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 - **contexts/factory:** :ref:`cfg=contexts/factory`
88 - **contexts/guard-size:** :ref:`cfg=contexts/guard-size`
89 - **contexts/nthreads:** :ref:`cfg=contexts/nthreads`
90 - **contexts/parallel-threshold:** :ref:`cfg=contexts/parallel-threshold`
91 - **contexts/stack-size:** :ref:`cfg=contexts/stack-size`
92 - **contexts/synchro:** :ref:`cfg=contexts/synchro`
94 - **cpu/maxmin-selective-update:** :ref:`Cpu Optimization Level <options_model_optim>`
95 - **cpu/model:** :ref:`options_model_select`
96 - **cpu/optim:** :ref:`Cpu Optimization Level <options_model_optim>`
98 - **debug/breakpoint:** :ref:`cfg=debug/breakpoint`
99 - **debug/clean-atexit:** :ref:`cfg=debug/clean-atexit`
100 - **debug/verbose-exit:** :ref:`cfg=debug/verbose-exit`
102 - **exception/cutpath:** :ref:`cfg=exception/cutpath`
104 - **host/model:** :ref:`options_model_select`
106 - **maxmin/precision:** :ref:`cfg=maxmin/precision`
107 - **maxmin/concurrency-limit:** :ref:`cfg=maxmin/concurrency-limit`
109 - **msg/debug-multiple-use:** :ref:`cfg=msg/debug-multiple-use`
111 - **model-check:** :ref:`options_modelchecking`
112 - **model-check/checkpoint:** :ref:`cfg=model-check/checkpoint`
113 - **model-check/communications-determinism:** :ref:`cfg=model-check/communications-determinism`
114 - **model-check/dot-output:** :ref:`cfg=model-check/dot-output`
115 - **model-check/max-depth:** :ref:`cfg=model-check/max-depth`
116 - **model-check/property:** :ref:`cfg=model-check/property`
117 - **model-check/reduction:** :ref:`cfg=model-check/reduction`
118 - **model-check/replay:** :ref:`cfg=model-check/replay`
119 - **model-check/send-determinism:** :ref:`cfg=model-check/send-determinism`
120 - **model-check/termination:** :ref:`cfg=model-check/termination`
121 - **model-check/timeout:** :ref:`cfg=model-check/timeout`
122 - **model-check/visited:** :ref:`cfg=model-check/visited`
124 - **network/bandwidth-factor:** :ref:`cfg=network/bandwidth-factor`
125 - **network/crosstraffic:** :ref:`cfg=network/crosstraffic`
126 - **network/latency-factor:** :ref:`cfg=network/latency-factor`
127 - **network/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 - **storage/max_file_descriptors:** :ref:`cfg=storage/max_file_descriptors`
139 - **surf/precision:** :ref:`cfg=surf/precision`
141 - **For collective operations of SMPI,** please refer to Section :ref:`cfg=smpi/coll-selector`
142 - **smpi/async-small-thresh:** :ref:`cfg=smpi/async-small-thresh`
143 - **smpi/bw-factor:** :ref:`cfg=smpi/bw-factor`
144 - **smpi/coll-selector:** :ref:`cfg=smpi/coll-selector`
145 - **smpi/comp-adjustment-file:** :ref:`cfg=smpi/comp-adjustment-file`
146 - **smpi/cpu-threshold:** :ref:`cfg=smpi/cpu-threshold`
147 - **smpi/display-timing:** :ref:`cfg=smpi/display-timing`
148 - **smpi/grow-injected-times:** :ref:`cfg=smpi/grow-injected-times`
149 - **smpi/host-speed:** :ref:`cfg=smpi/host-speed`
150 - **smpi/IB-penalty-factors:** :ref:`cfg=smpi/IB-penalty-factors`
151 - **smpi/iprobe:** :ref:`cfg=smpi/iprobe`
152 - **smpi/iprobe-cpu-usage:** :ref:`cfg=smpi/iprobe-cpu-usage`
153 - **smpi/init:** :ref:`cfg=smpi/init`
154 - **smpi/keep-temps:** :ref:`cfg=smpi/keep-temps`
155 - **smpi/lat-factor:** :ref:`cfg=smpi/lat-factor`
156 - **smpi/ois:** :ref:`cfg=smpi/ois`
157 - **smpi/or:** :ref:`cfg=smpi/or`
158 - **smpi/os:** :ref:`cfg=smpi/os`
159 - **smpi/papi-events:** :ref:`cfg=smpi/papi-events`
160 - **smpi/privatization:** :ref:`cfg=smpi/privatization`
161 - **smpi/privatize-libs:** :ref:`cfg=smpi/privatize-libs`
162 - **smpi/send-is-detached-thresh:** :ref:`cfg=smpi/send-is-detached-thresh`
163 - **smpi/shared-malloc:** :ref:`cfg=smpi/shared-malloc`
164 - **smpi/shared-malloc-hugepage:** :ref:`cfg=smpi/shared-malloc-hugepage`
165 - **smpi/simulate-computation:** :ref:`cfg=smpi/simulate-computation`
166 - **smpi/test:** :ref:`cfg=smpi/test`
167 - **smpi/wtime:** :ref:`cfg=smpi/wtime`
169 - **Tracing configuration options** can be found in Section :ref:`tracing_tracing_options`
171 - **storage/model:** :ref:`options_model_select`
173 - **vm/model:** :ref:`options_model_select`
177 Configuring the Platform Models
178 -------------------------------
180 .. _options_model_select:
182 Choosing the Platform Models
183 ............................
185 SimGrid comes with several network, CPU and storage models built in,
186 and you can change the used model at runtime by changing the passed
187 configuration. The three main configuration items are given below.
188 For each of these items, passing the special ``help`` value gives you
189 a short description of all possible values (for example,
190 ``--cfg=network/model:help`` will present all provided network
191 models). Also, ``--help-models`` should provide information about all
192 models for all existing resources.
