iprobe-cpu-usage:** :ref:`cfg=smpi/iprobe-cpu-usage`
156 - **smpi/init:** :ref:`cfg=smpi/init`
157 - **smpi/keep-temps:** :ref:`cfg=smpi/keep-temps`
158 - **smpi/lat-factor:** :ref:`cfg=smpi/lat-factor`
159 - **smpi/ois:** :ref:`cfg=smpi/ois`
160 - **smpi/or:** :ref:`cfg=smpi/or`
161 - **smpi/os:** :ref:`cfg=smpi/os`
162 - **smpi/papi-events:** :ref:`cfg=smpi/papi-events`
163 - **smpi/privatization:** :ref:`cfg=smpi/privatization`
164 - **smpi/privatize-libs:** :ref:`cfg=smpi/privatize-libs`
165 - **smpi/send-is-detached-thresh:** :ref:`cfg=smpi/send-is-detached-thresh`
166 - **smpi/shared-malloc:** :ref:`cfg=smpi/shared-malloc`
167 - **smpi/shared-malloc-hugepage:** :ref:`cfg=smpi/shared-malloc-hugepage`
168 - **smpi/simulate-computation:** :ref:`cfg=smpi/simulate-computation`
169 - **smpi/test:** :ref:`cfg=smpi/test`
170 - **smpi/wtime:** :ref:`cfg=smpi/wtime`
172 - **Tracing configuration options** can be found in Section :ref:`tracing_tracing_options`
174 - **storage/model:** :ref:`options_model_select`
176 - **vm/model:** :ref:`options_model_select`
180 Configuring the Platform Models
181 -------------------------------
183 .. _options_model_select:
185 Choosing the Platform Models
186 ............................
188 SimGrid comes with several network, CPU and disk models built in,
189 and you can change the used model at runtime by changing the passed
190 configuration. The three main configuration items are given below.
191 For each of these items, passing the special ``help`` value gives you
192 a short description of all possible values (for example,
193 ``--cfg=network/model:help`` will present all provided network
194 models). Also, ``--help-models`` should provide information about all
195 models for all existing resources.
197 - ``network/model``: specify the used network model. Possible values:
199 - **LV08 (default one):** Realistic network analytic model
200 (slow-start modeled by multiplying latency by 13.01, bandwidth by
201 .97; bottleneck sharing uses a payload of S=20537 for evaluating
202 RTT). Described in `Accuracy Study and Improvement of Network
203 Simulation in the SimGrid Framework
204 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
205 - **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 - **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). This model can be
211 :ref:`further configured <options_model_network>`.
212 - **IB:** Realistic network model specifically tailored for HPC
213 settings with InfiniBand networks (accurate modeling contention
214 behavior, based on the model explained in `this PhD work
215 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
216 This model can be :ref:`further configured <options_model_network>`.
217 - **CM02:** Legacy network analytic model. Very similar to LV08, but
218 without corrective factors. The timings of small messages are thus
219 poorly modeled. This model is described in `A Network Model for
220 Simulation of Grid Application
221 <https://hal.inria.fr/inria-00071989/document>`_.
222 - **ns-3** (only available if you compiled SimGrid accordingly):
223 Use the packet-level network
224 simulators as network models (see :ref:`model_ns3`).
225 This model can be :ref:`further configured <options_pls>`.
227 - ``cpu/model``: specify the used CPU model. We have only one model
230 - **Cas01:** Simplistic CPU model (time=size/speed)
232 - ``host/model``: The host concept is the aggregation of a CPU with a
233 network card. Three models exists, but actually, only 2 of them are
234 interesting. The "compound" one is simply due to the way our
235 internal code is organized, and can easily be ignored. So at the
236 end, you have two host models: The default one allows aggregation of
237 an existing CPU model with an existing network model, but does not
238 allow parallel tasks because these beasts need some collaboration
239 between the network and CPU model. That is why, ptask_07 is used by
240 default when using SimDag.
242 - **default:** Default host model. Currently, CPU:Cas01 and
243 network:LV08 (with cross traffic enabled)
244 - **compound:** Host model that is automatically chosen if
245 you change the network and CPU models
246 - **ptask_L07:** Host model somehow similar to Cas01+CM02 but
247 allowing "parallel tasks", that are intended to model the moldable
248 tasks of the grid scheduling literature.
250 - ``storage/model``: specify the used storage model. Only one model is
252 - ``vm/model``: specify the model for virtual machines. Only one model
255 .. todo: make 'compound' the default host model.
257 .. _options_model_optim:
262 The network and CPU models that are based on lmm_solve (that
263 is, all our analytical models) accept specific optimization
266 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
268 - **Lazy:** Lazy action management (partial invalidation in lmm +
269 heap in action remaining).
270 - **TI:** Trace integration. Highly optimized mode when using
271 availability traces (only available for the Cas01 CPU model for
273 - **Full:** Full update of remaining and variables. Slow but may be
274 useful when debugging.
276 - items ``network/maxmin-selective-update`` and
277 ``cpu/maxmin-selective-update``: configure whether the underlying
278 should be lazily updated or not. It should have no impact on the
279 computed timings, but should speed up the computation. |br| It is
280 still possible to disable this feature because it can reveal
281 counter-productive in very specific scenarios where the
282 interaction level is high. In particular, if all your
283 communication share a given backbone link, you should disable it:
284 without it, a simple regular loop is used to update each
285 communication. With it, each of them is still updated (because of
286 the dependency induced by the backbone), but through a complicated
287 and slow pattern that follows the actual dependencies.
