8 <object id="TOC" data="graphical-toc.svg" width="100%" 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 ').
44 Another solution is to use the ``<config>`` tag in the platform file. The
45 only restriction is that this tag must occure 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 - **clean-atexit:** :ref:`cfg=clean-atexit`
89 - **contexts/factory:** :ref:`cfg=contexts/factory`
90 - **contexts/guard-size:** :ref:`cfg=contexts/guard-size`
91 - **contexts/nthreads:** :ref:`cfg=contexts/nthreads`
92 - **contexts/parallel-threshold:** :ref:`cfg=contexts/parallel-threshold`
93 - **contexts/stack-size:** :ref:`cfg=contexts/stack-size`
94 - **contexts/synchro:** :ref:`cfg=contexts/synchro`
96 - **cpu/maxmin-selective-update:** :ref:`Cpu Optimization Level <options_model_optim>`
97 - **cpu/model:** :ref:`options_model_select`
98 - **cpu/optim:** :ref:`Cpu Optimization Level <options_model_optim>`
100 - **exception/cutpath:** :ref:`cfg=exception/cutpath`
102 - **host/model:** :ref:`options_model_select`
104 - **maxmin/precision:** :ref:`cfg=maxmin/precision`
105 - **maxmin/concurrency-limit:** :ref:`cfg=maxmin/concurrency-limit`
107 - **msg/debug-multiple-use:** :ref:`cfg=msg/debug-multiple-use`
109 - **model-check:** :ref:`options_modelchecking`
110 - **model-check/checkpoint:** :ref:`cfg=model-check/checkpoint`
111 - **model-check/communications-determinism:** :ref:`cfg=model-check/communications-determinism`
112 - **model-check/dot-output:** :ref:`cfg=model-check/dot-output`
113 - **model-check/hash:** :ref:`cfg=model-checker/hash`
114 - **model-check/max-depth:** :ref:`cfg=model-check/max-depth`
115 - **model-check/property:** :ref:`cfg=model-check/property`
116 - **model-check/reduction:** :ref:`cfg=model-check/reduction`
117 - **model-check/replay:** :ref:`cfg=model-check/replay`
118 - **model-check/send-determinism:** :ref:`cfg=model-check/send-determinism`
119 - **model-check/sparse-checkpoint:** :ref:`cfg=model-check/sparse-checkpoint`
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/maxmin-selective-update:** :ref:`Network Optimization Level <options_model_optim>`
128 - **network/model:** :ref:`options_model_select`
129 - **network/optim:** :ref:`Network Optimization Level <options_model_optim>`
130 - **network/TCP-gamma:** :ref:`cfg=network/TCP-gamma`
131 - **network/weight-S:** :ref:`cfg=network/weight-S`
133 - **ns3/TcpModel:** :ref:`options_pls`
134 - **path:** :ref:`cfg=path`
135 - **plugin:** :ref:`cfg=plugin`
137 - **simix/breakpoint:** :ref:`cfg=simix/breakpoint`
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/bw-factor:** :ref:`cfg=smpi/bw-factor`
146 - **smpi/coll-selector:** :ref:`cfg=smpi/coll-selector`
147 - **smpi/comp-adjustment-file:** :ref:`cfg=smpi/comp-adjustment-file`
148 - **smpi/cpu-threshold:** :ref:`cfg=smpi/cpu-threshold`
149 - **smpi/display-timing:** :ref:`cfg=smpi/display-timing`
150 - **smpi/grow-injected-times:** :ref:`cfg=smpi/grow-injected-times`
151 - **smpi/host-speed:** :ref:`cfg=smpi/host-speed`
152 - **smpi/IB-penalty-factors:** :ref:`cfg=smpi/IB-penalty-factors`
153 - **smpi/iprobe:** :ref:`cfg=smpi/iprobe`
154 - **smpi/iprobe-cpu-usage:** :ref:`cfg=smpi/iprobe-cpu-usage`
155 - **smpi/init:** :ref:`cfg=smpi/init`
156 - **smpi/keep-temps:** :ref:`cfg=smpi/keep-temps`
157 - **smpi/lat-factor:** :ref:`cfg=smpi/lat-factor`
158 - **smpi/ois:** :ref:`cfg=smpi/ois`
159 - **smpi/or:** :ref:`cfg=smpi/or`
160 - **smpi/os:** :ref:`cfg=smpi/os`
161 - **smpi/papi-events:** :ref:`cfg=smpi/papi-events`
162 - **smpi/privatization:** :ref:`cfg=smpi/privatization`
163 - **smpi/privatize-libs:** :ref:`cfg=smpi/privatize-libs`
164 - **smpi/send-is-detached-thresh:** :ref:`cfg=smpi/send-is-detached-thresh`
165 - **smpi/shared-malloc:** :ref:`cfg=smpi/shared-malloc`
166 - **smpi/shared-malloc-hugepage:** :ref:`cfg=smpi/shared-malloc-hugepage`
167 - **smpi/simulate-computation:** :ref:`cfg=smpi/simulate-computation`
168 - **smpi/test:** :ref:`cfg=smpi/test`
169 - **smpi/wtime:** :ref:`cfg=smpi/wtime`
171 - **Tracing configuration options** can be found in Section :ref:`tracing_tracing_options`
173 - **storage/model:** :ref:`options_model_select`
174 - **verbose-exit:** :ref:`cfg=verbose-exit`
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 storage 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 <ftp://ftp.ens-lyon.fr/pub/LIP/Rapports/RR/RR2002/RR2002-40.ps.gz>`_.
222 - **Reno/Reno2/Vegas:** Models from Steven H. Low using lagrange_solve instead of
223 lmm_solve (experts only; check the code for more info).
224 - **NS3** (only available if you compiled SimGrid accordingly):
225 Use the packet-level network
226 simulators as network models (see :ref:`pls_ns3`).