194 - ``network/model``: specify the used network model. Possible values:
196 - **LV08 (default one):** Realistic network analytic model
197 (slow-start modeled by multiplying latency by 13.01, bandwidth by
198 .97; bottleneck sharing uses a payload of S=20537 for evaluating
199 RTT). Described in `Accuracy Study and Improvement of Network
200 Simulation in the SimGrid Framework
201 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
202 - **Constant:** Simplistic network model where all communication
203 take a constant time (one second). This model provides the lowest
204 realism, but is (marginally) faster.
205 - **SMPI:** Realistic network model specifically tailored for HPC
206 settings (accurate modeling of slow start with correction factors on
207 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). This model can be
208 :ref:`further configured <options_model_network>`.
209 - **IB:** Realistic network model specifically tailored for HPC
210 settings with InfiniBand networks (accurate modeling contention
211 behavior, based on the model explained in `this PhD work
212 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
213 This model can be :ref:`further configured <options_model_network>`.
214 - **CM02:** Legacy network analytic model. Very similar to LV08, but
215 without corrective factors. The timings of small messages are thus
216 poorly modeled. This model is described in `A Network Model for
217 Simulation of Grid Application
218 <https://hal.inria.fr/inria-00071989/document>`_.
219 - **ns-3** (only available if you compiled SimGrid accordingly):
220 Use the packet-level network
221 simulators as network models (see :ref:`model_ns3`).
222 This model can be :ref:`further configured <options_pls>`.
224 - ``cpu/model``: specify the used CPU model. We have only one model
227 - **Cas01:** Simplistic CPU model (time=size/power)
229 - ``host/model``: The host concept is the aggregation of a CPU with a
230 network card. Three models exists, but actually, only 2 of them are
231 interesting. The "compound" one is simply due to the way our
232 internal code is organized, and can easily be ignored. So at the
233 end, you have two host models: The default one allows to aggregate
234 an existing CPU model with an existing network model, but does not
235 allow parallel tasks because these beasts need some collaboration
236 between the network and CPU model. That is why, ptask_07 is used by
237 default when using SimDag.
239 - **default:** Default host model. Currently, CPU:Cas01 and
240 network:LV08 (with cross traffic enabled)
241 - **compound:** Host model that is automatically chosen if
242 you change the network and CPU models
243 - **ptask_L07:** Host model somehow similar to Cas01+CM02 but
244 allowing "parallel tasks", that are intended to model the moldable
245 tasks of the grid scheduling literature.
247 - ``storage/model``: specify the used storage model. Only one model is
249 - ``vm/model``: specify the model for virtual machines. Only one model
252 .. todo: make 'compound' the default host model.
254 .. _options_model_optim:
259 The network and CPU models that are based on lmm_solve (that
260 is, all our analytical models) accept specific optimization
263 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
265 - **Lazy:** Lazy action management (partial invalidation in lmm +
266 heap in action remaining).
267 - **TI:** Trace integration. Highly optimized mode when using
268 availability traces (only available for the Cas01 CPU model for
270 - **Full:** Full update of remaining and variables. Slow but may be
271 useful when debugging.
273 - items ``network/maxmin-selective-update`` and
274 ``cpu/maxmin-selective-update``: configure whether the underlying
275 should be lazily updated or not. It should have no impact on the
276 computed timings, but should speed up the computation. |br| It is
277 still possible to disable this feature because it can reveal
278 counter-productive in very specific scenarios where the
279 interaction level is high. In particular, if all your
280 communication share a given backbone link, you should disable it:
281 without it, a simple regular loop is used to update each
282 communication. With it, each of them is still updated (because of
283 the dependency induced by the backbone), but through a complicated
284 and slow pattern that follows the actual dependencies.
286 .. _cfg=maxmin/precision:
287 .. _cfg=surf/precision:
292 **Option** ``maxmin/precision`` **Default:** 0.00001 (in flops or bytes) |br|
293 **Option** ``surf/precision`` **Default:** 0.00001 (in seconds)
295 The analytical models handle a lot of floating point values. It is
296 possible to change the epsilon used to update and compare them through
297 this configuration item. Changing it may speedup the simulation by
298 discarding very small actions, at the price of a reduced numerical
299 precision. You can modify separately the precision used to manipulate
300 timings (in seconds) and the one used to manipulate amounts of work
303 .. _cfg=maxmin/concurrency-limit:
308 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
310 The maximum number of variables per resource can be tuned through this
311 option. You can have as many simultaneous actions per resources as you
312 want. If your simulation presents a very high level of concurrency, it
313 may help to use e.g. 100 as a value here. It means that at most 100
314 actions can consume a resource at a given time. The extraneous actions
315 are queued and wait until the amount of concurrency of the considered
316 resource lowers under the given boundary.
318 Such limitations help both to the simulation speed and simulation accuracy
319 on highly constrained scenarios, but the simulation speed suffers of this
320 setting on regular (less constrained) scenarios so it is off by default.
322 .. _options_model_network:
324 Configuring the Network Model
325 .............................
327 .. _cfg=network/TCP-gamma:
329 Maximal TCP Window Size
330 ^^^^^^^^^^^^^^^^^^^^^^^
332 **Option** ``network/TCP-gamma`` **Default:** 4194304
334 The analytical models need to know the maximal TCP window size to take
335 the TCP congestion mechanism into account. On Linux, this value can
336 be retrieved using the following commands. Both give a set of values,
337 and you should use the last one, which is the maximal size.