289 .. _cfg=maxmin/precision:
290 .. _cfg=surf/precision:
295 **Option** ``maxmin/precision`` **Default:** 0.00001 (in flops or bytes) |br|
296 **Option** ``surf/precision`` **Default:** 0.00001 (in seconds)
298 The analytical models handle a lot of floating point values. It is
299 possible to change the epsilon used to update and compare them through
300 this configuration item. Changing it may speedup the simulation by
301 discarding very small actions, at the price of a reduced numerical
302 precision. You can modify separately the precision used to manipulate
303 timings (in seconds) and the one used to manipulate amounts of work
306 .. _cfg=maxmin/concurrency-limit:
311 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
313 The maximum number of variables per resource can be tuned through this
314 option. You can have as many simultaneous actions per resources as you
315 want. If your simulation presents a very high level of concurrency, it
316 may help to use e.g. 100 as a value here. It means that at most 100
317 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
319 resource 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 .. _options_model_network:
327 Configuring the Network Model
328 .............................
330 .. _cfg=network/TCP-gamma:
332 Maximal TCP Window Size
333 ^^^^^^^^^^^^^^^^^^^^^^^
335 **Option** ``network/TCP-gamma`` **Default:** 4194304
337 The analytical models need to know the maximal TCP window size to take
338 the TCP congestion mechanism into account. On Linux, this value can
339 be retrieved using the following commands. Both give a set of values,
340 and you should use the last one, which is the maximal size.
342 .. code-block:: shell
344 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
345 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
347 .. _cfg=smpi/IB-penalty-factors:
348 .. _cfg=network/bandwidth-factor:
349 .. _cfg=network/latency-factor:
350 .. _cfg=network/weight-S:
352 Correcting Important Network Parameters
353 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
355 SimGrid can take network irregularities such as a slow startup or
356 changing behavior depending on the message size into account. You
357 should not change these values unless you really know what you're
358 doing. The corresponding values were computed through data fitting
359 one the timings of packet-level simulators, as described in `Accuracy
360 Study and Improvement of Network Simulation in the SimGrid Framework
361 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
364 If you are using the SMPI model, these correction coefficients are
365 themselves corrected by constant values depending on the size of the
366 exchange. By default SMPI uses factors computed on the Stampede
367 Supercomputer at TACC, with optimal deployment of processes on
368 nodes. Again, only hardcore experts should bother about this fact.
370 InfiniBand network behavior can be modeled through 3 parameters
371 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `this PhD
373 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
375 .. todo:: This section should be rewritten, and actually explain the
376 options network/bandwidth-factor, network/latency-factor,
379 .. _cfg=network/crosstraffic:
381 Simulating Cross-Traffic
382 ^^^^^^^^^^^^^^^^^^^^^^^^
384 Since SimGrid v3.7, cross-traffic effects can be taken into account in
385 analytical simulations. It means that ongoing and incoming
386 communication flows are treated independently. In addition, the LV08
387 model adds 0.05 of usage on the opposite direction for each new
388 created flow. This can be useful to simulate some important TCP
389 phenomena such as ack compression.
391 For that to work, your platform must have two links for each
392 pair of interconnected hosts. An example of usable platform is
393 available in ``examples/platforms/crosstraffic.xml``.
395 This is activated through the ``network/crosstraffic`` item, that
396 can be set to 0 (disable this feature) or 1 (enable it).
398 Note that with the default host model this option is activated by default.
400 .. _cfg=network/loopback:
402 Configuring loopback link
403 ^^^^^^^^^^^^^^^^^^^^^^^^^
405 Several network model provide an implicit loopback link to account for local
406 communication on a host. By default it has a 10GBps bandwidth and a null latency.
407 This can be changed with ``network/loopback-lat`` and ``network/loopback-bw``
410 .. _cfg=smpi/async-small-thresh:
412 Simulating Asynchronous Send
413 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
415 (this configuration item is experimental and may change or disappear)
417 It is possible to specify that messages below a certain size will be
418 sent as soon as the call to MPI_Send is issued, without waiting for
419 the correspondant receive. This threshold can be configured through
420 the ``smpi/async-small-thresh`` item. The default value is 0. This
421 behavior can also be manually set for mailboxes, by setting the
422 receiving mode of the mailbox with a call to
423 :cpp:func:`MSG_mailbox_set_async`. After this, all messages sent to
424 this mailbox will have this behavior regardless of the message size.
426 This value needs to be smaller than or equals to the threshold set at
427 :ref:`cfg=smpi/send-is-detached-thresh`, because asynchronous messages
428 are meant to be detached as well.
435 **Option** ``ns3/TcpModel`` **Default:** "default" (ns-3 default)
437 When using ns-3, there is an extra item ``ns3/TcpModel``, corresponding
438 to the ``ns3::TcpL4Protocol::SocketType`` configuration item in
439 ns-3. The only valid values (enforced on the SimGrid side) are
440 'default' (no change to the ns-3 configuration), 'NewReno' or 'Reno' or
443 Configuring the Storage model
444 .............................
446 .. _cfg=storage/max_file_descriptors:
448 File Descriptor Cound per Host
449 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
451 **Option** ``storage/max_file_descriptors`` **Default:** 1024
453 Each host maintains a fixed-size array of its file descriptors. You
454 can change its size through this item to either enlarge it if your
455 application requires it or to reduce it to save memory space.
462 SimGrid plugins allow one to extend the framework without changing its
463 source code directly. Read the source code of the existing plugins to
464 learn how to do so (in ``src/plugins``), and ask your questions to the
465 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
466 that plugins usually register callbacks to some signals of interest.
467 If they need to store some information about a given object (Link, CPU
468 or Actor), they do so through the use of a dedicated object extension.
470 Some of the existing plugins can be activated from the command line,
471 meaning that you can activate them from the command line without any
472 modification to your simulation code. For example, you can activate
473 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
476 Here is a partial list of plugins that can be activated this way. You can get
477 the full list by passing ``--cfg=plugin:help`` to your simulator.
479 - :ref:`Host Energy <plugin_host_energy>`: models the energy dissipation of the compute units.
480 - :ref:`Link Energy <plugin_link_energy>`: models the energy dissipation of the network.
481 - :ref:`Host Load <plugin_host_load>`: monitors the load of the compute units.