227 This model can be :ref:`further configured <options_pls>`.
229 - ``cpu/model``: specify the used CPU model. We have only one model
232 - **Cas01:** Simplistic CPU model (time=size/power)
234 - ``host/model``: The host concept is the aggregation of a CPU with a
235 network card. Three models exists, but actually, only 2 of them are
236 interesting. The "compound" one is simply due to the way our
237 internal code is organized, and can easily be ignored. So at the
238 end, you have two host models: The default one allows to aggregate
239 an existing CPU model with an existing network model, but does not
240 allow parallel tasks because these beasts need some collaboration
241 between the network and CPU model. That is why, ptask_07 is used by
242 default when using SimDag.
244 - **default:** Default host model. Currently, CPU:Cas01 and
245 network:LV08 (with cross traffic enabled)
246 - **compound:** Host model that is automatically chosen if
247 you change the network and CPU models
248 - **ptask_L07:** Host model somehow similar to Cas01+CM02 but
249 allowing "parallel tasks", that are intended to model the moldable
250 tasks of the grid scheduling literature.
252 - ``storage/model``: specify the used storage model. Only one model is
254 - ``vm/model``: specify the model for virtual machines. Only one model
257 .. todo: make 'compound' the default host model.
259 .. _options_model_optim:
264 The network and CPU models that are based on lmm_solve (that
265 is, all our analytical models) accept specific optimization
268 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
270 - **Lazy:** Lazy action management (partial invalidation in lmm +
271 heap in action remaining).
272 - **TI:** Trace integration. Highly optimized mode when using
273 availability traces (only available for the Cas01 CPU model for
275 - **Full:** Full update of remaining and variables. Slow but may be
276 useful when debugging.
278 - items ``network/maxmin-selective-update`` and
279 ``cpu/maxmin-selective-update``: configure whether the underlying
280 should be lazily updated or not. It should have no impact on the
281 computed timings, but should speed up the computation. |br| It is
282 still possible to disable this feature because it can reveal
283 counter-productive in very specific scenarios where the
284 interaction level is high. In particular, if all your
285 communication share a given backbone link, you should disable it:
286 without it, a simple regular loop is used to update each
287 communication. With it, each of them is still updated (because of
288 the dependency induced by the backbone), but through a complicated
289 and slow pattern that follows the actual dependencies.
291 .. _cfg=maxmin/precision:
292 .. _cfg=surf/precision:
297 **Option** ``maxmin/precision`` **Default:** 0.00001 (in flops or bytes) |br|
298 **Option** ``surf/precision`` **Default:** 0.00001 (in seconds)
300 The analytical models handle a lot of floating point values. It is
301 possible to change the epsilon used to update and compare them through
302 this configuration item. Changing it may speedup the simulation by
303 discarding very small actions, at the price of a reduced numerical
304 precision. You can modify separately the precision used to manipulate
305 timings (in seconds) and the one used to manipulate amounts of work
308 .. _cfg=maxmin/concurrency-limit:
313 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
315 The maximum number of variables per resource can be tuned through this
316 option. You can have as many simultaneous actions per resources as you
317 want. If your simulation presents a very high level of concurrency, it
318 may help to use e.g. 100 as a value here. It means that at most 100
319 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
321 resource 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 .. _options_model_network:
329 Configuring the Network Model
330 .............................
332 .. _cfg=network/TCP-gamma:
334 Maximal TCP Window Size
335 ^^^^^^^^^^^^^^^^^^^^^^^
337 **Option** ``network/TCP-gamma`` **Default:** 4194304
339 The analytical models need to know the maximal TCP window size to take
340 the TCP congestion mechanism into account. On Linux, this value can
341 be retrieved using the following commands. Both give a set of values,
342 and you should use the last one, which is the maximal size.
344 .. code-block:: shell
346 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
347 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
349 .. _cfg=smpi/IB-penalty-factors:
350 .. _cfg=network/bandwidth-factor:
351 .. _cfg=network/latency-factor:
352 .. _cfg=network/weight-S:
354 Correcting Important Network Parameters
355 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
357 SimGrid can take network irregularities such as a slow startup or
358 changing behavior depending on the message size into account. You
359 should not change these values unless you really know what you're
360 doing. The corresponding values were computed through data fitting
361 one the timings of packet-level simulators, as described in `Accuracy
362 Study and Improvement of Network Simulation in the SimGrid Framework
363 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
366 If you are using the SMPI model, these correction coefficients are
367 themselves corrected by constant values depending on the size of the
368 exchange. By default SMPI uses factors computed on the Stampede
369 Supercomputer at TACC, with optimal deployment of processes on
370 nodes. Again, only hardcore experts should bother about this fact.
372 InfiniBand network behavior can be modeled through 3 parameters
373 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `this PhD
375 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
377 .. todo:: This section should be rewritten, and actually explain the
378 options network/bandwidth-factor, network/latency-factor,
381 .. _cfg=network/crosstraffic:
383 Simulating Cross-Traffic
384 ^^^^^^^^^^^^^^^^^^^^^^^^
386 Since SimGrid v3.7, cross-traffic effects can be taken into account in
387 analytical simulations. It means that ongoing and incoming
388 communication flows are treated independently. In addition, the LV08
389 model adds 0.05 of usage on the opposite direction for each new
390 created flow. This can be useful to simulate some important TCP
391 phenomena such as ack compression.
393 For that to work, your platform must have two links for each
394 pair of interconnected hosts. An example of usable platform is
395 available in ``examples/platforms/crosstraffic.xml``.
397 This is activated through the ``network/crosstraffic`` item, that
398 can be set to 0 (disable this feature) or 1 (enable it).
400 Note that with the default host model this option is activated by default.
402 .. _cfg=smpi/async-small-thresh:
404 Simulating Asyncronous Send
405 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
407 (this configuration item is experimental and may change or disapear)
409 It is possible to specify that messages below a certain size will be
410 sent as soon as the call to MPI_Send is issued, without waiting for
411 the correspondant receive. This threshold can be configured through
412 the ``smpi/async-small-thresh`` item. The default value is 0. This
413 behavior can also be manually set for mailboxes, by setting the
414 receiving mode of the mailbox with a call to
415 :cpp:func:`MSG_mailbox_set_async`. After this, all messages sent to
416 this mailbox will have this behavior regardless of the message size.