339 .. code-block:: shell
341 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
342 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
344 .. _cfg=smpi/IB-penalty-factors:
345 .. _cfg=network/bandwidth-factor:
346 .. _cfg=network/latency-factor:
347 .. _cfg=network/weight-S:
349 Correcting Important Network Parameters
350 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
352 SimGrid can take network irregularities such as a slow startup or
353 changing behavior depending on the message size into account. You
354 should not change these values unless you really know what you're
355 doing. The corresponding values were computed through data fitting
356 one the timings of packet-level simulators, as described in `Accuracy
357 Study and Improvement of Network Simulation in the SimGrid Framework
358 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
361 If you are using the SMPI model, these correction coefficients are
362 themselves corrected by constant values depending on the size of the
363 exchange. By default SMPI uses factors computed on the Stampede
364 Supercomputer at TACC, with optimal deployment of processes on
365 nodes. Again, only hardcore experts should bother about this fact.
367 InfiniBand network behavior can be modeled through 3 parameters
368 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `this PhD
370 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
372 .. todo:: This section should be rewritten, and actually explain the
373 options network/bandwidth-factor, network/latency-factor,
376 .. _cfg=network/crosstraffic:
378 Simulating Cross-Traffic
379 ^^^^^^^^^^^^^^^^^^^^^^^^
381 Since SimGrid v3.7, cross-traffic effects can be taken into account in
382 analytical simulations. It means that ongoing and incoming
383 communication flows are treated independently. In addition, the LV08
384 model adds 0.05 of usage on the opposite direction for each new
385 created flow. This can be useful to simulate some important TCP
386 phenomena such as ack compression.
388 For that to work, your platform must have two links for each
389 pair of interconnected hosts. An example of usable platform is
390 available in ``examples/platforms/crosstraffic.xml``.
392 This is activated through the ``network/crosstraffic`` item, that
393 can be set to 0 (disable this feature) or 1 (enable it).
395 Note that with the default host model this option is activated by default.
397 .. _cfg=smpi/async-small-thresh:
399 Simulating Asyncronous Send
400 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
402 (this configuration item is experimental and may change or disapear)
404 It is possible to specify that messages below a certain size will be
405 sent as soon as the call to MPI_Send is issued, without waiting for
406 the correspondant receive. This threshold can be configured through
407 the ``smpi/async-small-thresh`` item. The default value is 0. This
408 behavior can also be manually set for mailboxes, by setting the
409 receiving mode of the mailbox with a call to
410 :cpp:func:`MSG_mailbox_set_async`. After this, all messages sent to
411 this mailbox will have this behavior regardless of the message size.
413 This value needs to be smaller than or equals to the threshold set at
414 :ref:`cfg=smpi/send-is-detached-thresh`, because asynchronous messages
415 are meant to be detached as well.
422 **Option** ``ns3/TcpModel`` **Default:** "default" (ns-3 default)
424 When using ns-3, there is an extra item ``ns3/TcpModel``, corresponding
425 to the ``ns3::TcpL4Protocol::SocketType`` configuration item in
426 ns-3. The only valid values (enforced on the SimGrid side) are
427 'default' (no change to the ns-3 configuration), 'NewReno' or 'Reno' or
430 Configuring the Storage model
431 .............................
433 .. _cfg=storage/max_file_descriptors:
435 File Descriptor Cound per Host
436 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
438 **Option** ``storage/max_file_descriptors`` **Default:** 1024
440 Each host maintains a fixed-size array of its file descriptors. You
441 can change its size through this item to either enlarge it if your
442 application requires it or to reduce it to save memory space.
449 SimGrid plugins allow to extend the framework without changing its
450 source code directly. Read the source code of the existing plugins to
451 learn how to do so (in ``src/plugins``), and ask your questions to the
452 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
453 that plugins usually register callbacks to some signals of interest.
454 If they need to store some information about a given object (Link, CPU
455 or Actor), they do so through the use of a dedicated object extension.
457 Some of the existing plugins can be activated from the command line,
458 meaning that you can activate them from the command line without any
459 modification to your simulation code. For example, you can activate
460 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
463 Here is the full list of plugins that can be activated this way:
465 - **host_energy:** keeps track of the energy dissipated by
466 computations. More details in @ref plugin_energy.
467 - **link_energy:** keeps track of the energy dissipated by
468 communications. More details in @ref SURF_plugin_energy.
469 - **host_load:** keeps track of the computational load.
470 More details in @ref plugin_load.
472 .. _options_modelchecking:
474 Configuring the Model-Checking
475 ------------------------------
477 To enable the SimGrid model-checking support the program should
478 be executed using the simgrid-mc wrapper:
480 .. code-block:: shell
482 simgrid-mc ./my_program
484 Safety properties are expressed as assertions using the function
485 :cpp:func:`void MC_assert(int prop)`.
487 .. _cfg=model-check/property:
489 Specifying a liveness property
490 ..............................
492 **Option** ``model-check/property`` **Default:** unset
494 If you want to specify liveness properties, you have to pass them on
495 the command line, specifying the name of the file containing the
496 property, as formatted by the ltl2ba program.
499 .. code-block:: shell
501 simgrid-mc ./my_program --cfg=model-check/property:<filename>
503 .. _cfg=model-check/checkpoint:
505 Going for Stateful Verification
506 ...............................
508 By default, the system is backtracked to its initial state to explore
509 another path instead of backtracking to the exact step before the fork
510 that we want to explore (this is called stateless verification). This
511 is done this way because saving intermediate states can rapidly
512 exhaust the available memory. If you want, you can change the value of
513 the ``model-check/checkpoint`` item. For example,
514 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
515 step. Beware, this will certainly explode your memory. Larger values
516 are probably better, make sure to experiment a bit to find the right
517 setting for your specific system.
519 .. _cfg=model-check/reduction:
521 Specifying the kind of reduction
522 ................................
524 The main issue when using the model-checking is the state space
525 explosion. To counter that problem, you can chose a exploration
526 reduction techniques with
527 ``--cfg=model-check/reduction:<technique>``. For now, this
528 configuration variable can take 2 values:
530 - **none:** Do not apply any kind of reduction (mandatory for now for
532 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if
533 you verify local safety properties (default value for safety
536 There is unfortunately no silver bullet here, and the most efficient
537 reduction techniques cannot be applied to any properties. In
538 particular, the DPOR method cannot be applied on liveness properties
539 since our implementation of DPOR may break some cycles, while cycles
540 are very important to the soundness of the exploration for liveness
543 .. _cfg=model-check/visited:
545 Size of Cycle Detection Set
546 ...........................