483 .. _options_modelchecking:
485 Configuring the Model-Checking
486 ------------------------------
488 To enable SimGrid's model-checking support, the program should
489 be executed using the simgrid-mc wrapper:
491 .. code-block:: shell
493 simgrid-mc ./my_program
495 Safety properties are expressed as assertions using the function
496 :cpp:func:`void MC_assert(int prop)`.
498 .. _cfg=smpi/buffering:
500 Specifying the MPI buffering behavior
501 .....................................
503 **Option** ``smpi/buffering`` **Default:** infty
505 Buffering in MPI has a huge impact on the communication semantic. For example,
506 standard blocking sends are synchronous calls when the system buffers are full
507 while these calls can complete immediately without even requiring a matching
508 receive call for small messages sent when the system buffers are empty.
510 In SMPI, this depends on the message size, that is compared against two thresholds:
512 - if (size < :ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>`) then
513 MPI_Send returns immediately, even if the corresponding receive has not be issued yet.
514 - if (:ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>` < size < :ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>`) then
515 MPI_Send returns as soon as the corresponding receive has been issued. This is known as the eager mode.
516 - if (:ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>` < size) then
517 MPI_Send returns only when the message has actually been sent over the network. This is known as the rendez-vous mode.
519 The ``smpi/buffering`` option gives an easier interface to choose between these semantics. It can take two values:
521 - **zero:** means that buffering should be disabled. Blocking communications are actually blocking.
522 - **infty:** means that buffering should be made infinite. Blocking communications are non-blocking.
524 .. _cfg=model-check/property:
526 Specifying a liveness property
527 ..............................
529 **Option** ``model-check/property`` **Default:** unset
531 If you want to specify liveness properties, you have to pass them on
532 the command line, specifying the name of the file containing the
533 property, as formatted by the `ltl2ba <https://github.com/utwente-fmt/ltl2ba>`_ program.
534 Note that ltl2ba is not part of SimGrid and must be installed separatly.
536 .. code-block:: shell
538 simgrid-mc ./my_program --cfg=model-check/property:<filename>
540 .. _cfg=model-check/checkpoint:
542 Going for Stateful Verification
543 ...............................
545 By default, the system is backtracked to its initial state to explore
546 another path, instead of backtracking to the exact step before the fork
547 that we want to explore (this is called stateless verification). This
548 is done this way because saving intermediate states can rapidly
549 exhaust the available memory. If you want, you can change the value of
550 the ``model-check/checkpoint`` item. For example,
551 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
552 step. Beware, this will certainly explode your memory. Larger values
553 are probably better, make sure to experiment a bit to find the right
554 setting for your specific system.
556 .. _cfg=model-check/reduction:
558 Specifying the kind of reduction
559 ................................
561 The main issue when using the model-checking is the state space
562 explosion. You can activate some reduction technique with
563 ``--cfg=model-check/reduction:<technique>``. For now, this
564 configuration variable can take 2 values:
566 - **none:** Do not apply any kind of reduction (mandatory for
567 liveness properties, as our current DPOR algorithm breaks cycles)
568 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if
569 you verify local safety properties (default value for safety
572 Another way to mitigate the state space explosion is to search for
573 cycles in the exploration with the :ref:`cfg=model-check/visited`
574 configuration. Note that DPOR and state-equality reduction may not
575 play well together. You should choose between them.
577 Our current DPOR implementation could be improved in may ways. We are
578 currently improving its efficiency (both in term of reduction ability
579 and computational speed), and future work could make it compatible
580 with liveness properties.
582 .. _cfg=model-check/visited:
584 Size of Cycle Detection Set (state equality reduction)
585 ......................................................
587 Mc SimGrid can be asked to search for cycles during the exploration,
588 i.e. situations where a new explored state is in fact the same state
589 than a previous one.. This can prove useful to mitigate the state
590 space explosion with safety properties, and this is the crux when
591 searching for counter-examples to the liveness properties.
593 Note that this feature may break the current implementation of the
594 DPOR reduction technique.
596 The ``model-check/visited`` item is the maximum number of states, which
597 are stored in memory. If the maximum number of snapshotted state is
598 reached, some states will be removed from the memory and some cycles
599 might be missed. Small values can lead to incorrect verifications, but
600 large values can exhaust your memory and be CPU intensive as each new
601 state must be compared to that amount of older saved states.
603 The default settings depend on the kind of exploration. With safety
604 checking, no state is snapshotted and cycles cannot be detected. With
605 liveness checking, all states are snapshotted because missing a cycle
606 could hinder the exploration soundness.
608 .. _cfg=model-check/termination:
610 Non-Termination Detection
611 .........................
613 The ``model-check/termination`` configuration item can be used to
614 report if a non-termination execution path has been found. This is a
615 path with a cycle, which means that the program might never terminate.
617 This only works in safety mode, not in liveness mode.
619 This options is disabled by default.
621 .. _cfg=model-check/dot-output:
626 If set, the ``model-check/dot-output`` configuration item is the name
627 of a file in which to write a dot file of the path leading to the
628 property violation discovered (safety or liveness violation), as well
629 as the cycle for liveness properties. This dot file can then be fed to the
630 graphviz dot tool to generate an corresponding graphical representation.
632 .. _cfg=model-check/max-depth:
634 Exploration Depth Limit
635 .......................
637 The ``model-checker/max-depth`` can set the maximum depth of the
638 exploration graph of the model checker. If this limit is reached, a
639 logging message is sent and the results might not be exact.
641 By default, there is no depth limit.
643 .. _cfg=model-check/timeout:
648 By default, the model checker does not handle timeout conditions: the `wait`
649 operations never time out. With the ``model-check/timeout`` configuration item
650 set to **yes**, the model checker will explore timeouts of `wait` operations.
652 .. _cfg=model-check/communications-determinism:
653 .. _cfg=model-check/send-determinism:
655 Communication Determinism
656 .........................
