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/termination:** :ref:`cfg=model-check/termination`
120 - **model-check/timeout:** :ref:`cfg=model-check/timeout`
121 - **model-check/visited:** :ref:`cfg=model-check/visited`
123 - **network/bandwidth-factor:** :ref:`cfg=network/bandwidth-factor`
124 - **network/crosstraffic:** :ref:`cfg=network/crosstraffic`
125 - **network/latency-factor:** :ref:`cfg=network/latency-factor`
126 - **network/maxmin-selective-update:** :ref:`Network Optimization Level <options_model_optim>`
127 - **network/model:** :ref:`options_model_select`
128 - **network/optim:** :ref:`Network Optimization Level <options_model_optim>`
129 - **network/TCP-gamma:** :ref:`cfg=network/TCP-gamma`
130 - **network/weight-S:** :ref:`cfg=network/weight-S`
132 - **ns3/TcpModel:** :ref:`options_pls`
133 - **path:** :ref:`cfg=path`
134 - **plugin:** :ref:`cfg=plugin`
136 - **simix/breakpoint:** :ref:`cfg=simix/breakpoint`
138 - **storage/max_file_descriptors:** :ref:`cfg=storage/max_file_descriptors`
140 - **surf/precision:** :ref:`cfg=surf/precision`
142 - **For collective operations of SMPI,** please refer to Section :ref:`cfg=smpi/coll-selector`
143 - **smpi/async-small-thresh:** :ref:`cfg=smpi/async-small-thresh`
144 - **smpi/bw-factor:** :ref:`cfg=smpi/bw-factor`
145 - **smpi/coll-selector:** :ref:`cfg=smpi/coll-selector`
146 - **smpi/comp-adjustment-file:** :ref:`cfg=smpi/comp-adjustment-file`
147 - **smpi/cpu-threshold:** :ref:`cfg=smpi/cpu-threshold`
148 - **smpi/display-timing:** :ref:`cfg=smpi/display-timing`
149 - **smpi/grow-injected-times:** :ref:`cfg=smpi/grow-injected-times`
150 - **smpi/host-speed:** :ref:`cfg=smpi/host-speed`
151 - **smpi/IB-penalty-factors:** :ref:`cfg=smpi/IB-penalty-factors`
152 - **smpi/iprobe:** :ref:`cfg=smpi/iprobe`
153 - **smpi/iprobe-cpu-usage:** :ref:`cfg=smpi/iprobe-cpu-usage`
154 - **smpi/init:** :ref:`cfg=smpi/init`
155 - **smpi/keep-temps:** :ref:`cfg=smpi/keep-temps`
156 - **smpi/lat-factor:** :ref:`cfg=smpi/lat-factor`
157 - **smpi/ois:** :ref:`cfg=smpi/ois`
158 - **smpi/or:** :ref:`cfg=smpi/or`
159 - **smpi/os:** :ref:`cfg=smpi/os`
160 - **smpi/papi-events:** :ref:`cfg=smpi/papi-events`
161 - **smpi/privatization:** :ref:`cfg=smpi/privatization`
162 - **smpi/privatize-libs:** :ref:`cfg=smpi/privatize-libs`
163 - **smpi/send-is-detached-thresh:** :ref:`cfg=smpi/send-is-detached-thresh`
164 - **smpi/shared-malloc:** :ref:`cfg=smpi/shared-malloc`
165 - **smpi/shared-malloc-hugepage:** :ref:`cfg=smpi/shared-malloc-hugepage`
166 - **smpi/simulate-computation:** :ref:`cfg=smpi/simulate-computation`
167 - **smpi/test:** :ref:`cfg=smpi/test`
168 - **smpi/wtime:** :ref:`cfg=smpi/wtime`
170 - **Tracing configuration options** can be found in Section :ref:`tracing_tracing_options`
172 - **storage/model:** :ref:`options_model_select`
173 - **verbose-exit:** :ref:`cfg=verbose-exit`
175 - **vm/model:** :ref:`options_model_select`
179 Configuring the Platform Models
180 -------------------------------
182 .. _options_model_select:
184 Choosing the Platform Models
185 ............................
187 SimGrid comes with several network, CPU and storage models built in,
188 and you can change the used model at runtime by changing the passed
189 configuration. The three main configuration items are given below.
190 For each of these items, passing the special ``help`` value gives you
191 a short description of all possible values (for example,
192 ``--cfg=network/model:help`` will present all provided network
193 models). Also, ``--help-models`` should provide information about all
194 models for all existing resources.
196 - ``network/model``: specify the used network model. Possible values:
198 - **LV08 (default one):** Realistic network analytic model
199 (slow-start modeled by multiplying latency by 13.01, bandwidth by
200 .97; bottleneck sharing uses a payload of S=20537 for evaluating
201 RTT). Described in `Accuracy Study and Improvement of Network
202 Simulation in the SimGrid Framework
203 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
204 - **Constant:** Simplistic network model where all communication
205 take a constant time (one second). This model provides the lowest
206 realism, but is (marginally) faster.
207 - **SMPI:** Realistic network model specifically tailored for HPC
208 settings (accurate modeling of slow start with correction factors on
209 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). This model can be
210 :ref:`further configured <options_model_network>`.
211 - **IB:** Realistic network model specifically tailored for HPC
212 settings with InfiniBand networks (accurate modeling contention
213 behavior, based on the model explained in `this PhD work
214 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
215 This model can be :ref:`further configured <options_model_network>`.
216 - **CM02:** Legacy network analytic model. Very similar to LV08, but
217 without corrective factors. The timings of small messages are thus
218 poorly modeled. This model is described in `A Network Model for
219 Simulation of Grid Application
220 <https://hal.inria.fr/inria-00071989/document>`_.
221 - **Reno/Reno2/Vegas:** Models from Steven H. Low using lagrange_solve instead of
222 lmm_solve (experts only; check the code for more info).
223 - **NS3** (only available if you compiled SimGrid accordingly):
224 Use the packet-level network
225 simulators as network models (see :ref:`pls_ns3`).