548 In order to detect cycles, the model-checker needs to check if a new
549 explored state is in fact the same state than a previous one. For
550 that, the model-checker can take a snapshot of each visited state:
551 this snapshot is then used to compare it with subsequent states in the
554 The ``model-check/visited`` item is the maximum number of states which
555 are stored in memory. If the maximum number of snapshotted state is
556 reached, some states will be removed from the memory and some cycles
557 might be missed. Small values can lead to incorrect verifications, but
558 large value can exhaust your memory, so choose carefully.
560 By default, no state is snapshotted and cycles cannot be detected.
562 .. _cfg=model-check/termination:
564 Non-Termination Detection
565 .........................
567 The ``model-check/termination`` configuration item can be used to
568 report if a non-termination execution path has been found. This is a
569 path with a cycle which means that the program might never terminate.
571 This only works in safety mode, not in liveness mode.
573 This options is disabled by default.
575 .. _cfg=model-check/dot-output:
580 If set, the ``model-check/dot-output`` configuration item is the name
581 of a file in which to write a dot file of the path leading the found
582 property (safety or liveness violation) as well as the cycle for
583 liveness properties. This dot file can then fed to the graphviz dot
584 tool to generate an corresponding graphical representation.
586 .. _cfg=model-check/max-depth:
588 Exploration Depth Limit
589 .......................
591 The ``model-checker/max-depth`` can set the maximum depth of the
592 exploration graph of the model-checker. If this limit is reached, a
593 logging message is sent and the results might not be exact.
595 By default, there is not depth limit.
597 .. _cfg=model-check/timeout:
602 By default, the model-checker does not handle timeout conditions: the `wait`
603 operations never time out. With the ``model-check/timeout`` configuration item
604 set to **yes**, the model-checker will explore timeouts of `wait` operations.
606 .. _cfg=model-check/communications-determinism:
607 .. _cfg=model-check/send-determinism:
609 Communication Determinism
610 .........................
612 The ``model-check/communications-determinism`` and
613 ``model-check/send-determinism`` items can be used to select the
614 communication determinism mode of the model-checker which checks
615 determinism properties of the communications of an application.
617 Verification Performance Considerations
618 .......................................
620 The size of the stacks can have a huge impact on the memory
621 consumption when using model-checking. By default, each snapshot will
622 save a copy of the whole stacks and not only of the part which is
623 really meaningful: you should expect the contribution of the memory
624 consumption of the snapshots to be @f$ @mbox{number of processes}
625 @times @mbox{stack size} @times @mbox{number of states} @f$.
627 When compiled against the model checker, the stacks are not
628 protected with guards: if the stack size is too small for your
629 application, the stack will silently overflow on other parts of the
630 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
632 .. _cfg=model-check/replay:
634 Replaying buggy execution paths out of the model-checker
635 ........................................................
637 Debugging the problems reported by the model-checker is challenging: First, the
638 application under verification cannot be debugged with gdb because the
639 model-checker already traces it. Then, the model-checker may explore several
640 execution paths before encountering the issue, making it very difficult to
641 understand the outputs. Fortunately, SimGrid provides the execution path leading
642 to any reported issue so that you can replay this path out of the model checker,
643 enabling the usage of classical debugging tools.
645 When the model-checker finds an interesting path in the application
646 execution graph (where a safety or liveness property is violated), it
647 generates an identifier for this path. Here is an example of output:
649 .. code-block:: shell
651 [ 0.000000] (0:@) Check a safety property
652 [ 0.000000] (0:@) **************************
653 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
654 [ 0.000000] (0:@) **************************
655 [ 0.000000] (0:@) Counter-example execution trace:
656 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
657 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
658 [ 0.000000] (0:@) Path = 1/3;1/4
659 [ 0.000000] (0:@) Expanded states = 27
660 [ 0.000000] (0:@) Visited states = 68
661 [ 0.000000] (0:@) Executed transitions = 46
663 The interesting line is ``Path = 1/3;1/4``, which means that you should use
664 ``--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
665 execution path. All options (but the model-checker related ones) must
666 remain the same. In particular, if you ran your application with
667 ``smpirun -wrapper simgrid-mc``, then do it again. Remove all
668 MC-related options, keep the other ones and add
669 ``--cfg=model-check/replay``.
671 Currently, if the path is of the form ``X;Y;Z``, each number denotes
672 the actor's pid that is selected at each indecision point. If it's of
673 the form ``X/a;Y/b``, the X and Y are the selected pids while the a
674 and b are the return values of their simcalls. In the previouse
675 example, ``1/3;1/4``, you can see from the full output that the actor
676 1 is doing MC_RANDOM simcalls, so the 3 and 4 simply denote the values
677 that these simcall return.
679 Configuring the User Code Virtualization
680 ----------------------------------------
682 .. _cfg=contexts/factory:
684 Selecting the Virtualization Factory
685 ....................................
687 **Option** contexts/factory **Default:** "raw"
689 In SimGrid, the user code is virtualized in a specific mechanism that
690 allows the simulation kernel to control its execution: when a user
691 process requires a blocking action (such as sending a message), it is
692 interrupted, and only gets released when the simulated clock reaches
693 the point where the blocking operation is done. This is explained
694 graphically in the `relevant tutorial, available online
695 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
697 In SimGrid, the containers in which user processes are virtualized are
698 called contexts. Several context factory are provided, and you can
699 select the one you want to use with the ``contexts/factory``
700 configuration item. Some of the following may not exist on your
701 machine because of portability issues. In any case, the default one
702 should be the most effcient one (please report bugs if the
703 auto-detection fails for you). They are approximately sorted here from
704 the slowest to the most efficient:
706 - **thread:** very slow factory using full featured threads (either
707 pthreads or windows native threads). They are slow but very
708 standard. Some debuggers or profilers only work with this factory.