658 The ``model-check/communications-determinism`` and
659 ``model-check/send-determinism`` items can be used to select the
660 communication determinism mode of the model checker, which checks
661 determinism properties of the communications of an application.
665 Verification Performance Considerations
666 .......................................
668 The size of the stacks can have a huge impact on the memory
669 consumption when using model-checking. By default, each snapshot will
670 save a copy of the whole stacks and not only of the part that is
671 really meaningful: you should expect the contribution of the memory
672 consumption of the snapshots to be:
673 :math:`\text{number of processes} \times \text{stack size} \times \text{number of states}`.
675 When compiled against the model checker, the stacks are not
676 protected with guards: if the stack size is too small for your
677 application, the stack will silently overflow into other parts of the
678 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
680 .. _cfg=model-check/replay:
682 Replaying buggy execution paths from the model checker
683 ......................................................
685 Debugging the problems reported by the model checker is challenging:
686 First, the application under verification cannot be debugged with gdb
687 because the model checker already traces it. Then, the model checker may
688 explore several execution paths before encountering the issue, making it
689 very difficult to understand the output. Fortunately, SimGrid provides
690 the execution path leading to any reported issue so that you can replay
691 this path reported by the model checker, enabling the usage of classical
694 When the model checker finds an interesting path in the application
695 execution graph (where a safety or liveness property is violated), it
696 generates an identifier for this path. Here is an example of the output:
698 .. code-block:: shell
700 [ 0.000000] (0:@) Check a safety property
701 [ 0.000000] (0:@) **************************
702 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
703 [ 0.000000] (0:@) **************************
704 [ 0.000000] (0:@) Counter-example execution trace:
705 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
706 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
707 [ 0.000000] (0:@) Path = 1/3;1/4
708 [ 0.000000] (0:@) Expanded states = 27
709 [ 0.000000] (0:@) Visited states = 68
710 [ 0.000000] (0:@) Executed transitions = 46
712 The interesting line is ``Path = 1/3;1/4``, which means that you should use
713 ``--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
714 execution path. All options (but the model checker related ones) must
715 remain the same. In particular, if you ran your application with
716 ``smpirun -wrapper simgrid-mc``, then do it again. Remove all
717 MC-related options, keep non-MC-related ones and add
718 ``--cfg=model-check/replay:???``.
720 Currently, if the path is of the form ``X;Y;Z``, each number denotes
721 the actor's pid that is selected at each indecision point. If it's of
722 the form ``X/a;Y/b``, the X and Y are the selected pids while the a
723 and b are the return values of their simcalls. In the previouse
724 example, ``1/3;1/4``, you can see from the full output that the actor
725 1 is doing MC_RANDOM simcalls, so the 3 and 4 simply denote the values
726 that these simcall return.
728 Configuring the User Code Virtualization
729 ----------------------------------------
731 .. _cfg=contexts/factory:
733 Selecting the Virtualization Factory
734 ....................................
736 **Option** contexts/factory **Default:** "raw"
738 In SimGrid, the user code is virtualized in a specific mechanism that
739 allows the simulation kernel to control its execution: when a user
740 process requires a blocking action (such as sending a message), it is
741 interrupted, and only gets released when the simulated clock reaches
742 the point where the blocking operation is done. This is explained
743 graphically in the `relevant tutorial, available online
744 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
746 In SimGrid, the containers in which user processes are virtualized are
747 called contexts. Several context factory are provided, and you can
748 select the one you want to use with the ``contexts/factory``
749 configuration item. Some of the following may not exist on your
750 machine because of portability issues. In any case, the default one
751 should be the most effcient one (please report bugs if the
752 auto-detection fails for you). They are approximately sorted here from
753 the slowest to the most efficient:
755 - **thread:** very slow factory using full featured threads (either
756 pthreads or windows native threads). They are slow but very
757 standard. Some debuggers or profilers only work with this factory.
758 - **java:** Java applications are virtualized onto java threads (that
759 are regular pthreads registered to the JVM)
760 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
761 - **boost:** This uses the `context
762 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
763 of the boost library for a performance that is comparable to our
765 |br| Install the relevant library (e.g. with the
766 libboost-contexts-dev package on Debian/Ubuntu) and recompile
768 - **raw:** amazingly fast factory using a context switching mechanism
769 of our own, directly implemented in assembly (only available for x86
770 and amd64 platforms for now) and without any unneeded system call.
772 The main reason to change this setting is when the debugging tools become
773 fooled by the optimized context factories. Threads are the most
774 debugging-friendly contexts, as they allow one to set breakpoints
775 anywhere with gdb and visualize backtraces for all processes, in order
776 to debug concurrency issues. Valgrind is also more comfortable with
777 threads, but it should be usable with all factories (Exception: the
778 callgrind tool really dislikes raw and ucontext factories).
780 .. _cfg=contexts/stack-size:
782 Adapting the Stack Size
783 .......................
785 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
787 Each virtualized used process is executed using a specific system
788 stack. The size of this stack has a huge impact on the simulation
789 scalability, but its default value is rather large. This is because
790 the error messages that you get when the stack size is too small are
791 rather disturbing: this leads to stack overflow (overwriting other
792 stacks), leading to segfaults with corrupted stack traces.
794 If you want to push the scalability limits of your code, you might
795 want to reduce the ``contexts/stack-size`` item. Its default value is
796 8192 (in KiB), while our Chord simulation works with stacks as small
797 as 16 KiB, for example. You can ensure that some actors have a specific
798 size by simply changing the value of this configuration item before
799 creating these actors. The :cpp:func:`simgrid::s4u::Engine::set_config`
800 functions are handy for that.
802 This *setting is ignored* when using the thread factory (because there
803 is no way to modify the stack size with C++ system threads). Instead,
804 you should compile SimGrid and your application with
805 ``-fsplit-stack``. Note that this compilation flag is not compatible
806 with the model checker right now.