226 This model can be :ref:`further configured <options_pls>`.
228 - ``cpu/model``: specify the used CPU model. We have only one model
231 - **Cas01:** Simplistic CPU model (time=size/power)
233 - ``host/model``: The host concept is the aggregation of a CPU with a
234 network card. Three models exists, but actually, only 2 of them are
235 interesting. The "compound" one is simply due to the way our
236 internal code is organized, and can easily be ignored. So at the
237 end, you have two host models: The default one allows to aggregate
238 an existing CPU model with an existing network model, but does not
239 allow parallel tasks because these beasts need some collaboration
240 between the network and CPU model. That is why, ptask_07 is used by
241 default when using SimDag.
243 - **default:** Default host model. Currently, CPU:Cas01 and
244 network:LV08 (with cross traffic enabled)
245 - **compound:** Host model that is automatically chosen if
246 you change the network and CPU models
247 - **ptask_L07:** Host model somehow similar to Cas01+CM02 but
248 allowing "parallel tasks", that are intended to model the moldable
249 tasks of the grid scheduling literature.
251 - ``storage/model``: specify the used storage model. Only one model is
253 - ``vm/model``: specify the model for virtual machines. Only one model
256 .. todo: make 'compound' the default host model.
258 .. _options_model_optim:
263 The network and CPU models that are based on lmm_solve (that
264 is, all our analytical models) accept specific optimization
267 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
269 - **Lazy:** Lazy action management (partial invalidation in lmm +
270 heap in action remaining).
271 - **TI:** Trace integration. Highly optimized mode when using
272 availability traces (only available for the Cas01 CPU model for
274 - **Full:** Full update of remaining and variables. Slow but may be
275 useful when debugging.
277 - items ``network/maxmin-selective-update`` and
278 ``cpu/maxmin-selective-update``: configure whether the underlying
279 should be lazily updated or not. It should have no impact on the
280 computed timings, but should speed up the computation. |br| It is
281 still possible to disable this feature because it can reveal
282 counter-productive in very specific scenarios where the
283 interaction level is high. In particular, if all your
284 communication share a given backbone link, you should disable it:
285 without it, a simple regular loop is used to update each
286 communication. With it, each of them is still updated (because of
287 the dependency induced by the backbone), but through a complicated
288 and slow pattern that follows the actual dependencies.
290 .. _cfg=maxmin/precision:
291 .. _cfg=surf/precision:
296 **Option** ``maxmin/precision`` **Default:** 0.00001 (in flops or bytes) |br|
297 **Option** ``surf/precision`` **Default:** 0.00001 (in seconds)
299 The analytical models handle a lot of floating point values. It is
300 possible to change the epsilon used to update and compare them through
301 this configuration item. Changing it may speedup the simulation by
302 discarding very small actions, at the price of a reduced numerical
303 precision. You can modify separately the precision used to manipulate
304 timings (in seconds) and the one used to manipulate amounts of work
307 .. _cfg=maxmin/concurrency-limit:
312 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
314 The maximum number of variables per resource can be tuned through this
315 option. You can have as many simultaneous actions per resources as you
316 want. If your simulation presents a very high level of concurrency, it
317 may help to use e.g. 100 as a value here. It means that at most 100
318 actions can consume a resource at a given time. The extraneous actions
319 are queued and wait until the amount of concurrency of the considered
320 resource lowers under the given boundary.
322 Such limitations help both to the simulation speed and simulation accuracy
323 on highly constrained scenarios, but the simulation speed suffers of this
324 setting on regular (less constrained) scenarios so it is off by default.
326 .. _options_model_network:
328 Configuring the Network Model
329 .............................
331 .. _cfg=network/TCP-gamma:
333 Maximal TCP Window Size
334 ^^^^^^^^^^^^^^^^^^^^^^^
336 **Option** ``network/TCP-gamma`` **Default:** 4194304
338 The analytical models need to know the maximal TCP window size to take
339 the TCP congestion mechanism into account. On Linux, this value can
340 be retrieved using the following commands. Both give a set of values,
341 and you should use the last one, which is the maximal size.
343 .. code-block:: shell
345 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
346 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
348 .. _cfg=smpi/IB-penalty-factors:
349 .. _cfg=network/bandwidth-factor:
350 .. _cfg=network/latency-factor:
351 .. _cfg=network/weight-S:
353 Correcting Important Network Parameters
354 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
356 SimGrid can take network irregularities such as a slow startup or
357 changing behavior depending on the message size into account. You
358 should not change these values unless you really know what you're
359 doing. The corresponding values were computed through data fitting
360 one the timings of packet-level simulators, as described in `Accuracy
361 Study and Improvement of Network Simulation in the SimGrid Framework
362 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
365 If you are using the SMPI model, these correction coefficients are
366 themselves corrected by constant values depending on the size of the
367 exchange. By default SMPI uses factors computed on the Stampede
368 Supercomputer at TACC, with optimal deployment of processes on
369 nodes. Again, only hardcore experts should bother about this fact.
371 InfiniBand network behavior can be modeled through 3 parameters
372 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `this PhD
374 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
376 .. todo:: This section should be rewritten, and actually explain the
377 options network/bandwidth-factor, network/latency-factor,
380 .. _cfg=network/crosstraffic:
382 Simulating Cross-Traffic
383 ^^^^^^^^^^^^^^^^^^^^^^^^
385 Since SimGrid v3.7, cross-traffic effects can be taken into account in
386 analytical simulations. It means that ongoing and incoming
387 communication flows are treated independently. In addition, the LV08
388 model adds 0.05 of usage on the opposite direction for each new
389 created flow. This can be useful to simulate some important TCP
390 phenomena such as ack compression.
392 For that to work, your platform must have two links for each
393 pair of interconnected hosts. An example of usable platform is
394 available in ``examples/platforms/crosstraffic.xml``.
396 This is activated through the ``network/crosstraffic`` item, that
397 can be set to 0 (disable this feature) or 1 (enable it).