709 - **java:** Java applications are virtualized onto java threads (that
710 are regular pthreads registered to the JVM)
711 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
712 - **boost:** This uses the `context
713 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
714 of the boost library for a performance that is comparable to our
716 |br| Install the relevant library (e.g. with the
717 libboost-contexts-dev package on Debian/Ubuntu) and recompile
719 - **raw:** amazingly fast factory using a context switching mechanism
720 of our own, directly implemented in assembly (only available for x86
721 and amd64 platforms for now) and without any unneeded system call.
723 The main reason to change this setting is when the debugging tools get
724 fooled by the optimized context factories. Threads are the most
725 debugging-friendly contextes, as they allow to set breakpoints
726 anywhere with gdb and visualize backtraces for all processes, in order
727 to debug concurrency issues. Valgrind is also more comfortable with
728 threads, but it should be usable with all factories (Exception: the
729 callgrind tool really dislikes raw and ucontext factories).
731 .. _cfg=contexts/stack-size:
733 Adapting the Stack Size
734 .......................
736 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
738 Each virtualized used process is executed using a specific system
739 stack. The size of this stack has a huge impact on the simulation
740 scalability, but its default value is rather large. This is because
741 the error messages that you get when the stack size is too small are
742 rather disturbing: this leads to stack overflow (overwriting other
743 stacks), leading to segfaults with corrupted stack traces.
745 If you want to push the scalability limits of your code, you might
746 want to reduce the ``contexts/stack-size`` item. Its default value is
747 8192 (in KiB), while our Chord simulation works with stacks as small
748 as 16 KiB, for example. This *setting is ignored* when using the
749 thread factory. Instead, you should compile SimGrid and your
750 application with ``-fsplit-stack``. Note that this compilation flag is
751 not compatible with the model-checker right now.
753 The operating system should only allocate memory for the pages of the
754 stack which are actually used and you might not need to use this in
755 most cases. However, this setting is very important when using the
756 model checker (see :ref:`options_mc_perf`).
758 .. _cfg=contexts/guard-size:
760 Disabling Stack Guard Pages
761 ...........................
763 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages on Windows or with MC)
765 Unless you use the threads context factory (see
766 :ref:`cfg=contexts/factory`), a stack guard page is usually used
767 which prevents the stack of a given actor from overflowing on another
768 stack. But the performance impact may become prohibitive when the
769 amount of actors increases. The option ``contexts/guard-size`` is the
770 number of stack guard pages used. By setting it to 0, no guard pages
771 will be used: in this case, you should avoid using small stacks (with
772 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
773 will silently overflow on other parts of the memory.
775 When no stack guard page is created, stacks may then silently overflow
776 on other parts of the memory if their size is too small for the
779 .. _cfg=contexts/nthreads:
780 .. _cfg=contexts/parallel-threshold:
781 .. _cfg=contexts/synchro:
783 Running User Code in Parallel
784 .............................
786 Parallel execution of the user code is only considered stable in
787 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
788 simulations may well fail in parallel mode. It is described in
789 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
791 If you are using the **ucontext** or **raw** context factories, you can
792 request to execute the user code in parallel. Several threads are
793 launched, each of them handling as much user contexts at each run. To
794 actiave this, set the ``contexts/nthreads`` item to the amount of
795 cores that you have in your computer (or lower than 1 to have
796 the amount of cores auto-detected).
798 Even if you asked several worker threads using the previous option,
799 you can request to start the parallel execution (and pay the
800 associated synchronization costs) only if the potential parallelism is
801 large enough. For that, set the ``contexts/parallel-threshold``
802 item to the minimal amount of user contexts needed to start the
803 parallel execution. In any given simulation round, if that amount is
804 not reached, the contexts will be run sequentially directly by the
805 main thread (thus saving the synchronization costs). Note that this
806 option is mainly useful when the grain of the user code is very fine,
807 because our synchronization is now very efficient.
809 When parallel execution is activated, you can choose the
810 synchronization schema used with the ``contexts/synchro`` item,
811 which value is either:
813 - **futex:** ultra optimized synchronisation schema, based on futexes
814 (fast user-mode mutexes), and thus only available on Linux systems.
815 This is the default mode when available.
816 - **posix:** slow but portable synchronisation using only POSIX
818 - **busy_wait:** not really a synchronisation: the worker threads
819 constantly request new contexts to execute. It should be the most
820 efficient synchronisation schema, but it loads all the cores of
821 your machine for no good reason. You probably prefer the other less
824 Configuring the Tracing
825 -----------------------
827 The :ref:`tracing subsystem <outcomes_vizu>` can be configured in
828 several different ways depending on the nature of the simulator (MSG,
829 SimDag, SMPI) and the kind of traces that need to be obtained. See the
830 :ref:`Tracing Configuration Options subsection
831 <tracing_tracing_options>` to get a detailed description of each
832 configuration option.
834 We detail here a simple way to get the traces working for you, even if
835 you never used the tracing API.
838 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
840 .. code-block:: shell
842 --cfg=tracing:yes --cfg=tracing/uncategorized:yes --cfg=triva/uncategorized:uncat.plist
844 The first parameter activates the tracing subsystem, the second
845 tells it to trace host and link utilization (without any
846 categorization) and the third creates a graph configuration file to
847 configure Triva when analysing the resulting trace file.
849 - MSG or SimDag-based simulator and categorized traces (you need to
850 declare categories and classify your tasks according to them)
852 .. code-block:: shell
854 --cfg=tracing:yes --cfg=tracing/categorized:yes --cfg=triva/categorized:cat.plist
856 The first parameter activates the tracing subsystem, the second
857 tells it to trace host and link categorized utilization and the
858 third creates a graph configuration file to configure Triva when
859 analysing the resulting trace file.