808 The operating system should only allocate memory for the pages of the
809 stack which are actually used and you might not need to use this in
810 most cases. However, this setting is very important when using the
811 model checker (see :ref:`options_mc_perf`).
813 .. _cfg=contexts/guard-size:
815 Disabling Stack Guard Pages
816 ...........................
818 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages on Windows or with MC)
820 Unless you use the threads context factory (see
821 :ref:`cfg=contexts/factory`), a stack guard page is usually used
822 which prevents the stack of a given actor from overflowing on another
823 stack. But the performance impact may become prohibitive when the
824 amount of actors increases. The option ``contexts/guard-size`` is the
825 number of stack guard pages used. By setting it to 0, no guard pages
826 will be used: in this case, you should avoid using small stacks (with
827 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
828 will silently overflow on other parts of the memory.
830 When no stack guard page is created, stacks may then silently overflow
831 on other parts of the memory if their size is too small for the
834 .. _cfg=contexts/nthreads:
835 .. _cfg=contexts/parallel-threshold:
836 .. _cfg=contexts/synchro:
838 Running User Code in Parallel
839 .............................
841 Parallel execution of the user code is only considered stable in
842 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
843 simulations may well fail in parallel mode. It is described in
844 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
846 If you are using the **ucontext** or **raw** context factories, you can
847 request to execute the user code in parallel. Several threads are
848 launched, each of them handling the same number of user contexts at each
849 run. To activate this, set the ``contexts/nthreads`` item to the amount
850 of cores that you have in your computer (or lower than 1 to have the
851 amount of cores auto-detected).
853 Even if you asked several worker threads using the previous option,
854 you can request to start the parallel execution (and pay the
855 associated synchronization costs) only if the potential parallelism is
856 large enough. For that, set the ``contexts/parallel-threshold``
857 item to the minimal amount of user contexts needed to start the
858 parallel execution. In any given simulation round, if that amount is
859 not reached, the contexts will be run sequentially directly by the
860 main thread (thus saving the synchronization costs). Note that this
861 option is mainly useful when the grain of the user code is very fine,
862 because our synchronization is now very efficient.
864 When parallel execution is activated, you can choose the
865 synchronization schema used with the ``contexts/synchro`` item,
866 which value is either:
868 - **futex:** ultra optimized synchronisation schema, based on futexes
869 (fast user-mode mutexes), and thus only available on Linux systems.
870 This is the default mode when available.
871 - **posix:** slow but portable synchronisation using only POSIX
873 - **busy_wait:** not really a synchronisation: the worker threads
874 constantly request new contexts to execute. It should be the most
875 efficient synchronisation schema, but it loads all the cores of
876 your machine for no good reason. You probably prefer the other less
879 Configuring the Tracing
880 -----------------------
882 The :ref:`tracing subsystem <outcomes_vizu>` can be configured in
883 several different ways depending on the nature of the simulator (MSG,
884 SimDag, SMPI) and the kind of traces that need to be obtained. See the
885 :ref:`Tracing Configuration Options subsection
886 <tracing_tracing_options>` to get a detailed description of each
887 configuration option.
889 We detail here a simple way to get the traces working for you, even if
890 you never used the tracing API.
893 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
895 .. code-block:: shell
897 --cfg=tracing:yes --cfg=tracing/uncategorized:yes
899 The first parameter activates the tracing subsystem, and the second
900 tells it to trace host and link utilization (without any
903 - MSG or SimDag-based simulator and categorized traces (you need to
904 declare categories and classify your tasks according to them)
906 .. code-block:: shell
908 --cfg=tracing:yes --cfg=tracing/categorized:yes
910 The first parameter activates the tracing subsystem, and the second
911 tells it to trace host and link categorized utilization.
913 - SMPI simulator and traces for a space/time view:
915 .. code-block:: shell
919 The `-trace` parameter for the smpirun script runs the simulation
920 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
921 smpirun's `-help` parameter for additional tracing options.
923 Sometimes you might want to put additional information on the trace to
924 correctly identify them later, or to provide data that can be used to
925 reproduce an experiment. You have two ways to do that:
927 - Add a string on top of the trace file as comment:
929 .. code-block:: shell
931 --cfg=tracing/comment:my_simulation_identifier
933 - Add the contents of a textual file on top of the trace file as comment:
935 .. code-block:: shell
937 --cfg=tracing/comment-file:my_file_with_additional_information.txt
939 Please, use these two parameters (for comments) to make reproducible
940 simulations. For additional details about this and all tracing
941 options, check See the :ref:`tracing_tracing_options`.
946 .. _cfg=msg/debug-multiple-use:
951 **Option** ``msg/debug-multiple-use`` **Default:** off
953 Sometimes your application may try to send a task that is still being
954 executed somewhere else, making it impossible to send this task. However,
955 for debugging purposes, one may want to know what the other host is/was
956 doing. This option shows a backtrace of the other process.
961 The SMPI interface provides several specific configuration items.
962 These are not easy to see, since the code is usually launched through the
963 ``smiprun`` script directly.
965 .. _cfg=smpi/host-speed:
966 .. _cfg=smpi/cpu-threshold:
967 .. _cfg=smpi/simulate-computation:
969 Automatic Benchmarking of SMPI Code
970 ...................................
972 In SMPI, the sequential code is automatically benchmarked, and these
973 computations are automatically reported to the simulator. That is to
974 say that if you have a large computation between a ``MPI_Recv()`` and
975 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
976 this code, and create an execution task within the simulator to take
977 this into account. For that, the actual duration is measured on the
978 host machine and then scaled to the power of the corresponding
979 simulated machine. The variable ``smpi/host-speed`` allows one to specify
980 the computational speed of the host machine (in flop/s) to use when
981 scaling the execution times. It defaults to 20000, but you really want
982 to adjust it to get accurate simulation results.