399 Note that with the default host model this option is activated by default.
401 .. _cfg=smpi/async-small-thresh:
403 Simulating Asyncronous Send
404 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
406 (this configuration item is experimental and may change or disapear)
408 It is possible to specify that messages below a certain size will be
409 sent as soon as the call to MPI_Send is issued, without waiting for
410 the correspondant receive. This threshold can be configured through
411 the ``smpi/async-small-thresh`` item. The default value is 0. This
412 behavior can also be manually set for mailboxes, by setting the
413 receiving mode of the mailbox with a call to
414 :cpp:func:`MSG_mailbox_set_async`. After this, all messages sent to
415 this mailbox will have this behavior regardless of the message size.
417 This value needs to be smaller than or equals to the threshold set at
418 @ref options_model_smpi_detached , because asynchronous messages are
419 meant to be detached as well.
426 **Option** ``ns3/TcpModel`` **Default:** "default" (NS3 default)
428 When using NS3, there is an extra item ``ns3/TcpModel``, corresponding
429 to the ``ns3::TcpL4Protocol::SocketType`` configuration item in
430 NS3. The only valid values (enforced on the SimGrid side) are
431 'default' (no change to the NS3 configuration), 'NewReno' or 'Reno' or
434 Configuring the Storage model
435 .............................
437 .. _cfg=storage/max_file_descriptors:
439 File Descriptor Cound per Host
440 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
442 **Option** ``storage/max_file_descriptors`` **Default:** 1024
444 Each host maintains a fixed-size array of its file descriptors. You
445 can change its size through this item to either enlarge it if your
446 application requires it or to reduce it to save memory space.
453 SimGrid plugins allow to extend the framework without changing its
454 source code directly. Read the source code of the existing plugins to
455 learn how to do so (in ``src/plugins``), and ask your questions to the
456 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
457 that plugins usually register callbacks to some signals of interest.
458 If they need to store some information about a given object (Link, CPU
459 or Actor), they do so through the use of a dedicated object extension.
461 Some of the existing plugins can be activated from the command line,
462 meaning that you can activate them from the command line without any
463 modification to your simulation code. For example, you can activate
464 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
467 Here is the full list of plugins that can be activated this way:
469 - **host_energy:** keeps track of the energy dissipated by
470 computations. More details in @ref plugin_energy.
471 - **link_energy:** keeps track of the energy dissipated by
472 communications. More details in @ref SURF_plugin_energy.
473 - **host_load:** keeps track of the computational load.
474 More details in @ref plugin_load.
476 .. _options_modelchecking:
478 Configuring the Model-Checking
479 ------------------------------
481 To enable the SimGrid model-checking support the program should
482 be executed using the simgrid-mc wrapper:
484 .. code-block:: shell
486 simgrid-mc ./my_program
488 Safety properties are expressed as assertions using the function
489 :cpp:func:`void MC_assert(int prop)`.
491 .. _cfg=model-check/property:
493 Specifying a liveness property
494 ..............................
496 **Option** ``model-check/property`` **Default:** unset
498 If you want to specify liveness properties, you have to pass them on
499 the command line, specifying the name of the file containing the
500 property, as formatted by the ltl2ba program.
503 .. code-block:: shell
505 simgrid-mc ./my_program --cfg=model-check/property:<filename>
507 .. _cfg=model-check/checkpoint:
509 Going for Stateful Verification
510 ...............................
512 By default, the system is backtracked to its initial state to explore
513 another path instead of backtracking to the exact step before the fork
514 that we want to explore (this is called stateless verification). This
515 is done this way because saving intermediate states can rapidly
516 exhaust the available memory. If you want, you can change the value of
517 the ``model-check/checkpoint`` item. For example,
518 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
519 step. Beware, this will certainly explode your memory. Larger values
520 are probably better, make sure to experiment a bit to find the right
521 setting for your specific system.
523 .. _cfg=model-check/reduction:
525 Specifying the kind of reduction
526 ................................
528 The main issue when using the model-checking is the state space
529 explosion. To counter that problem, you can chose a exploration
530 reduction techniques with
531 ``--cfg=model-check/reduction:<technique>``. For now, this
532 configuration variable can take 2 values:
534 - **none:** Do not apply any kind of reduction (mandatory for now for
536 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if
537 you verify local safety properties (default value for safety
540 There is unfortunately no silver bullet here, and the most efficient
541 reduction techniques cannot be applied to any properties. In
542 particular, the DPOR method cannot be applied on liveness properties
543 since our implementation of DPOR may break some cycles, while cycles
544 are very important to the soundness of the exploration for liveness
547 .. _cfg=model-check/visited:
549 Size of Cycle Detection Set
550 ...........................
552 In order to detect cycles, the model-checker needs to check if a new
553 explored state is in fact the same state than a previous one. For
554 that, the model-checker can take a snapshot of each visited state:
555 this snapshot is then used to compare it with subsequent states in the
558 The ``model-check/visited`` item is the maximum number of states which
559 are stored in memory. If the maximum number of snapshotted state is
560 reached, some states will be removed from the memory and some cycles
561 might be missed. Small values can lead to incorrect verifications, but
562 large value can exhaust your memory, so choose carefully.
564 By default, no state is snapshotted and cycles cannot be detected.
566 .. _cfg=model-check/termination:
568 Non-Termination Detection
569 .........................
571 The ``model-check/termination`` configuration item can be used to
572 report if a non-termination execution path has been found. This is a
573 path with a cycle which means that the program might never terminate.
575 This only works in safety mode, not in liveness mode.
577 This options is disabled by default.
579 .. _cfg=model-check/dot-output:
584 If set, the ``model-check/dot-output`` configuration item is the name
585 of a file in which to write a dot file of the path leading the found
586 property (safety or liveness violation) as well as the cycle for
587 liveness properties. This dot file can then fed to the graphviz dot
588 tool to generate an corresponding graphical representation.