861 - SMPI simulator and traces for a space/time view:
863 .. code-block:: shell
867 The `-trace` parameter for the smpirun script runs the simulation
868 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
869 smpirun's `-help` parameter for additional tracing options.
871 Sometimes you might want to put additional information on the trace to
872 correctly identify them later, or to provide data that can be used to
873 reproduce an experiment. You have two ways to do that:
875 - Add a string on top of the trace file as comment:
877 .. code-block:: shell
879 --cfg=tracing/comment:my_simulation_identifier
881 - Add the contents of a textual file on top of the trace file as comment:
883 .. code-block:: shell
885 --cfg=tracing/comment-file:my_file_with_additional_information.txt
887 Please, use these two parameters (for comments) to make reproducible
888 simulations. For additional details about this and all tracing
889 options, check See the :ref:`tracing_tracing_options`.
894 .. _cfg=msg/debug-multiple-use:
899 **Option** ``msg/debug-multiple-use`` **Default:** off
901 Sometimes your application may try to send a task that is still being
902 executed somewhere else, making it impossible to send this task. However,
903 for debugging purposes, one may want to know what the other host is/was
904 doing. This option shows a backtrace of the other process.
909 The SMPI interface provides several specific configuration items.
910 These are uneasy to see since the code is usually launched through the
911 ``smiprun`` script directly.
913 .. _cfg=smpi/host-speed:
914 .. _cfg=smpi/cpu-threshold:
915 .. _cfg=smpi/simulate-computation:
917 Automatic Benchmarking of SMPI Code
918 ...................................
920 In SMPI, the sequential code is automatically benchmarked, and these
921 computations are automatically reported to the simulator. That is to
922 say that if you have a large computation between a ``MPI_Recv()`` and
923 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
924 this code, and create an execution task within the simulator to take
925 this into account. For that, the actual duration is measured on the
926 host machine and then scaled to the power of the corresponding
927 simulated machine. The variable ``smpi/host-speed`` allows to specify
928 the computational speed of the host machine (in flop/s) to use when
929 scaling the execution times. It defaults to 20000, but you really want
930 to update it to get accurate simulation results.
932 When the code is constituted of numerous consecutive MPI calls, the
933 previous mechanism feeds the simulation kernel with numerous tiny
934 computations. The ``smpi/cpu-threshold`` item becomes handy when this
935 impacts badly the simulation performance. It specifies a threshold (in
936 seconds) below which the execution chunks are not reported to the
937 simulation kernel (default value: 1e-6).
939 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
940 time spent below this threshold. SMPI does not consider the
941 `amount` of these computations; there is no offset for this. Hence,
942 a value that is too small, may lead to unreliable simulation
945 In some cases, however, one may wish to disable simulation of
946 application computation. This is the case when SMPI is used not to
947 simulate an MPI applications, but instead an MPI code that performs
948 "live replay" of another MPI app (e.g., ScalaTrace's replay tool,
949 various on-line simulators that run an app at scale). In this case the
950 computation of the replay/simulation logic should not be simulated by
951 SMPI. Instead, the replay tool or on-line simulator will issue
952 "computation events", which correspond to the actual MPI simulation
953 being replayed/simulated. At the moment, these computation events can
954 be simulated using SMPI by calling internal smpi_execute*() functions.
956 To disable the benchmarking/simulation of computation in the simulated
957 application, the variable ``smpi/simulate-computation`` should be set
958 to no. This option just ignores the timings in your simulation; it
959 still executes the computations itself. If you want to stop SMPI from
960 doing that, you should check the SMPI_SAMPLE macros, documented in
961 Section :ref:`SMPI_adapting_speed`.
963 +------------------------------------+-------------------------+-----------------------------+
964 | Solution | Computations executed? | Computations simulated? |
965 +====================================+=========================+=============================+
966 | --cfg=smpi/simulate-computation:no | Yes | Never |
967 +------------------------------------+-------------------------+-----------------------------+
968 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
969 +------------------------------------+-------------------------+-----------------------------+
970 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
971 +------------------------------------+-------------------------+-----------------------------+
973 .. _cfg=smpi/comp-adjustment-file:
975 Slow-down or speed-up parts of your code
976 ........................................
978 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
980 This option allows you to pass a file that contains two columns: The
981 first column defines the section that will be subject to a speedup;
982 the second column is the speedup. For instance:
984 .. code-block:: shell
987 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
989 The first line is the header - you must include it. The following
990 line means that the code between two consecutive MPI calls on line 30
991 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
992 of 1.18244559422142. The value for the second column is therefore a
993 speedup, if it is larger than 1 and a slow-down if it is smaller
994 than 1. Nothing will be changed if it is equal to 1.
996 Of course, you can set any arbitrary filenames you want (so the start
997 and end don't have to be in the same file), but be aware that this
998 mechanism only supports `consecutive calls!`
1000 Please note that you must pass the ``-trace-call-location`` flag to
1001 smpicc or smpiff, respectively. This flag activates some internal
1002 macro definitions that help with obtaining the call location.
1004 .. _cfg=smpi/bw-factor:
1009 **Option** ``smpi/bw-factor``
1010 |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
1012 The possible throughput of network links is often dependent on the
1013 message sizes, as protocols may adapt to different message sizes. With
1014 this option, a series of message sizes and factors are given, helping
1015 the simulation to be more realistic. For instance, the current default
1016 value means that messages with size 65472 and more will get a total of
1017 MAX_BANDWIDTH*0.940694, messages of size 15424 to 65471 will get
1018 MAX_BANDWIDTH*0.697866 and so on (where MAX_BANDWIDTH denotes the
1019 bandwidth of the link).
1021 An experimental script to compute these factors is available online. See
1022 https://framagit.org/simgrid/platform-calibration/
1023 https://simgrid.org/contrib/smpi-saturation-doc.html
1025 .. _cfg=smpi/display-timing:
1027 Reporting Simulation Time
1028 .........................