418 This value needs to be smaller than or equals to the threshold set at
419 @ref options_model_smpi_detached , because asynchronous messages are
420 meant to be detached as well.
427 **Option** ``ns3/TcpModel`` **Default:** "default" (NS3 default)
429 When using NS3, there is an extra item ``ns3/TcpModel``, corresponding
430 to the ``ns3::TcpL4Protocol::SocketType`` configuration item in
431 NS3. The only valid values (enforced on the SimGrid side) are
432 'default' (no change to the NS3 configuration), 'NewReno' or 'Reno' or
435 Configuring the Storage model
436 .............................
438 .. _cfg=storage/max_file_descriptors:
440 File Descriptor Cound per Host
441 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
443 **Option** ``storage/max_file_descriptors`` **Default:** 1024
445 Each host maintains a fixed-size array of its file descriptors. You
446 can change its size through this item to either enlarge it if your
447 application requires it or to reduce it to save memory space.
454 SimGrid plugins allow to extend the framework without changing its
455 source code directly. Read the source code of the existing plugins to
456 learn how to do so (in ``src/plugins``), and ask your questions to the
457 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
458 that plugins usually register callbacks to some signals of interest.
459 If they need to store some information about a given object (Link, CPU
460 or Actor), they do so through the use of a dedicated object extension.
462 Some of the existing plugins can be activated from the command line,
463 meaning that you can activate them from the command line without any
464 modification to your simulation code. For example, you can activate
465 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
468 Here is the full list of plugins that can be activated this way:
470 - **host_energy:** keeps track of the energy dissipated by
471 computations. More details in @ref plugin_energy.
472 - **link_energy:** keeps track of the energy dissipated by
473 communications. More details in @ref SURF_plugin_energy.
474 - **host_load:** keeps track of the computational load.
475 More details in @ref plugin_load.
477 .. _options_modelchecking:
479 Configuring the Model-Checking
480 ------------------------------
482 To enable the SimGrid model-checking support the program should
483 be executed using the simgrid-mc wrapper:
485 .. code-block:: shell
487 simgrid-mc ./my_program
489 Safety properties are expressed as assertions using the function
490 :cpp:func:`void MC_assert(int prop)`.
492 .. _cfg=model-check/property:
494 Specifying a liveness property
495 ..............................
497 **Option** ``model-check/property`` **Default:** unset
499 If you want to specify liveness properties, you have to pass them on
500 the command line, specifying the name of the file containing the
501 property, as formatted by the ltl2ba program.
504 .. code-block:: shell
506 simgrid-mc ./my_program --cfg=model-check/property:<filename>
508 .. _cfg=model-check/checkpoint:
510 Going for Stateful Verification
511 ...............................
513 By default, the system is backtracked to its initial state to explore
514 another path instead of backtracking to the exact step before the fork
515 that we want to explore (this is called stateless verification). This
516 is done this way because saving intermediate states can rapidly
517 exhaust the available memory. If you want, you can change the value of
518 the ``model-check/checkpoint`` item. For example,
519 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
520 step. Beware, this will certainly explode your memory. Larger values
521 are probably better, make sure to experiment a bit to find the right
522 setting for your specific system.
524 .. _cfg=model-check/reduction:
526 Specifying the kind of reduction
527 ................................
529 The main issue when using the model-checking is the state space
530 explosion. To counter that problem, you can chose a exploration
531 reduction techniques with
532 ``--cfg=model-check/reduction:<technique>``. For now, this
533 configuration variable can take 2 values:
535 - **none:** Do not apply any kind of reduction (mandatory for now for
537 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if
538 you verify local safety properties (default value for safety
541 There is unfortunately no silver bullet here, and the most efficient
542 reduction techniques cannot be applied to any properties. In
543 particular, the DPOR method cannot be applied on liveness properties
544 since our implementation of DPOR may break some cycles, while cycles
545 are very important to the soundness of the exploration for liveness
548 .. _cfg=model-check/visited:
550 Size of Cycle Detection Set
551 ...........................
553 In order to detect cycles, the model-checker needs to check if a new
554 explored state is in fact the same state than a previous one. For
555 that, the model-checker can take a snapshot of each visited state:
556 this snapshot is then used to compare it with subsequent states in the
559 The ``model-check/visited`` item is the maximum number of states which
560 are stored in memory. If the maximum number of snapshotted state is
561 reached, some states will be removed from the memory and some cycles
562 might be missed. Small values can lead to incorrect verifications, but
563 large value can exhaust your memory, so choose carefully.
565 By default, no state is snapshotted and cycles cannot be detected.
567 .. _cfg=model-check/termination:
569 Non-Termination Detection
570 .........................
572 The ``model-check/termination`` configuration item can be used to
573 report if a non-termination execution path has been found. This is a
574 path with a cycle which means that the program might never terminate.
576 This only works in safety mode, not in liveness mode.
578 This options is disabled by default.
580 .. _cfg=model-check/dot-output:
585 If set, the ``model-check/dot-output`` configuration item is the name
586 of a file in which to write a dot file of the path leading the found
587 property (safety or liveness violation) as well as the cycle for
588 liveness properties. This dot file can then fed to the graphviz dot
589 tool to generate an corresponding graphical representation.
591 .. _cfg=model-check/max-depth:
593 Exploration Depth Limit
594 .......................
596 The ``model-checker/max-depth`` can set the maximum depth of the
597 exploration graph of the model-checker. If this limit is reached, a
598 logging message is sent and the results might not be exact.
600 By default, there is not depth limit.
602 .. _cfg=model-check/timeout:
607 By default, the model-checker does not handle timeout conditions: the `wait`
608 operations never time out. With the ``model-check/timeout`` configuration item
609 set to **yes**, the model-checker will explore timeouts of `wait` operations.
611 .. _cfg=model-check/communications-determinism:
612 .. _cfg=model-check/send-determinism:
614 Communication Determinism
615 .........................
617 The ``model-check/communications-determinism`` and
618 ``model-check/send-determinism`` items can be used to select the
619 communication determinism mode of the model-checker which checks
620 determinism properties of the communications of an application.