984 When the code consists of numerous consecutive MPI calls, the
985 previous mechanism feeds the simulation kernel with numerous tiny
986 computations. The ``smpi/cpu-threshold`` item becomes handy when this
987 impacts badly on the simulation performance. It specifies a threshold (in
988 seconds) below which the execution chunks are not reported to the
989 simulation kernel (default value: 1e-6).
991 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
992 time spent below this threshold. SMPI does not consider the
993 `amount of time` of these computations; there is no offset for
994 this. Hence, a value that is too small, may lead to unreliable
997 In some cases, however, one may wish to disable simulation of
998 the computation of an application. This is the case when SMPI is used not to
999 simulate an MPI application, but instead an MPI code that performs
1000 "live replay" of another MPI app (e.g., ScalaTrace's replay tool, or
1001 various on-line simulators that run an app at scale). In this case the
1002 computation of the replay/simulation logic should not be simulated by
1003 SMPI. Instead, the replay tool or on-line simulator will issue
1004 "computation events", which correspond to the actual MPI simulation
1005 being replayed/simulated. At the moment, these computation events can
1006 be simulated using SMPI by calling internal smpi_execute*() functions.
1008 To disable the benchmarking/simulation of a computation in the simulated
1009 application, the variable ``smpi/simulate-computation`` should be set
1010 to **no**. This option just ignores the timings in your simulation; it
1011 still executes the computations itself. If you want to stop SMPI from
1012 doing that, you should check the SMPI_SAMPLE macros, documented in
1013 Section :ref:`SMPI_use_faster`.
1015 +------------------------------------+-------------------------+-----------------------------+
1016 | Solution | Computations executed? | Computations simulated? |
1017 +====================================+=========================+=============================+
1018 | --cfg=smpi/simulate-computation:no | Yes | Never |
1019 +------------------------------------+-------------------------+-----------------------------+
1020 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
1021 +------------------------------------+-------------------------+-----------------------------+
1022 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
1023 +------------------------------------+-------------------------+-----------------------------+
1025 .. _cfg=smpi/comp-adjustment-file:
1027 Slow-down or speed-up parts of your code
1028 ........................................
1030 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
1032 This option allows you to pass a file that contains two columns: The
1033 first column defines the section that will be subject to a speedup;
1034 the second column is the speedup. For instance:
1036 .. code-block:: shell
1038 "start:stop","ratio"
1039 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1041 The first line is the header - you must include it. The following
1042 line means that the code between two consecutive MPI calls on line 30
1043 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1044 of 1.18244559422142. The value for the second column is therefore a
1045 speedup, if it is larger than 1 and a slowdown if it is smaller
1046 than 1. Nothing will be changed if it is equal to 1.
1048 Of course, you can set any arbitrary filenames you want (so the start
1049 and end don't have to be in the same file), but be aware that this
1050 mechanism only supports `consecutive calls!`
1052 Please note that you must pass the ``-trace-call-location`` flag to
1053 smpicc or smpiff, respectively. This flag activates some internal
1054 macro definitions that help with obtaining the call location.
1056 .. _cfg=smpi/bw-factor:
1061 **Option** ``smpi/bw-factor``
1062 |br| **Default:** 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
1064 The possible throughput of network links is often dependent on the
1065 message sizes, as protocols may adapt to different message sizes. With
1066 this option, a series of message sizes and factors are given, helping
1067 the simulation to be more realistic. For instance, the current default
1068 value means that messages with size 65472 and more will get a total of
1069 MAX_BANDWIDTH*0.940694, messages of size 15424 to 65471 will get
1070 MAX_BANDWIDTH*0.697866, and so on (where MAX_BANDWIDTH denotes the
1071 bandwidth of the link).
1073 An experimental script to compute these factors is available online. See
1074 https://framagit.org/simgrid/platform-calibration/
1075 https://simgrid.org/contrib/smpi-saturation-doc.html
1077 .. _cfg=smpi/display-timing:
1079 Reporting Simulation Time
1080 .........................
1082 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1084 Most of the time, you run MPI code with SMPI to compute the time it
1085 would take to run it on a platform. But since the code is run through
1086 the ``smpirun`` script, you don't have any control on the launcher
1087 code, making it difficult to report the simulated time when the
1088 simulation ends. If you enable the ``smpi/display-timing`` item,
1089 ``smpirun`` will display this information when the simulation
1092 .. _cfg=smpi/keep-temps:
1094 Keeping temporary files after simulation
1095 ........................................
1097 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1099 SMPI usually generates a lot of temporary files that are cleaned after
1100 use. This option requests to preserve them, for example to debug or
1101 profile your code. Indeed, the binary files are removed very early
1102 under the dlopen privatization schema, which tends to fool the
1105 .. _cfg=smpi/lat-factor:
1110 **Option** ``smpi/lat-factor`` |br|
1111 **default:** 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
1113 The motivation and syntax for this option is identical to the motivation/syntax
1114 of :ref:`cfg=smpi/bw-factor`.
1116 There is an important difference, though: While smpi/bw-factor `reduces` the
1117 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
1118 increase the latency, i.e., values larger than or equal to 1 are valid here.
1120 .. _cfg=smpi/papi-events:
1122 Trace hardware counters with PAPI
1123 .................................
1125 **Option** ``smpi/papi-events`` **default:** unset
1127 When the PAPI support is compiled into SimGrid, this option takes the
1128 names of PAPI counters and adds their respective values to the trace
1129 files (See Section :ref:`tracing_tracing_options`).
1133 This feature currently requires superuser privileges, as registers
1134 are queried. Only use this feature with code you trust! Call
1135 smpirun for instance via ``smpirun -wrapper "sudo "
1136 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1137 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1138 will not be required.
1140 It is planned to make this feature available on a per-process (or per-thread?) basis.
1141 The first draft, however, just implements a "global" (i.e., for all processes) set
1142 of counters, the "default" set.