1030 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1032 Most of the time, you run MPI code with SMPI to compute the time it
1033 would take to run it on a platform. But since the code is run through
1034 the ``smpirun`` script, you don't have any control on the launcher
1035 code, making it difficult to report the simulated time when the
1036 simulation ends. If you enable the ``smpi/display-timing`` item,
1037 ``smpirun`` will display this information when the simulation
1040 .. _cfg=smpi/keep-temps:
1042 Keeping temporary files after simulation
1043 ........................................
1045 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1047 SMPI usually generates a lot of temporary files that are cleaned after
1048 use. This option request to preserve them, for example to debug or
1049 profile your code. Indeed, the binary files are removed very early
1050 under the dlopen privatization schema, which tend to fool the
1053 .. _cfg=smpi/lat-factor:
1058 **Option** ``smpi/lat-factor`` |br|
1059 **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
1061 The motivation and syntax for this option is identical to the motivation/syntax
1062 of :ref:`cfg=smpi/bw-factor`.
1064 There is an important difference, though: While smpi/bw-factor `reduces` the
1065 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
1066 increase the latency, i.e., values larger than or equal to 1 are valid here.
1068 .. _cfg=smpi/papi-events:
1070 Trace hardware counters with PAPI
1071 .................................
1073 **Option** ``smpi/papi-events`` **default:** unset
1075 When the PAPI support was compiled in SimGrid, this option takes the
1076 names of PAPI counters and adds their respective values to the trace
1077 files (See Section :ref:`tracing_tracing_options`).
1081 This feature currently requires superuser privileges, as registers
1082 are queried. Only use this feature with code you trust! Call
1083 smpirun for instance via ``smpirun -wrapper "sudo "
1084 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1085 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1086 will not be required.
1088 It is planned to make this feature available on a per-process (or per-thread?) basis.
1089 The first draft, however, just implements a "global" (i.e., for all processes) set
1090 of counters, the "default" set.
1092 .. code-block:: shell
1094 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1096 .. _cfg=smpi/privatization:
1098 Automatic Privatization of Global Variables
1099 ...........................................
1101 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1103 MPI executables are usually meant to be executed in separated
1104 processes, but SMPI is executed in only one process. Global variables
1105 from executables will be placed in the same memory zone and shared
1106 between processes, causing intricate bugs. Several options are
1107 possible to avoid this, as described in the main `SMPI publication
1108 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1109 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1110 automatically privatizing the globals, and this option allows to
1111 choose between them.
1113 - **no** (default when not using smpirun): Do not automatically
1114 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1116 - **dlopen** or **yes** (default when using smpirun): Link multiple
1117 times against the binary.
1118 - **mmap** (slower, but maybe somewhat more stable):
1119 Runtime automatic switching of the data segments.
1122 This configuration option cannot be set in your platform file. You can only
1123 pass it as an argument to smpirun.
1125 .. _cfg=smpi/privatize-libs:
1127 Automatic privatization of global variables inside external libraries
1128 .....................................................................
1130 **Option** ``smpi/privatize-libs`` **default:** unset
1132 **Linux/BSD only:** When using dlopen (default) privatization,
1133 privatize specific shared libraries with internal global variables, if
1134 they can't be linked statically. For example libgfortran is usually
1135 used for Fortran I/O and indexes in files can be mixed up.
1137 Multiple libraries can be given, semicolon separated.
1139 This configuration option can only use either full paths to libraries,
1140 or full names. Check with ldd the name of the library you want to
1143 .. code-block:: shell
1147 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1150 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1151 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1152 but not ``libgfortran`` nor ``libgfortran.so``.
1154 .. _cfg=smpi/send-is-detached-thresh:
1156 Simulating MPI detached send
1157 ............................
1159 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1161 This threshold specifies the size in bytes under which the send will
1162 return immediately. This is different from the threshold detailed in
1163 :ref:`cfg=smpi/async-small-thresh` because the message is not
1164 effectively sent when the send is posted. SMPI still waits for the
1165 correspondant receive to be posted to perform the communication
1168 .. _cfg=smpi/coll-selector:
1170 Simulating MPI collective algorithms
1171 ....................................
1173 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1175 SMPI implements more than 100 different algorithms for MPI collective
1176 communication, to accurately simulate the behavior of most of the
1177 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1178 use the decision logic of either OpenMPI or MPICH libraries (by
1179 default SMPI uses naive version of collective operations).
1181 Each collective operation can be manually selected with a
1182 ``smpi/collective_name:algo_name``. Available algorithms are listed in
1183 :ref:`SMPI_use_colls`.
1185 .. TODO:: All available collective algorithms will be made available
1186 via the ``smpirun --help-coll`` command.
1188 .. _cfg=smpi/iprobe:
1190 Inject constant times for MPI_Iprobe
1191 ....................................
1193 **Option** ``smpi/iprobe`` **default:** 0.0001
1195 The behavior and motivation for this configuration option is identical
1196 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1199 .. _cfg=smpi/iprobe-cpu-usage:
1201 Reduce speed for iprobe calls
1202 .............................
1204 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1206 MPI_Iprobe calls can be heavily used in applications. To account
1207 correctly for the energy cores spend probing, it is necessary to
1208 reduce the load that these calls cause inside SimGrid.
1210 For instance, we measured a max power consumption of 220 W for a
1211 particular application but only 180 W while this application was
1212 probing. Hence, the correct factor that should be passed to this
1213 option would be 180/220 = 0.81.
1217 Inject constant times for MPI_Init
1218 ..................................
1220 **Option** ``smpi/init`` **default:** 0
1222 The behavior and motivation for this configuration option is identical
1223 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1227 Inject constant times for MPI_Isend()
1228 .....................................