622 .. _cfg=model-check/sparse-checkpoint:
624 Incremental Checkpoints
625 .......................
627 When the model-checker is configured to take a snapshot of each
628 explored state (with the ``model-checker/visited`` item), the memory
629 consumption can rapidly reach GiB ou Tib of memory. However, for many
630 workloads, the memory does not change much between different snapshots
631 and taking a complete copy of each snapshot is a waste of memory.
633 The ``model-check/sparse-checkpoint`` option item can be set to
634 **yes** to avoid making a complete copy of each snapshot. Instead,
635 each snapshot will be decomposed in blocks which will be stored
636 separately. If multiple snapshots share the same block (or if the
637 same block is used in the same snapshot), the same copy of the block
638 will be shared leading to a reduction of the memory footprint.
640 For many applications, this option considerably reduces the memory
641 consumption. In somes cases, the model-checker might be slightly
642 slower because of the time taken to manage the metadata about the
643 blocks. In other cases however, this snapshotting strategy will be
644 much faster by reducing the cache consumption. When the memory
645 consumption is important, by avoiding to hit the swap or reducing the
646 swap usage, this option might be much faster than the basic
647 snapshotting strategy.
649 This option is currently disabled by default.
651 Verification Performance Considerations
652 .......................................
654 The size of the stacks can have a huge impact on the memory
655 consumption when using model-checking. By default, each snapshot will
656 save a copy of the whole stacks and not only of the part which is
657 really meaningful: you should expect the contribution of the memory
658 consumption of the snapshots to be @f$ @mbox{number of processes}
659 @times @mbox{stack size} @times @mbox{number of states} @f$.
661 The ``model-check/sparse-checkpoint`` can be used to reduce the memory
662 consumption by trying to share memory between the different snapshots.
664 When compiled against the model checker, the stacks are not
665 protected with guards: if the stack size is too small for your
666 application, the stack will silently overflow on other parts of the
667 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
669 .. _cfg=model-checker/hash:
674 Usually most of the time of the model-checker is spent comparing states. This
675 process is complicated and consumes a lot of bandwidth and cache.
676 In order to speedup the state comparison, the experimental ``model-checker/hash``
677 configuration item enables the computation of a hash summarizing as much
678 information of the state as possible into a single value. This hash can be used
679 to avoid most of the comparisons: the costly comparison is then only used when
680 the hashes are identical.
682 Currently most of the state is not included in the hash because the
683 implementation was found to be buggy and this options is not as useful as
684 it could be. For this reason, it is currently disabled by default.
686 .. _cfg=model-check/replay:
688 Replaying buggy execution paths out of the model-checker
689 ........................................................
691 Debugging the problems reported by the model-checker is challenging: First, the
692 application under verification cannot be debugged with gdb because the
693 model-checker already traces it. Then, the model-checker may explore several
694 execution paths before encountering the issue, making it very difficult to
695 understand the outputs. Fortunately, SimGrid provides the execution path leading
696 to any reported issue so that you can replay this path out of the model checker,
697 enabling the usage of classical debugging tools.
699 When the model-checker finds an interesting path in the application
700 execution graph (where a safety or liveness property is violated), it
701 generates an identifier for this path. Here is an example of output:
703 .. code-block:: shell
705 [ 0.000000] (0:@) Check a safety property
706 [ 0.000000] (0:@) **************************
707 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
708 [ 0.000000] (0:@) **************************
709 [ 0.000000] (0:@) Counter-example execution trace:
710 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
711 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
712 [ 0.000000] (0:@) Path = 1/3;1/4
713 [ 0.000000] (0:@) Expanded states = 27
714 [ 0.000000] (0:@) Visited states = 68
715 [ 0.000000] (0:@) Executed transitions = 46
717 The interesting line is ``Path = 1/3;1/4``, which means that you should use
718 `--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
719 execution path. The other options should be the same (but the model-checker
720 should be disabled). Note that format and meaning of the path may change between
723 Configuring the User Code Virtualization
724 ----------------------------------------
726 .. _cfg=contexts/factory:
728 Selecting the Virtualization Factory
729 ....................................
731 **Option** contexts/factory **Default:** "raw"
733 In SimGrid, the user code is virtualized in a specific mechanism that
734 allows the simulation kernel to control its execution: when a user
735 process requires a blocking action (such as sending a message), it is
736 interrupted, and only gets released when the simulated clock reaches
737 the point where the blocking operation is done. This is explained
738 graphically in the `relevant tutorial, available online
739 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
741 In SimGrid, the containers in which user processes are virtualized are
742 called contexts. Several context factory are provided, and you can
743 select the one you want to use with the ``contexts/factory``
744 configuration item. Some of the following may not exist on your
745 machine because of portability issues. In any case, the default one
746 should be the most effcient one (please report bugs if the
747 auto-detection fails for you). They are approximately sorted here from
748 the slowest to the most efficient:
750 - **thread:** very slow factory using full featured threads (either
751 pthreads or windows native threads). They are slow but very
752 standard. Some debuggers or profilers only work with this factory.
753 - **java:** Java applications are virtualized onto java threads (that
754 are regular pthreads registered to the JVM)
755 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
756 - **boost:** This uses the `context
757 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
758 of the boost library for a performance that is comparable to our
760 |br| Install the relevant library (e.g. with the
761 libboost-contexts-dev package on Debian/Ubuntu) and recompile
763 - **raw:** amazingly fast factory using a context switching mechanism
764 of our own, directly implemented in assembly (only available for x86
765 and amd64 platforms for now) and without any unneeded system call.
767 The main reason to change this setting is when the debugging tools get
768 fooled by the optimized context factories. Threads are the most
769 debugging-friendly contextes, as they allow to set breakpoints
770 anywhere with gdb and visualize backtraces for all processes, in order
771 to debug concurrency issues. Valgrind is also more comfortable with
772 threads, but it should be usable with all factories (Exception: the
773 callgrind tool really dislikes raw and ucontext factories).
775 .. _cfg=contexts/stack-size:
777 Adapting the Stack Size
778 .......................