1144 .. code-block:: shell
1146 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1148 .. _cfg=smpi/privatization:
1150 Automatic Privatization of Global Variables
1151 ...........................................
1153 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1155 MPI executables are usually meant to be executed in separate
1156 processes, but SMPI is executed in only one process. Global variables
1157 from executables will be placed in the same memory region and shared
1158 between processes, causing intricate bugs. Several options are
1159 possible to avoid this, as described in the main `SMPI publication
1160 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1161 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1162 automatically privatizing the globals, and this option allows one to
1163 choose between them.
1165 - **no** (default when not using smpirun): Do not automatically
1166 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1168 - **dlopen** or **yes** (default when using smpirun): Link multiple
1169 times against the binary.
1170 - **mmap** (slower, but maybe somewhat more stable):
1171 Runtime automatic switching of the data segments.
1174 This configuration option cannot be set in your platform file. You can only
1175 pass it as an argument to smpirun.
1177 .. _cfg=smpi/privatize-libs:
1179 Automatic privatization of global variables inside external libraries
1180 .....................................................................
1182 **Option** ``smpi/privatize-libs`` **default:** unset
1184 **Linux/BSD only:** When using dlopen (default) privatization,
1185 privatize specific shared libraries with internal global variables, if
1186 they can't be linked statically. For example libgfortran is usually
1187 used for Fortran I/O and indexes in files can be mixed up.
1189 Multiple libraries can be given, semicolon separated.
1191 This configuration option can only use either full paths to libraries,
1192 or full names. Check with ldd the name of the library you want to
1195 .. code-block:: shell
1199 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1202 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1203 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1204 but not ``libgfortran`` nor ``libgfortran.so``.
1206 .. _cfg=smpi/send-is-detached-thresh:
1208 Simulating MPI detached send
1209 ............................
1211 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1213 This threshold specifies the size in bytes under which the send will
1214 return immediately. This is different from the threshold detailed in
1215 :ref:`cfg=smpi/async-small-thresh` because the message is not
1216 really sent when the send is posted. SMPI still waits for the
1217 corresponding receive to be posted, in order to perform the communication
1220 .. _cfg=smpi/coll-selector:
1222 Simulating MPI collective algorithms
1223 ....................................
1225 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1227 SMPI implements more than 100 different algorithms for MPI collective
1228 communication, to accurately simulate the behavior of most of the
1229 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1230 select the decision logic either of the OpenMPI or the MPICH libraries. (By
1231 default SMPI uses naive version of collective operations.)
1233 Each collective operation can be manually selected with a
1234 ``smpi/collective_name:algo_name``. Available algorithms are listed in
1235 :ref:`SMPI_use_colls`.
1237 .. TODO:: All available collective algorithms will be made available
1238 via the ``smpirun --help-coll`` command.
1240 .. _cfg=smpi/iprobe:
1242 Inject constant times for MPI_Iprobe
1243 ....................................
1245 **Option** ``smpi/iprobe`` **default:** 0.0001
1247 The behavior and motivation for this configuration option is identical
1248 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1251 .. _cfg=smpi/iprobe-cpu-usage:
1253 Reduce speed for iprobe calls
1254 .............................
1256 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1258 MPI_Iprobe calls can be heavily used in applications. To account
1259 correctly for the energy that cores spend probing, it is necessary to
1260 reduce the load that these calls cause inside SimGrid.
1262 For instance, we measured a maximum power consumption of 220 W for a
1263 particular application but only 180 W while this application was
1264 probing. Hence, the correct factor that should be passed to this
1265 option would be 180/220 = 0.81.
1269 Inject constant times for MPI_Init
1270 ..................................
1272 **Option** ``smpi/init`` **default:** 0
1274 The behavior and motivation for this configuration option is identical
1275 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1279 Inject constant times for MPI_Isend()
1280 .....................................
1282 **Option** ``smpi/ois``
1284 The behavior and motivation for this configuration option is identical
1285 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1289 Inject constant times for MPI_send()
1290 ....................................
1292 **Option** ``smpi/os``
1294 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1295 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1296 time). SMPI can factor these costs in as well, but the user has to
1297 configure SMPI accordingly as these values may vary by machine. This
1298 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1299 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1300 exactly as ``smpi/ois``.
1302 This item can consist of multiple sections; each section takes three
1303 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1304 so this example contains two sections. Furthermore, each section
1305 consists of three values.
1307 1. The first value denotes the minimum size for this section to take effect;
1308 read it as "if message size is greater than this value (and other section has a larger
1309 first value that is also smaller than the message size), use this".
1310 In the first section above, this value is "1".
1312 2. The second value is the startup time; this is a constant value that will always
1313 be charged, no matter what the size of the message. In the first section above,
1316 3. The third value is the `per-byte` cost. That is, it is charged for every
1317 byte of the message (incurring cost messageSize*cost_per_byte)
1318 and hence accounts also for larger messages. In the first
1319 section of the example above, this value is "2".
1321 Now, SMPI always checks which section it should use for a given
1322 message; that is, if a message of size 11 is sent with the
1323 configuration of the example above, only the second section will be
1324 used, not the first, as the first value of the second section is
1325 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1326 message of size 11 incurs the following cost inside MPI_Send:
1327 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1329 Note that the order of sections can be arbitrary; they will be ordered internally.
1333 Inject constant times for MPI_Recv()
1334 ....................................
1336 **Option** ``smpi/or``
1338 The behavior and motivation for this configuration option is identical
1339 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1342 .. _cfg=smpi/grow-injected-times:
1344 Inject constant times for MPI_Test
1345 ..................................