590 .. _cfg=model-check/max-depth:
592 Exploration Depth Limit
593 .......................
595 The ``model-checker/max-depth`` can set the maximum depth of the
596 exploration graph of the model-checker. If this limit is reached, a
597 logging message is sent and the results might not be exact.
599 By default, there is not depth limit.
601 .. _cfg=model-check/timeout:
606 By default, the model-checker does not handle timeout conditions: the `wait`
607 operations never time out. With the ``model-check/timeout`` configuration item
608 set to **yes**, the model-checker will explore timeouts of `wait` operations.
610 .. _cfg=model-check/communications-determinism:
611 .. _cfg=model-check/send-determinism:
613 Communication Determinism
614 .........................
616 The ``model-check/communications-determinism`` and
617 ``model-check/send-determinism`` items can be used to select the
618 communication determinism mode of the model-checker which checks
619 determinism properties of the communications of an application.
621 Verification Performance Considerations
622 .......................................
624 The size of the stacks can have a huge impact on the memory
625 consumption when using model-checking. By default, each snapshot will
626 save a copy of the whole stacks and not only of the part which is
627 really meaningful: you should expect the contribution of the memory
628 consumption of the snapshots to be @f$ @mbox{number of processes}
629 @times @mbox{stack size} @times @mbox{number of states} @f$.
631 When compiled against the model checker, the stacks are not
632 protected with guards: if the stack size is too small for your
633 application, the stack will silently overflow on other parts of the
634 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
636 .. _cfg=model-checker/hash:
641 Usually most of the time of the model-checker is spent comparing states. This
642 process is complicated and consumes a lot of bandwidth and cache.
643 In order to speedup the state comparison, the experimental ``model-checker/hash``
644 configuration item enables the computation of a hash summarizing as much
645 information of the state as possible into a single value. This hash can be used
646 to avoid most of the comparisons: the costly comparison is then only used when
647 the hashes are identical.
649 Currently most of the state is not included in the hash because the
650 implementation was found to be buggy and this options is not as useful as
651 it could be. For this reason, it is currently disabled by default.
653 .. _cfg=model-check/replay:
655 Replaying buggy execution paths out of the model-checker
656 ........................................................
658 Debugging the problems reported by the model-checker is challenging: First, the
659 application under verification cannot be debugged with gdb because the
660 model-checker already traces it. Then, the model-checker may explore several
661 execution paths before encountering the issue, making it very difficult to
662 understand the outputs. Fortunately, SimGrid provides the execution path leading
663 to any reported issue so that you can replay this path out of the model checker,
664 enabling the usage of classical debugging tools.
666 When the model-checker finds an interesting path in the application
667 execution graph (where a safety or liveness property is violated), it
668 generates an identifier for this path. Here is an example of output:
670 .. code-block:: shell
672 [ 0.000000] (0:@) Check a safety property
673 [ 0.000000] (0:@) **************************
674 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
675 [ 0.000000] (0:@) **************************
676 [ 0.000000] (0:@) Counter-example execution trace:
677 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
678 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
679 [ 0.000000] (0:@) Path = 1/3;1/4
680 [ 0.000000] (0:@) Expanded states = 27
681 [ 0.000000] (0:@) Visited states = 68
682 [ 0.000000] (0:@) Executed transitions = 46
684 The interesting line is ``Path = 1/3;1/4``, which means that you should use
685 `--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
686 execution path. The other options should be the same (but the model-checker
687 should be disabled). Note that format and meaning of the path may change between
690 Configuring the User Code Virtualization
691 ----------------------------------------
693 .. _cfg=contexts/factory:
695 Selecting the Virtualization Factory
696 ....................................
698 **Option** contexts/factory **Default:** "raw"
700 In SimGrid, the user code is virtualized in a specific mechanism that
701 allows the simulation kernel to control its execution: when a user
702 process requires a blocking action (such as sending a message), it is
703 interrupted, and only gets released when the simulated clock reaches
704 the point where the blocking operation is done. This is explained
705 graphically in the `relevant tutorial, available online
706 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
708 In SimGrid, the containers in which user processes are virtualized are
709 called contexts. Several context factory are provided, and you can
710 select the one you want to use with the ``contexts/factory``
711 configuration item. Some of the following may not exist on your
712 machine because of portability issues. In any case, the default one
713 should be the most effcient one (please report bugs if the
714 auto-detection fails for you). They are approximately sorted here from
715 the slowest to the most efficient:
717 - **thread:** very slow factory using full featured threads (either
718 pthreads or windows native threads). They are slow but very
719 standard. Some debuggers or profilers only work with this factory.
720 - **java:** Java applications are virtualized onto java threads (that
721 are regular pthreads registered to the JVM)
722 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
723 - **boost:** This uses the `context
724 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
725 of the boost library for a performance that is comparable to our
727 |br| Install the relevant library (e.g. with the
728 libboost-contexts-dev package on Debian/Ubuntu) and recompile
730 - **raw:** amazingly fast factory using a context switching mechanism
731 of our own, directly implemented in assembly (only available for x86
732 and amd64 platforms for now) and without any unneeded system call.
734 The main reason to change this setting is when the debugging tools get
735 fooled by the optimized context factories. Threads are the most
736 debugging-friendly contextes, as they allow to set breakpoints
737 anywhere with gdb and visualize backtraces for all processes, in order
738 to debug concurrency issues. Valgrind is also more comfortable with
739 threads, but it should be usable with all factories (Exception: the
740 callgrind tool really dislikes raw and ucontext factories).
742 .. _cfg=contexts/stack-size:
744 Adapting the Stack Size
745 .......................
747 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
749 Each virtualized used process is executed using a specific system
750 stack. The size of this stack has a huge impact on the simulation
751 scalability, but its default value is rather large. This is because
752 the error messages that you get when the stack size is too small are
753 rather disturbing: this leads to stack overflow (overwriting other
754 stacks), leading to segfaults with corrupted stack traces.