780 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
782 Each virtualized used process is executed using a specific system
783 stack. The size of this stack has a huge impact on the simulation
784 scalability, but its default value is rather large. This is because
785 the error messages that you get when the stack size is too small are
786 rather disturbing: this leads to stack overflow (overwriting other
787 stacks), leading to segfaults with corrupted stack traces.
789 If you want to push the scalability limits of your code, you might
790 want to reduce the ``contexts/stack-size`` item. Its default value is
791 8192 (in KiB), while our Chord simulation works with stacks as small
792 as 16 KiB, for example. This *setting is ignored* when using the
793 thread factory. Instead, you should compile SimGrid and your
794 application with ``-fsplit-stack``. Note that this compilation flag is
795 not compatible with the model-checker right now.
797 The operating system should only allocate memory for the pages of the
798 stack which are actually used and you might not need to use this in
799 most cases. However, this setting is very important when using the
800 model checker (see :ref:`options_mc_perf`).
802 .. _cfg=contexts/guard-size:
804 Disabling Stack Guard Pages
805 ...........................
807 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages on Windows or with MC)
809 Unless you use the threads context factory (see
810 :ref:`cfg=contexts/factory`), a stack guard page is usually used
811 which prevents the stack of a given actor from overflowing on another
812 stack. But the performance impact may become prohibitive when the
813 amount of actors increases. The option ``contexts/guard-size`` is the
814 number of stack guard pages used. By setting it to 0, no guard pages
815 will be used: in this case, you should avoid using small stacks (with
816 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
817 will silently overflow on other parts of the memory.
819 When no stack guard page is created, stacks may then silently overflow
820 on other parts of the memory if their size is too small for the
823 .. _cfg=contexts/nthreads:
824 .. _cfg=contexts/parallel-threshold:
825 .. _cfg=contexts/synchro:
827 Running User Code in Parallel
828 .............................
830 Parallel execution of the user code is only considered stable in
831 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
832 simulations may well fail in parallel mode. It is described in
833 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
835 If you are using the **ucontext** or **raw** context factories, you can
836 request to execute the user code in parallel. Several threads are
837 launched, each of them handling as much user contexts at each run. To
838 actiave this, set the ``contexts/nthreads`` item to the amount of
839 cores that you have in your computer (or lower than 1 to have
840 the amount of cores auto-detected).
842 Even if you asked several worker threads using the previous option,
843 you can request to start the parallel execution (and pay the
844 associated synchronization costs) only if the potential parallelism is
845 large enough. For that, set the ``contexts/parallel-threshold``
846 item to the minimal amount of user contexts needed to start the
847 parallel execution. In any given simulation round, if that amount is
848 not reached, the contexts will be run sequentially directly by the
849 main thread (thus saving the synchronization costs). Note that this
850 option is mainly useful when the grain of the user code is very fine,
851 because our synchronization is now very efficient.
853 When parallel execution is activated, you can choose the
854 synchronization schema used with the ``contexts/synchro`` item,
855 which value is either:
857 - **futex:** ultra optimized synchronisation schema, based on futexes
858 (fast user-mode mutexes), and thus only available on Linux systems.
859 This is the default mode when available.
860 - **posix:** slow but portable synchronisation using only POSIX
862 - **busy_wait:** not really a synchronisation: the worker threads
863 constantly request new contexts to execute. It should be the most
864 efficient synchronisation schema, but it loads all the cores of
865 your machine for no good reason. You probably prefer the other less
868 Configuring the Tracing
869 -----------------------
871 The :ref:`tracing subsystem <outcomes_vizu>` can be configured in
872 several different ways depending on the nature of the simulator (MSG,
873 SimDag, SMPI) and the kind of traces that need to be obtained. See the
874 :ref:`Tracing Configuration Options subsection
875 <tracing_tracing_options>` to get a detailed description of each
876 configuration option.
878 We detail here a simple way to get the traces working for you, even if
879 you never used the tracing API.
882 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
884 .. code-block:: shell
886 --cfg=tracing:yes --cfg=tracing/uncategorized:yes --cfg=triva/uncategorized:uncat.plist
888 The first parameter activates the tracing subsystem, the second
889 tells it to trace host and link utilization (without any
890 categorization) and the third creates a graph configuration file to
891 configure Triva when analysing the resulting trace file.
893 - MSG or SimDag-based simulator and categorized traces (you need to
894 declare categories and classify your tasks according to them)
896 .. code-block:: shell
898 --cfg=tracing:yes --cfg=tracing/categorized:yes --cfg=triva/categorized:cat.plist
900 The first parameter activates the tracing subsystem, the second
901 tells it to trace host and link categorized utilization and the
902 third creates a graph configuration file to configure Triva when
903 analysing the resulting trace file.
905 - SMPI simulator and traces for a space/time view:
907 .. code-block:: shell
911 The `-trace` parameter for the smpirun script runs the simulation
912 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
913 smpirun's `-help` parameter for additional tracing options.
915 Sometimes you might want to put additional information on the trace to
916 correctly identify them later, or to provide data that can be used to
917 reproduce an experiment. You have two ways to do that:
919 - Add a string on top of the trace file as comment:
921 .. code-block:: shell
923 --cfg=tracing/comment:my_simulation_identifier
925 - Add the contents of a textual file on top of the trace file as comment:
927 .. code-block:: shell
929 --cfg=tracing/comment-file:my_file_with_additional_information.txt
931 Please, use these two parameters (for comments) to make reproducible
932 simulations. For additional details about this and all tracing
933 options, check See the :ref:`tracing_tracing_options`.
938 .. _cfg=msg/debug-multiple-use:
943 **Option** ``msg/debug-multiple-use`` **Default:** off
945 Sometimes your application may try to send a task that is still being
946 executed somewhere else, making it impossible to send this task. However,
947 for debugging purposes, one may want to know what the other host is/was
948 doing. This option shows a backtrace of the other process.
953 The SMPI interface provides several specific configuration items.
954 These are uneasy to see since the code is usually launched through the
955 ``smiprun`` script directly.