1347 **Option** ``smpi/test`` **default:** 0.0001
1349 By setting this option, you can control the amount of time a process
1350 sleeps when MPI_Test() is called; this is important, because SimGrid
1351 normally only advances the time while communication is happening and
1352 thus, MPI_Test will not add to the time, resulting in deadlock if it is
1353 used as a break-condition as in the following example:
1358 MPI_Test(request, flag, status);
1362 To speed up execution, we use a counter to keep track of how often we
1363 checked if the handle is now valid or not. Hence, we actually
1364 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1365 process to sleep increases linearly with the number of previously
1366 failed tests. This behavior can be disabled by setting
1367 ``smpi/grow-injected-times`` to **no**. This will also disable this
1368 behavior for MPI_Iprobe.
1370 .. _cfg=smpi/shared-malloc:
1371 .. _cfg=smpi/shared-malloc-hugepage:
1376 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1378 If your simulation consumes too much memory, you may want to modify
1379 your code so that the working areas are shared by all MPI ranks. For
1380 example, in a block-cyclic matrix multiplication, you will only
1381 allocate one set of blocks, and all processes will share them.
1382 Naturally, this will lead to very wrong results, but this will save a
1383 lot of memory. So this is still desirable for some studies. For more on
1384 the motivation for that feature, please refer to the `relevant section
1385 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1386 of the SMPI CourseWare (see Activity #2.2 of the pointed
1387 assignment). In practice, change the calls for malloc() and free() into
1388 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1390 SMPI provides two algorithms for this feature. The first one, called
1391 ``local``, allocates one block per call to SMPI_SHARED_MALLOC()
1392 (each call site gets its own block) ,and this block is shared
1393 among all MPI ranks. This is implemented with the shm_* functions
1394 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1395 for each shared block.
1397 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1398 returns a new address, but it only points to a shadow block: its memory
1399 area is mapped on a 1 MiB file on disk. If the returned block is of size
1400 N MiB, then the same file is mapped N times to cover the whole bloc.
1401 At the end, no matter how many times you call SMPI_SHARED_MALLOC, this will
1402 only consume 1 MiB in memory.
1404 You can disable this behavior and come back to regular mallocs (for
1405 example for debugging purposes) using ``no`` as a value.
1407 If you want to keep private some parts of the buffer, for instance if these
1408 parts are used by the application logic and should not be corrupted, you
1409 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). For example:
1413 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1415 This will allocate 500 bytes to mem, such that mem[27..41] and
1416 mem[100..199] are shared while other area remain private.
1418 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1420 When smpi/shared-malloc:global is used, the memory consumption problem
1421 is solved, but it may induce too much load on the kernel's pages table.
1422 In this case, you should use huge pages so that the kernel creates only one
1423 entry per MB of malloced data instead of one entry per 4 kB.
1424 To activate this, you must mount a hugetlbfs on your system and allocate
1425 at least one huge page:
1427 .. code-block:: shell
1430 sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1431 sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1433 Then, you can pass the option
1434 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1435 actually activate the huge page support in shared mallocs.
1439 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1440 ...................................................................
1442 **Option** ``smpi/wtime`` **default:** 10 ns
1444 This option controls the amount of (simulated) time spent in calls to
1445 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1446 to 0, the simulated clock is not advanced in these calls, which leads
1447 to issues if your application contains such a loop:
1451 while(MPI_Wtime() < some_time_bound) {
1452 /* some tests, with no communication nor computation */
1455 When the option smpi/wtime is set to 0, the time advances only on
1456 communications and computations. So the previous code results in an
1457 infinite loop: the current [simulated] time will never reach
1458 ``some_time_bound``. This infinite loop is avoided when that option
1459 is set to a small value, as it is by default since SimGrid v3.21.
1461 Note that if your application does not contain any loop depending on
1462 the current time only, then setting this option to a non-zero value
1463 will slow down your simulations by a tiny bit: the simulation loop has
1464 to be broken out of and reset each time your code asks for the current time.
1465 If the simulation speed really matters to you, you can avoid this
1466 extra delay by setting smpi/wtime to 0.
1468 Other Configurations
1469 --------------------
1471 .. _cfg=debug/clean-atexit:
1473 Cleanup at Termination
1474 ......................
1476 **Option** ``debug/clean-atexit`` **default:** on
1478 If your code is segfaulting during its finalization, it may help to
1479 disable this option to request that SimGrid not attempt any cleanups at
1480 the end of the simulation. Since the Unix process is ending anyway,
1481 the operating system will wipe it all.
1488 **Option** ``path`` **default:** . (current dir)
1490 It is possible to specify a list of directories to search in for the
1491 trace files (see :ref:`pf_trace`) by using this configuration
1492 item. To add several directory to the path, set the configuration
1493 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1495 .. _cfg=debug/breakpoint:
1500 **Option** ``debug/breakpoint`` **default:** unset
1502 This configuration option sets a breakpoint: when the simulated clock
1503 reaches the given time, a SIGTRAP is raised. This can be used to stop
1504 the execution and get a backtrace with a debugger.
1506 It is also possible to set the breakpoint from inside the debugger, by
1507 writing in global variable simgrid::simix::breakpoint. For example,
1510 .. code-block:: shell
1512 set variable simgrid::simix::breakpoint = 3.1416
1514 .. _cfg=debug/verbose-exit:
1519 **Option** ``debug/verbose-exit`` **default:** on
1521 By default, when Ctrl-C is pressed, the status of all existing actors
1522 is displayed before exiting the simulation. This is very useful to
1523 debug your code, but it can become troublesome if you have many
1524 actors. Set this configuration item to **off** to disable this
1527 .. _cfg=exception/cutpath:
1529 Truncate local path from exception backtrace
1530 ............................................
1532 **Option** ``exception/cutpath`` **default:** off
1534 This configuration option is used to remove the path from the
1535 backtrace shown when an exception is thrown. This is mainly useful for
1536 the tests: the full file path would makes the tests non-reproducible because
1537 the paths of source files depend of the build settings. That would
1538 break most of the tests since their output is continually compared.
1540 Logging Configuration
1541 ---------------------
1543 This can be done by using XBT. Go to :ref:`XBT_log` for more details.