756 If you want to push the scalability limits of your code, you might
757 want to reduce the ``contexts/stack-size`` item. Its default value is
758 8192 (in KiB), while our Chord simulation works with stacks as small
759 as 16 KiB, for example. This *setting is ignored* when using the
760 thread factory. Instead, you should compile SimGrid and your
761 application with ``-fsplit-stack``. Note that this compilation flag is
762 not compatible with the model-checker right now.
764 The operating system should only allocate memory for the pages of the
765 stack which are actually used and you might not need to use this in
766 most cases. However, this setting is very important when using the
767 model checker (see :ref:`options_mc_perf`).
769 .. _cfg=contexts/guard-size:
771 Disabling Stack Guard Pages
772 ...........................
774 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages on Windows or with MC)
776 Unless you use the threads context factory (see
777 :ref:`cfg=contexts/factory`), a stack guard page is usually used
778 which prevents the stack of a given actor from overflowing on another
779 stack. But the performance impact may become prohibitive when the
780 amount of actors increases. The option ``contexts/guard-size`` is the
781 number of stack guard pages used. By setting it to 0, no guard pages
782 will be used: in this case, you should avoid using small stacks (with
783 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
784 will silently overflow on other parts of the memory.
786 When no stack guard page is created, stacks may then silently overflow
787 on other parts of the memory if their size is too small for the
790 .. _cfg=contexts/nthreads:
791 .. _cfg=contexts/parallel-threshold:
792 .. _cfg=contexts/synchro:
794 Running User Code in Parallel
795 .............................
797 Parallel execution of the user code is only considered stable in
798 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
799 simulations may well fail in parallel mode. It is described in
800 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
802 If you are using the **ucontext** or **raw** context factories, you can
803 request to execute the user code in parallel. Several threads are
804 launched, each of them handling as much user contexts at each run. To
805 actiave this, set the ``contexts/nthreads`` item to the amount of
806 cores that you have in your computer (or lower than 1 to have
807 the amount of cores auto-detected).
809 Even if you asked several worker threads using the previous option,
810 you can request to start the parallel execution (and pay the
811 associated synchronization costs) only if the potential parallelism is
812 large enough. For that, set the ``contexts/parallel-threshold``
813 item to the minimal amount of user contexts needed to start the
814 parallel execution. In any given simulation round, if that amount is
815 not reached, the contexts will be run sequentially directly by the
816 main thread (thus saving the synchronization costs). Note that this
817 option is mainly useful when the grain of the user code is very fine,
818 because our synchronization is now very efficient.
820 When parallel execution is activated, you can choose the
821 synchronization schema used with the ``contexts/synchro`` item,
822 which value is either:
824 - **futex:** ultra optimized synchronisation schema, based on futexes
825 (fast user-mode mutexes), and thus only available on Linux systems.
826 This is the default mode when available.
827 - **posix:** slow but portable synchronisation using only POSIX
829 - **busy_wait:** not really a synchronisation: the worker threads
830 constantly request new contexts to execute. It should be the most
831 efficient synchronisation schema, but it loads all the cores of
832 your machine for no good reason. You probably prefer the other less
835 Configuring the Tracing
836 -----------------------
838 The :ref:`tracing subsystem <outcomes_vizu>` can be configured in
839 several different ways depending on the nature of the simulator (MSG,
840 SimDag, SMPI) and the kind of traces that need to be obtained. See the
841 :ref:`Tracing Configuration Options subsection
842 <tracing_tracing_options>` to get a detailed description of each
843 configuration option.
845 We detail here a simple way to get the traces working for you, even if
846 you never used the tracing API.
849 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
851 .. code-block:: shell
853 --cfg=tracing:yes --cfg=tracing/uncategorized:yes --cfg=triva/uncategorized:uncat.plist
855 The first parameter activates the tracing subsystem, the second
856 tells it to trace host and link utilization (without any
857 categorization) and the third creates a graph configuration file to
858 configure Triva when analysing the resulting trace file.
860 - MSG or SimDag-based simulator and categorized traces (you need to
861 declare categories and classify your tasks according to them)
863 .. code-block:: shell
865 --cfg=tracing:yes --cfg=tracing/categorized:yes --cfg=triva/categorized:cat.plist
867 The first parameter activates the tracing subsystem, the second
868 tells it to trace host and link categorized utilization and the
869 third creates a graph configuration file to configure Triva when
870 analysing the resulting trace file.
872 - SMPI simulator and traces for a space/time view:
874 .. code-block:: shell
878 The `-trace` parameter for the smpirun script runs the simulation
879 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
880 smpirun's `-help` parameter for additional tracing options.
882 Sometimes you might want to put additional information on the trace to
883 correctly identify them later, or to provide data that can be used to
884 reproduce an experiment. You have two ways to do that:
886 - Add a string on top of the trace file as comment:
888 .. code-block:: shell
890 --cfg=tracing/comment:my_simulation_identifier
892 - Add the contents of a textual file on top of the trace file as comment:
894 .. code-block:: shell
896 --cfg=tracing/comment-file:my_file_with_additional_information.txt
898 Please, use these two parameters (for comments) to make reproducible
899 simulations. For additional details about this and all tracing
900 options, check See the :ref:`tracing_tracing_options`.
905 .. _cfg=msg/debug-multiple-use:
910 **Option** ``msg/debug-multiple-use`` **Default:** off
912 Sometimes your application may try to send a task that is still being
913 executed somewhere else, making it impossible to send this task. However,
914 for debugging purposes, one may want to know what the other host is/was
915 doing. This option shows a backtrace of the other process.
920 The SMPI interface provides several specific configuration items.
921 These are uneasy to see since the code is usually launched through the
922 ``smiprun`` script directly.
924 .. _cfg=smpi/host-speed:
925 .. _cfg=smpi/cpu-threshold:
926 .. _cfg=smpi/simulate-computation:
928 Automatic Benchmarking of SMPI Code
929 ...................................