957 .. _cfg=smpi/host-speed:
958 .. _cfg=smpi/cpu-threshold:
959 .. _cfg=smpi/simulate-computation:
961 Automatic Benchmarking of SMPI Code
962 ...................................
964 In SMPI, the sequential code is automatically benchmarked, and these
965 computations are automatically reported to the simulator. That is to
966 say that if you have a large computation between a ``MPI_Recv()`` and
967 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
968 this code, and create an execution task within the simulator to take
969 this into account. For that, the actual duration is measured on the
970 host machine and then scaled to the power of the corresponding
971 simulated machine. The variable ``smpi/host-speed`` allows to specify
972 the computational speed of the host machine (in flop/s) to use when
973 scaling the execution times. It defaults to 20000, but you really want
974 to update it to get accurate simulation results.
976 When the code is constituted of numerous consecutive MPI calls, the
977 previous mechanism feeds the simulation kernel with numerous tiny
978 computations. The ``smpi/cpu-threshold`` item becomes handy when this
979 impacts badly the simulation performance. It specifies a threshold (in
980 seconds) below which the execution chunks are not reported to the
981 simulation kernel (default value: 1e-6).
983 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
984 time spent below this threshold. SMPI does not consider the
985 `amount` of these computations; there is no offset for this. Hence,
986 a value that is too small, may lead to unreliable simulation
989 In some cases, however, one may wish to disable simulation of
990 application computation. This is the case when SMPI is used not to
991 simulate an MPI applications, but instead an MPI code that performs
992 "live replay" of another MPI app (e.g., ScalaTrace's replay tool,
993 various on-line simulators that run an app at scale). In this case the
994 computation of the replay/simulation logic should not be simulated by
995 SMPI. Instead, the replay tool or on-line simulator will issue
996 "computation events", which correspond to the actual MPI simulation
997 being replayed/simulated. At the moment, these computation events can
998 be simulated using SMPI by calling internal smpi_execute*() functions.
1000 To disable the benchmarking/simulation of computation in the simulated
1001 application, the variable ``smpi/simulate-computation`` should be set
1002 to no. This option just ignores the timings in your simulation; it
1003 still executes the computations itself. If you want to stop SMPI from
1004 doing that, you should check the SMPI_SAMPLE macros, documented in
1005 Section :ref:`SMPI_adapting_speed`.
1007 +------------------------------------+-------------------------+-----------------------------+
1008 | Solution | Computations executed? | Computations simulated? |
1009 +====================================+=========================+=============================+
1010 | --cfg=smpi/simulate-computation:no | Yes | Never |
1011 +------------------------------------+-------------------------+-----------------------------+
1012 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
1013 +------------------------------------+-------------------------+-----------------------------+
1014 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
1015 +------------------------------------+-------------------------+-----------------------------+
1017 .. _cfg=smpi/comp-adjustment-file:
1019 Slow-down or speed-up parts of your code
1020 ........................................
1022 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
1024 This option allows you to pass a file that contains two columns: The
1025 first column defines the section that will be subject to a speedup;
1026 the second column is the speedup. For instance:
1028 .. code-block:: shell
1030 "start:stop","ratio"
1031 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1033 The first line is the header - you must include it. The following
1034 line means that the code between two consecutive MPI calls on line 30
1035 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1036 of 1.18244559422142. The value for the second column is therefore a
1037 speedup, if it is larger than 1 and a slow-down if it is smaller
1038 than 1. Nothing will be changed if it is equal to 1.
1040 Of course, you can set any arbitrary filenames you want (so the start
1041 and end don't have to be in the same file), but be aware that this
1042 mechanism only supports `consecutive calls!`
1044 Please note that you must pass the ``-trace-call-location`` flag to
1045 smpicc or smpiff, respectively. This flag activates some internal
1046 macro definitions that help with obtaining the call location.
1048 .. _cfg=smpi/bw-factor:
1053 **Option** ``smpi/bw-factor``
1054 |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
1056 The possible throughput of network links is often dependent on the
1057 message sizes, as protocols may adapt to different message sizes. With
1058 this option, a series of message sizes and factors are given, helping
1059 the simulation to be more realistic. For instance, the current default
1060 value means that messages with size 65472 and more will get a total of
1061 MAX_BANDWIDTH*0.940694, messages of size 15424 to 65471 will get
1062 MAX_BANDWIDTH*0.697866 and so on (where MAX_BANDWIDTH denotes the
1063 bandwidth of the link).
1065 An experimental script to compute these factors is available online. See
1066 http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
1067 http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
1069 .. _cfg=smpi/display-timing:
1071 Reporting Simulation Time
1072 .........................
1074 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1076 Most of the time, you run MPI code with SMPI to compute the time it
1077 would take to run it on a platform. But since the code is run through
1078 the ``smpirun`` script, you don't have any control on the launcher
1079 code, making it difficult to report the simulated time when the
1080 simulation ends. If you enable the ``smpi/display-timing`` item,
1081 ``smpirun`` will display this information when the simulation
1084 .. _cfg=smpi/keep-temps:
1086 Keeping temporary files after simulation
1087 ........................................
1089 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1091 SMPI usually generates a lot of temporary files that are cleaned after
1092 use. This option request to preserve them, for example to debug or
1093 profile your code. Indeed, the binary files are removed very early
1094 under the dlopen privatization schema, which tend to fool the
1097 .. _cfg=smpi/lat-factor:
1102 **Option** ``smpi/lat-factor`` |br|
1103 **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
1105 The motivation and syntax for this option is identical to the motivation/syntax
1106 of :ref:`cfg=smpi/bw-factor`.
1108 There is an important difference, though: While smpi/bw-factor `reduces` the
1109 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
1110 increase the latency, i.e., values larger than or equal to 1 are valid here.
1112 .. _cfg=smpi/papi-events:
1114 Trace hardware counters with PAPI
1115 .................................
1117 **Option** ``smpi/papi-events`` **default:** unset
1119 When the PAPI support was compiled in SimGrid, this option takes the
1120 names of PAPI counters and adds their respective values to the trace
1121 files (See Section :ref:`tracing_tracing_options`).