931 In SMPI, the sequential code is automatically benchmarked, and these
932 computations are automatically reported to the simulator. That is to
933 say that if you have a large computation between a ``MPI_Recv()`` and
934 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
935 this code, and create an execution task within the simulator to take
936 this into account. For that, the actual duration is measured on the
937 host machine and then scaled to the power of the corresponding
938 simulated machine. The variable ``smpi/host-speed`` allows to specify
939 the computational speed of the host machine (in flop/s) to use when
940 scaling the execution times. It defaults to 20000, but you really want
941 to update it to get accurate simulation results.
943 When the code is constituted of numerous consecutive MPI calls, the
944 previous mechanism feeds the simulation kernel with numerous tiny
945 computations. The ``smpi/cpu-threshold`` item becomes handy when this
946 impacts badly the simulation performance. It specifies a threshold (in
947 seconds) below which the execution chunks are not reported to the
948 simulation kernel (default value: 1e-6).
950 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
951 time spent below this threshold. SMPI does not consider the
952 `amount` of these computations; there is no offset for this. Hence,
953 a value that is too small, may lead to unreliable simulation
956 In some cases, however, one may wish to disable simulation of
957 application computation. This is the case when SMPI is used not to
958 simulate an MPI applications, but instead an MPI code that performs
959 "live replay" of another MPI app (e.g., ScalaTrace's replay tool,
960 various on-line simulators that run an app at scale). In this case the
961 computation of the replay/simulation logic should not be simulated by
962 SMPI. Instead, the replay tool or on-line simulator will issue
963 "computation events", which correspond to the actual MPI simulation
964 being replayed/simulated. At the moment, these computation events can
965 be simulated using SMPI by calling internal smpi_execute*() functions.
967 To disable the benchmarking/simulation of computation in the simulated
968 application, the variable ``smpi/simulate-computation`` should be set
969 to no. This option just ignores the timings in your simulation; it
970 still executes the computations itself. If you want to stop SMPI from
971 doing that, you should check the SMPI_SAMPLE macros, documented in
972 Section :ref:`SMPI_adapting_speed`.
974 +------------------------------------+-------------------------+-----------------------------+
975 | Solution | Computations executed? | Computations simulated? |
976 +====================================+=========================+=============================+
977 | --cfg=smpi/simulate-computation:no | Yes | Never |
978 +------------------------------------+-------------------------+-----------------------------+
979 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
980 +------------------------------------+-------------------------+-----------------------------+
981 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
982 +------------------------------------+-------------------------+-----------------------------+
984 .. _cfg=smpi/comp-adjustment-file:
986 Slow-down or speed-up parts of your code
987 ........................................
989 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
991 This option allows you to pass a file that contains two columns: The
992 first column defines the section that will be subject to a speedup;
993 the second column is the speedup. For instance:
995 .. code-block:: shell
998 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1000 The first line is the header - you must include it. The following
1001 line means that the code between two consecutive MPI calls on line 30
1002 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1003 of 1.18244559422142. The value for the second column is therefore a
1004 speedup, if it is larger than 1 and a slow-down if it is smaller
1005 than 1. Nothing will be changed if it is equal to 1.
1007 Of course, you can set any arbitrary filenames you want (so the start
1008 and end don't have to be in the same file), but be aware that this
1009 mechanism only supports `consecutive calls!`
1011 Please note that you must pass the ``-trace-call-location`` flag to
1012 smpicc or smpiff, respectively. This flag activates some internal
1013 macro definitions that help with obtaining the call location.
1015 .. _cfg=smpi/bw-factor:
1020 **Option** ``smpi/bw-factor``
1021 |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
1023 The possible throughput of network links is often dependent on the
1024 message sizes, as protocols may adapt to different message sizes. With
1025 this option, a series of message sizes and factors are given, helping
1026 the simulation to be more realistic. For instance, the current default
1027 value means that messages with size 65472 and more will get a total of
1028 MAX_BANDWIDTH*0.940694, messages of size 15424 to 65471 will get
1029 MAX_BANDWIDTH*0.697866 and so on (where MAX_BANDWIDTH denotes the
1030 bandwidth of the link).
1032 An experimental script to compute these factors is available online. See
1033 https://framagit.org/simgrid/platform-calibration/
1034 https://simgrid.org/contrib/smpi-saturation-doc.html
1036 .. _cfg=smpi/display-timing:
1038 Reporting Simulation Time
1039 .........................
1041 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1043 Most of the time, you run MPI code with SMPI to compute the time it
1044 would take to run it on a platform. But since the code is run through
1045 the ``smpirun`` script, you don't have any control on the launcher
1046 code, making it difficult to report the simulated time when the
1047 simulation ends. If you enable the ``smpi/display-timing`` item,
1048 ``smpirun`` will display this information when the simulation
1051 .. _cfg=smpi/keep-temps:
1053 Keeping temporary files after simulation
1054 ........................................
1056 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1058 SMPI usually generates a lot of temporary files that are cleaned after
1059 use. This option request to preserve them, for example to debug or
1060 profile your code. Indeed, the binary files are removed very early
1061 under the dlopen privatization schema, which tend to fool the
1064 .. _cfg=smpi/lat-factor:
1069 **Option** ``smpi/lat-factor`` |br|
1070 **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
1072 The motivation and syntax for this option is identical to the motivation/syntax
1073 of :ref:`cfg=smpi/bw-factor`.
1075 There is an important difference, though: While smpi/bw-factor `reduces` the
1076 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
1077 increase the latency, i.e., values larger than or equal to 1 are valid here.
1079 .. _cfg=smpi/papi-events:
1081 Trace hardware counters with PAPI
1082 .................................
1084 **Option** ``smpi/papi-events`` **default:** unset
1086 When the PAPI support was compiled in SimGrid, this option takes the
1087 names of PAPI counters and adds their respective values to the trace
1088 files (See Section :ref:`tracing_tracing_options`).
1092 This feature currently requires superuser privileges, as registers
1093 are queried. Only use this feature with code you trust! Call
1094 smpirun for instance via ``smpirun -wrapper "sudo "
1095 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1096 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1097 will not be required.
1099 It is planned to make this feature available on a per-process (or per-thread?) basis.
1100 The first draft, however, just implements a "global" (i.e., for all processes) set
1101 of counters, the "default" set.
1103 .. code-block:: shell
1105 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1107 .. _cfg=smpi/privatization:
1109 Automatic Privatization of Global Variables
1110 ...........................................
1112 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1114 MPI executables are usually meant to be executed in separated
1115 processes, but SMPI is executed in only one process. Global variables
1116 from executables will be placed in the same memory zone and shared
1117 between processes, causing intricate bugs. Several options are
1118 possible to avoid this, as described in the main `SMPI publication
1119 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1120 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1121 automatically privatizing the globals, and this option allows to
1122 choose between them.
1124 - **no** (default when not using smpirun): Do not automatically
1125 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1127 - **dlopen** or **yes** (default when using smpirun): Link multiple
1128 times against the binary.
1129 - **mmap** (slower, but maybe somewhat more stable):
1130 Runtime automatic switching of the data segments.
1133 This configuration option cannot be set in your platform file. You can only
1134 pass it as an argument to smpirun.
1136 .. _cfg=smpi/privatize-libs:
1138 Automatic privatization of global variables inside external libraries
1139 .....................................................................
1141 **Option** ``smpi/privatize-libs`` **default:** unset
1143 **Linux/BSD only:** When using dlopen (default) privatization,
1144 privatize specific shared libraries with internal global variables, if
1145 they can't be linked statically. For example libgfortran is usually
1146 used for Fortran I/O and indexes in files can be mixed up.
1148 Multiple libraries can be given, semicolon separated.
1150 This configuration option can only use either full paths to libraries,
1151 or full names. Check with ldd the name of the library you want to
1154 .. code-block:: shell
1158 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1161 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1162 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1163 but not ``libgfortran`` nor ``libgfortran.so``.
1165 .. _cfg=smpi/send-is-detached-thresh:
1167 Simulating MPI detached send
1168 ............................
1170 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1172 This threshold specifies the size in bytes under which the send will
1173 return immediately. This is different from the threshold detailed in
1174 :ref:`options_model_network_asyncsend` because the message is not
1175 effectively sent when the send is posted. SMPI still waits for the
1176 correspondant receive to be posted to perform the communication
1179 .. _cfg=smpi/coll-selector:
1181 Simulating MPI collective algorithms
1182 ....................................
1184 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1186 SMPI implements more than 100 different algorithms for MPI collective
1187 communication, to accurately simulate the behavior of most of the
1188 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1189 use the decision logic of either OpenMPI or MPICH libraries (by
1190 default SMPI uses naive version of collective operations).
1192 Each collective operation can be manually selected with a
1193 ``smpi/collective_name:algo_name``. Available algorithms are listed in
1194 :ref:`SMPI_use_colls`.
1196 .. TODO:: All available collective algorithms will be made available
1197 via the ``smpirun --help-coll`` command.
1199 .. _cfg=smpi/iprobe:
1201 Inject constant times for MPI_Iprobe
1202 ....................................
1204 **Option** ``smpi/iprobe`` **default:** 0.0001
1206 The behavior and motivation for this configuration option is identical
1207 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1210 .. _cfg=smpi/iprobe-cpu-usage:
1212 Reduce speed for iprobe calls
1213 .............................
1215 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1217 MPI_Iprobe calls can be heavily used in applications. To account
1218 correctly for the energy cores spend probing, it is necessary to
1219 reduce the load that these calls cause inside SimGrid.
1221 For instance, we measured a max power consumption of 220 W for a
1222 particular application but only 180 W while this application was
1223 probing. Hence, the correct factor that should be passed to this
1224 option would be 180/220 = 0.81.
1228 Inject constant times for MPI_Init
1229 ..................................
1231 **Option** ``smpi/init`` **default:** 0
1233 The behavior and motivation for this configuration option is identical
1234 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1238 Inject constant times for MPI_Isend()
1239 .....................................
1241 **Option** ``smpi/ois``
1243 The behavior and motivation for this configuration option is identical
1244 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1248 Inject constant times for MPI_send()
1249 ....................................
1251 **Option** ``smpi/os``
1253 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1254 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1255 time). SMPI can factor these costs in as well, but the user has to
1256 configure SMPI accordingly as these values may vary by machine. This
1257 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1258 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1259 exactly as ``smpi/ois``.
1261 This item can consist of multiple sections; each section takes three
1262 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1263 so this example contains two sections. Furthermore, each section
1264 consists of three values.
1266 1. The first value denotes the minimum size for this section to take effect;
1267 read it as "if message size is greater than this value (and other section has a larger
1268 first value that is also smaller than the message size), use this".
1269 In the first section above, this value is "1".
1271 2. The second value is the startup time; this is a constant value that will always
1272 be charged, no matter what the size of the message. In the first section above,
1275 3. The third value is the `per-byte` cost. That is, it is charged for every
1276 byte of the message (incurring cost messageSize*cost_per_byte)
1277 and hence accounts also for larger messages. In the first
1278 section of the example above, this value is "2".
1280 Now, SMPI always checks which section it should take for a given
1281 message; that is, if a message of size 11 is sent with the
1282 configuration of the example above, only the second section will be
1283 used, not the first, as the first value of the second section is
1284 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1285 message of size 11 incurs the following cost inside MPI_Send:
1286 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1288 Note that the order of sections can be arbitrary; they will be ordered internally.
1292 Inject constant times for MPI_Recv()
1293 ....................................
1295 **Option** ``smpi/or``
1297 The behavior and motivation for this configuration option is identical
1298 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1301 .. _cfg=smpi/grow-injected-times:
1303 Inject constant times for MPI_Test
1304 ..................................