1125 This feature currently requires superuser privileges, as registers
1126 are queried. Only use this feature with code you trust! Call
1127 smpirun for instance via ``smpirun -wrapper "sudo "
1128 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1129 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1130 will not be required.
1132 It is planned to make this feature available on a per-process (or per-thread?) basis.
1133 The first draft, however, just implements a "global" (i.e., for all processes) set
1134 of counters, the "default" set.
1136 .. code-block:: shell
1138 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1140 .. _cfg=smpi/privatization:
1142 Automatic Privatization of Global Variables
1143 ...........................................
1145 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1147 MPI executables are usually meant to be executed in separated
1148 processes, but SMPI is executed in only one process. Global variables
1149 from executables will be placed in the same memory zone and shared
1150 between processes, causing intricate bugs. Several options are
1151 possible to avoid this, as described in the main `SMPI publication
1152 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1153 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1154 automatically privatizing the globals, and this option allows to
1155 choose between them.
1157 - **no** (default when not using smpirun): Do not automatically
1158 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1160 - **dlopen** or **yes** (default when using smpirun): Link multiple
1161 times against the binary.
1162 - **mmap** (slower, but maybe somewhat more stable):
1163 Runtime automatic switching of the data segments.
1166 This configuration option cannot be set in your platform file. You can only
1167 pass it as an argument to smpirun.
1169 .. _cfg=smpi/privatize-libs:
1171 Automatic privatization of global variables inside external libraries
1172 .....................................................................
1174 **Option** ``smpi/privatize-libs`` **default:** unset
1176 **Linux/BSD only:** When using dlopen (default) privatization,
1177 privatize specific shared libraries with internal global variables, if
1178 they can't be linked statically. For example libgfortran is usually
1179 used for Fortran I/O and indexes in files can be mixed up.
1181 Multiple libraries can be given, semicolon separated.
1183 This configuration option can only use either full paths to libraries,
1184 or full names. Check with ldd the name of the library you want to
1187 .. code-block:: shell
1191 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1194 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1195 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1196 but not ``libgfortran`` nor ``libgfortran.so``.
1198 .. _cfg=smpi/send-is-detached-thresh:
1200 Simulating MPI detached send
1201 ............................
1203 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1205 This threshold specifies the size in bytes under which the send will
1206 return immediately. This is different from the threshold detailed in
1207 :ref:`options_model_network_asyncsend` because the message is not
1208 effectively sent when the send is posted. SMPI still waits for the
1209 correspondant receive to be posted to perform the communication
1212 .. _cfg=smpi/coll-selector:
1214 Simulating MPI collective algorithms
1215 ....................................
1217 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1219 SMPI implements more than 100 different algorithms for MPI collective
1220 communication, to accurately simulate the behavior of most of the
1221 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1222 use the decision logic of either OpenMPI or MPICH libraries (by
1223 default SMPI uses naive version of collective operations).
1225 Each collective operation can be manually selected with a
1226 ``smpi/collective_name:algo_name``. Available algorithms are listed in
1227 :ref:`SMPI_use_colls`.
1229 .. TODO:: All available collective algorithms will be made available
1230 via the ``smpirun --help-coll`` command.
1232 .. _cfg=smpi/iprobe:
1234 Inject constant times for MPI_Iprobe
1235 ....................................
1237 **Option** ``smpi/iprobe`` **default:** 0.0001
1239 The behavior and motivation for this configuration option is identical
1240 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1243 .. _cfg=smpi/iprobe-cpu-usage:
1245 Reduce speed for iprobe calls
1246 .............................
1248 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1250 MPI_Iprobe calls can be heavily used in applications. To account
1251 correctly for the energy cores spend probing, it is necessary to
1252 reduce the load that these calls cause inside SimGrid.
1254 For instance, we measured a max power consumption of 220 W for a
1255 particular application but only 180 W while this application was
1256 probing. Hence, the correct factor that should be passed to this
1257 option would be 180/220 = 0.81.
1261 Inject constant times for MPI_Init
1262 ..................................
1264 **Option** ``smpi/init`` **default:** 0
1266 The behavior and motivation for this configuration option is identical
1267 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1271 Inject constant times for MPI_Isend()
1272 .....................................
1274 **Option** ``smpi/ois``
1276 The behavior and motivation for this configuration option is identical
1277 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1281 Inject constant times for MPI_send()
1282 ....................................
1284 **Option** ``smpi/os``
1286 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1287 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1288 time). SMPI can factor these costs in as well, but the user has to
1289 configure SMPI accordingly as these values may vary by machine. This
1290 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1291 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1292 exactly as ``smpi/ois``.
1294 This item can consist of multiple sections; each section takes three
1295 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1296 so this example contains two sections. Furthermore, each section
1297 consists of three values.
1299 1. The first value denotes the minimum size for this section to take effect;
1300 read it as "if message size is greater than this value (and other section has a larger
1301 first value that is also smaller than the message size), use this".
1302 In the first section above, this value is "1".
1304 2. The second value is the startup time; this is a constant value that will always
1305 be charged, no matter what the size of the message. In the first section above,
1308 3. The third value is the `per-byte` cost. That is, it is charged for every
1309 byte of the message (incurring cost messageSize*cost_per_byte)
1310 and hence accounts also for larger messages. In the first
1311 section of the example above, this value is "2".
1313 Now, SMPI always checks which section it should take for a given
1314 message; that is, if a message of size 11 is sent with the
1315 configuration of the example above, only the second section will be
1316 used, not the first, as the first value of the second section is
1317 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1318 message of size 11 incurs the following cost inside MPI_Send:
1319 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1321 Note that the order of sections can be arbitrary; they will be ordered internally.
1325 Inject constant times for MPI_Recv()
1326 ....................................
1328 **Option** ``smpi/or``
1330 The behavior and motivation for this configuration option is identical
1331 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1334 .. _cfg=smpi/grow-injected-times:
1336 Inject constant times for MPI_Test
1337 ..................................