1306 **Option** ``smpi/test`` **default:** 0.0001
1308 By setting this option, you can control the amount of time a process
1309 sleeps when MPI_Test() is called; this is important, because SimGrid
1310 normally only advances the time while communication is happening and
1311 thus, MPI_Test will not add to the time, resulting in a deadlock if
1312 used as a break-condition as in the following example:
1317 MPI_Test(request, flag, status);
1321 To speed up execution, we use a counter to keep track on how often we
1322 already checked if the handle is now valid or not. Hence, we actually
1323 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1324 process to sleep increases linearly with the number of previously
1325 failed tests. This behavior can be disabled by setting
1326 ``smpi/grow-injected-times`` to **no**. This will also disable this
1327 behavior for MPI_Iprobe.
1329 .. _cfg=smpi/shared-malloc:
1330 .. _cfg=smpi/shared-malloc-hugepage:
1335 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1337 If your simulation consumes too much memory, you may want to modify
1338 your code so that the working areas are shared by all MPI ranks. For
1339 example, in a bloc-cyclic matrix multiplication, you will only
1340 allocate one set of blocs, and every processes will share them.
1341 Naturally, this will lead to very wrong results, but this will save a
1342 lot of memory so this is still desirable for some studies. For more on
1343 the motivation for that feature, please refer to the `relevant section
1344 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1345 of the SMPI CourseWare (see Activity #2.2 of the pointed
1346 assignment). In practice, change the call to malloc() and free() into
1347 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1349 SMPI provides two algorithms for this feature. The first one, called
1350 ``local``, allocates one bloc per call to SMPI_SHARED_MALLOC() in your
1351 code (each call location gets its own bloc) and this bloc is shared
1352 amongst all MPI ranks. This is implemented with the shm_* functions
1353 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1354 for each shared bloc.
1356 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1357 returns a new adress, but it only points to a shadow bloc: its memory
1358 area is mapped on a 1MiB file on disk. If the returned bloc is of size
1359 N MiB, then the same file is mapped N times to cover the whole bloc.
1360 At the end, no matter how many SMPI_SHARED_MALLOC you do, this will
1361 only consume 1 MiB in memory.
1363 You can disable this behavior and come back to regular mallocs (for
1364 example for debugging purposes) using @c "no" as a value.
1366 If you want to keep private some parts of the buffer, for instance if these
1367 parts are used by the application logic and should not be corrupted, you
1368 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). Example:
1372 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1374 This will allocate 500 bytes to mem, such that mem[27..41] and
1375 mem[100..199] are shared while other area remain private.
1377 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1379 When smpi/shared-malloc:global is used, the memory consumption problem
1380 is solved, but it may induce too much load on the kernel's pages table.
1381 In this case, you should use huge pages so that we create only one
1382 entry per Mb of malloced data instead of one entry per 4k.
1383 To activate this, you must mount a hugetlbfs on your system and allocate
1384 at least one huge page:
1386 .. code-block:: shell
1389 sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1390 sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1392 Then, you can pass the option
1393 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1394 actually activate the huge page support in shared mallocs.
1398 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1399 ...................................................................
1401 **Option** ``smpi/wtime`` **default:** 10 ns
1403 This option controls the amount of (simulated) time spent in calls to
1404 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1405 to 0, the simulated clock is not advanced in these calls, which leads
1406 to issue if your application contains such a loop:
1410 while(MPI_Wtime() < some_time_bound) {
1411 /* some tests, with no communication nor computation */
1414 When the option smpi/wtime is set to 0, the time advances only on
1415 communications and computations, so the previous code results in an
1416 infinite loop: the current [simulated] time will never reach
1417 ``some_time_bound``. This infinite loop is avoided when that option
1418 is set to a small amount, as it is by default since SimGrid v3.21.
1420 Note that if your application does not contain any loop depending on
1421 the current time only, then setting this option to a non-zero value
1422 will slow down your simulations by a tiny bit: the simulation loop has
1423 to be broken and reset each time your code ask for the current time.
1424 If the simulation speed really matters to you, you can avoid this
1425 extra delay by setting smpi/wtime to 0.
1427 Other Configurations
1428 --------------------
1430 .. _cfg=clean-atexit:
1432 Cleanup at Termination
1433 ......................
1435 **Option** ``clean-atexit`` **default:** on
1437 If your code is segfaulting during its finalization, it may help to
1438 disable this option to request SimGrid to not attempt any cleanups at
1439 the end of the simulation. Since the Unix process is ending anyway,
1440 the operating system will wipe it all.
1447 **Option** ``path`` **default:** . (current dir)
1449 It is possible to specify a list of directories to search into for the
1450 trace files (see :ref:`pf_trace`) by using this configuration
1451 item. To add several directory to the path, set the configuration
1452 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1454 .. _cfg=simix/breakpoint:
1459 **Option** ``simix/breakpoint`` **default:** unset
1461 This configuration option sets a breakpoint: when the simulated clock
1462 reaches the given time, a SIGTRAP is raised. This can be used to stop
1463 the execution and get a backtrace with a debugger.
1465 It is also possible to set the breakpoint from inside the debugger, by
1466 writing in global variable simgrid::simix::breakpoint. For example,
1469 .. code-block:: shell
1471 set variable simgrid::simix::breakpoint = 3.1416
1473 .. _cfg=verbose-exit:
1478 **Option** ``verbose-exit`` **default:** on
1480 By default, when Ctrl-C is pressed, the status of all existing actors
1481 is displayed before exiting the simulation. This is very useful to
1482 debug your code, but it can reveal troublesome if you have many
1483 actors. Set this configuration item to **off** to disable this
1486 .. _cfg=exception/cutpath:
1488 Truncate local path from exception backtrace
1489 ............................................
1491 **Option** ``exception/cutpath`` **default:** off
1493 This configuration option is used to remove the path from the
1494 backtrace shown when an exception is thrown. This is mainly useful for
1495 the tests: the full file path makes the tests not reproducible because
1496 the path of source files depend of the build settings. That would
1497 break most of our tests as we keep comparing output.
1499 Logging Configuration
1500 ---------------------
1502 It can be done by using XBT. Go to :ref:`XBT_log` for more details.