1339 **Option** ``smpi/test`` **default:** 0.0001
1341 By setting this option, you can control the amount of time a process
1342 sleeps when MPI_Test() is called; this is important, because SimGrid
1343 normally only advances the time while communication is happening and
1344 thus, MPI_Test will not add to the time, resulting in a deadlock if
1345 used as a break-condition as in the following example:
1350 MPI_Test(request, flag, status);
1354 To speed up execution, we use a counter to keep track on how often we
1355 already checked if the handle is now valid or not. Hence, we actually
1356 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1357 process to sleep increases linearly with the number of previously
1358 failed tests. This behavior can be disabled by setting
1359 ``smpi/grow-injected-times`` to **no**. This will also disable this
1360 behavior for MPI_Iprobe.
1362 .. _cfg=smpi/shared-malloc:
1363 .. _cfg=smpi/shared-malloc-hugepage:
1368 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1370 If your simulation consumes too much memory, you may want to modify
1371 your code so that the working areas are shared by all MPI ranks. For
1372 example, in a bloc-cyclic matrix multiplication, you will only
1373 allocate one set of blocs, and every processes will share them.
1374 Naturally, this will lead to very wrong results, but this will save a
1375 lot of memory so this is still desirable for some studies. For more on
1376 the motivation for that feature, please refer to the `relevant section
1377 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1378 of the SMPI CourseWare (see Activity #2.2 of the pointed
1379 assignment). In practice, change the call to malloc() and free() into
1380 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1382 SMPI provides two algorithms for this feature. The first one, called
1383 ``local``, allocates one bloc per call to SMPI_SHARED_MALLOC() in your
1384 code (each call location gets its own bloc) and this bloc is shared
1385 amongst all MPI ranks. This is implemented with the shm_* functions
1386 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1387 for each shared bloc.
1389 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1390 returns a new adress, but it only points to a shadow bloc: its memory
1391 area is mapped on a 1MiB file on disk. If the returned bloc is of size
1392 N MiB, then the same file is mapped N times to cover the whole bloc.
1393 At the end, no matter how many SMPI_SHARED_MALLOC you do, this will
1394 only consume 1 MiB in memory.
1396 You can disable this behavior and come back to regular mallocs (for
1397 example for debugging purposes) using @c "no" as a value.
1399 If you want to keep private some parts of the buffer, for instance if these
1400 parts are used by the application logic and should not be corrupted, you
1401 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). Example:
1405 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1407 This will allocate 500 bytes to mem, such that mem[27..41] and
1408 mem[100..199] are shared while other area remain private.
1410 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1412 When smpi/shared-malloc:global is used, the memory consumption problem
1413 is solved, but it may induce too much load on the kernel's pages table.
1414 In this case, you should use huge pages so that we create only one
1415 entry per Mb of malloced data instead of one entry per 4k.
1416 To activate this, you must mount a hugetlbfs on your system and allocate
1417 at least one huge page:
1419 .. code-block:: shell
1422 sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1423 sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1425 Then, you can pass the option
1426 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1427 actually activate the huge page support in shared mallocs.
1431 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1432 ...................................................................
1434 **Option** ``smpi/wtime`` **default:** 10 ns
1436 This option controls the amount of (simulated) time spent in calls to
1437 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1438 to 0, the simulated clock is not advanced in these calls, which leads
1439 to issue if your application contains such a loop:
1443 while(MPI_Wtime() < some_time_bound) {
1444 /* some tests, with no communication nor computation */
1447 When the option smpi/wtime is set to 0, the time advances only on
1448 communications and computations, so the previous code results in an
1449 infinite loop: the current [simulated] time will never reach
1450 ``some_time_bound``. This infinite loop is avoided when that option
1451 is set to a small amount, as it is by default since SimGrid v3.21.
1453 Note that if your application does not contain any loop depending on
1454 the current time only, then setting this option to a non-zero value
1455 will slow down your simulations by a tiny bit: the simulation loop has
1456 to be broken and reset each time your code ask for the current time.
1457 If the simulation speed really matters to you, you can avoid this
1458 extra delay by setting smpi/wtime to 0.
1460 Other Configurations
1461 --------------------
1463 .. _cfg=clean-atexit:
1465 Cleanup at Termination
1466 ......................
1468 **Option** ``clean-atexit`` **default:** on
1470 If your code is segfaulting during its finalization, it may help to
1471 disable this option to request SimGrid to not attempt any cleanups at
1472 the end of the simulation. Since the Unix process is ending anyway,
1473 the operating system will wipe it all.
1480 **Option** ``path`` **default:** . (current dir)
1482 It is possible to specify a list of directories to search into for the
1483 trace files (see :ref:`pf_trace`) by using this configuration
1484 item. To add several directory to the path, set the configuration
1485 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1487 .. _cfg=simix/breakpoint:
1492 **Option** ``simix/breakpoint`` **default:** unset
1494 This configuration option sets a breakpoint: when the simulated clock
1495 reaches the given time, a SIGTRAP is raised. This can be used to stop
1496 the execution and get a backtrace with a debugger.
1498 It is also possible to set the breakpoint from inside the debugger, by
1499 writing in global variable simgrid::simix::breakpoint. For example,
1502 .. code-block:: shell
1504 set variable simgrid::simix::breakpoint = 3.1416
1506 .. _cfg=verbose-exit:
1511 **Option** ``verbose-exit`` **default:** on
1513 By default, when Ctrl-C is pressed, the status of all existing actors
1514 is displayed before exiting the simulation. This is very useful to
1515 debug your code, but it can reveal troublesome if you have many
1516 actors. Set this configuration item to **off** to disable this
1519 .. _cfg=exception/cutpath:
1521 Truncate local path from exception backtrace
1522 ............................................
1524 **Option** ``exception/cutpath`` **default:** off
1526 This configuration option is used to remove the path from the
1527 backtrace shown when an exception is thrown. This is mainly useful for
1528 the tests: the full file path makes the tests not reproducible because
1529 the path of source files depend of the build settings. That would
1530 break most of our tests as we keep comparing output.
1532 Logging Configuration
1533 ---------------------
1535 It can be done by using XBT. Go to :ref:`XBT_log` for more details.