1230 **Option** ``smpi/ois``
1232 The behavior and motivation for this configuration option is identical
1233 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1237 Inject constant times for MPI_send()
1238 ....................................
1240 **Option** ``smpi/os``
1242 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1243 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1244 time). SMPI can factor these costs in as well, but the user has to
1245 configure SMPI accordingly as these values may vary by machine. This
1246 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1247 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1248 exactly as ``smpi/ois``.
1250 This item can consist of multiple sections; each section takes three
1251 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1252 so this example contains two sections. Furthermore, each section
1253 consists of three values.
1255 1. The first value denotes the minimum size for this section to take effect;
1256 read it as "if message size is greater than this value (and other section has a larger
1257 first value that is also smaller than the message size), use this".
1258 In the first section above, this value is "1".
1260 2. The second value is the startup time; this is a constant value that will always
1261 be charged, no matter what the size of the message. In the first section above,
1264 3. The third value is the `per-byte` cost. That is, it is charged for every
1265 byte of the message (incurring cost messageSize*cost_per_byte)
1266 and hence accounts also for larger messages. In the first
1267 section of the example above, this value is "2".
1269 Now, SMPI always checks which section it should take for a given
1270 message; that is, if a message of size 11 is sent with the
1271 configuration of the example above, only the second section will be
1272 used, not the first, as the first value of the second section is
1273 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1274 message of size 11 incurs the following cost inside MPI_Send:
1275 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1277 Note that the order of sections can be arbitrary; they will be ordered internally.
1281 Inject constant times for MPI_Recv()
1282 ....................................
1284 **Option** ``smpi/or``
1286 The behavior and motivation for this configuration option is identical
1287 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1290 .. _cfg=smpi/grow-injected-times:
1292 Inject constant times for MPI_Test
1293 ..................................
1295 **Option** ``smpi/test`` **default:** 0.0001
1297 By setting this option, you can control the amount of time a process
1298 sleeps when MPI_Test() is called; this is important, because SimGrid
1299 normally only advances the time while communication is happening and
1300 thus, MPI_Test will not add to the time, resulting in a deadlock if
1301 used as a break-condition as in the following example:
1306 MPI_Test(request, flag, status);
1310 To speed up execution, we use a counter to keep track on how often we
1311 already checked if the handle is now valid or not. Hence, we actually
1312 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1313 process to sleep increases linearly with the number of previously
1314 failed tests. This behavior can be disabled by setting
1315 ``smpi/grow-injected-times`` to **no**. This will also disable this
1316 behavior for MPI_Iprobe.
1318 .. _cfg=smpi/shared-malloc:
1319 .. _cfg=smpi/shared-malloc-hugepage:
1324 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1326 If your simulation consumes too much memory, you may want to modify
1327 your code so that the working areas are shared by all MPI ranks. For
1328 example, in a bloc-cyclic matrix multiplication, you will only
1329 allocate one set of blocs, and every processes will share them.
1330 Naturally, this will lead to very wrong results, but this will save a
1331 lot of memory so this is still desirable for some studies. For more on
1332 the motivation for that feature, please refer to the `relevant section
1333 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1334 of the SMPI CourseWare (see Activity #2.2 of the pointed
1335 assignment). In practice, change the call to malloc() and free() into
1336 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1338 SMPI provides two algorithms for this feature. The first one, called
1339 ``local``, allocates one bloc per call to SMPI_SHARED_MALLOC() in your
1340 code (each call location gets its own bloc) and this bloc is shared
1341 amongst all MPI ranks. This is implemented with the shm_* functions
1342 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1343 for each shared bloc.
1345 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1346 returns a new adress, but it only points to a shadow bloc: its memory
1347 area is mapped on a 1MiB file on disk. If the returned bloc is of size
1348 N MiB, then the same file is mapped N times to cover the whole bloc.
1349 At the end, no matter how many SMPI_SHARED_MALLOC you do, this will
1350 only consume 1 MiB in memory.
1352 You can disable this behavior and come back to regular mallocs (for
1353 example for debugging purposes) using @c "no" as a value.
1355 If you want to keep private some parts of the buffer, for instance if these
1356 parts are used by the application logic and should not be corrupted, you
1357 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). Example:
1361 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1363 This will allocate 500 bytes to mem, such that mem[27..41] and
1364 mem[100..199] are shared while other area remain private.
1366 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1368 When smpi/shared-malloc:global is used, the memory consumption problem
1369 is solved, but it may induce too much load on the kernel's pages table.
1370 In this case, you should use huge pages so that we create only one
1371 entry per Mb of malloced data instead of one entry per 4k.
1372 To activate this, you must mount a hugetlbfs on your system and allocate
1373 at least one huge page:
1375 .. code-block:: shell
1378 sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1379 sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1381 Then, you can pass the option
1382 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1383 actually activate the huge page support in shared mallocs.
1387 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1388 ...................................................................
1390 **Option** ``smpi/wtime`` **default:** 10 ns
1392 This option controls the amount of (simulated) time spent in calls to
1393 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1394 to 0, the simulated clock is not advanced in these calls, which leads
1395 to issue if your application contains such a loop:
1399 while(MPI_Wtime() < some_time_bound) {
1400 /* some tests, with no communication nor computation */
1403 When the option smpi/wtime is set to 0, the time advances only on
1404 communications and computations, so the previous code results in an
1405 infinite loop: the current [simulated] time will never reach
1406 ``some_time_bound``. This infinite loop is avoided when that option
1407 is set to a small amount, as it is by default since SimGrid v3.21.
1409 Note that if your application does not contain any loop depending on
1410 the current time only, then setting this option to a non-zero value
1411 will slow down your simulations by a tiny bit: the simulation loop has
1412 to be broken and reset each time your code ask for the current time.
1413 If the simulation speed really matters to you, you can avoid this
1414 extra delay by setting smpi/wtime to 0.
1416 Other Configurations
1417 --------------------
1419 .. _cfg=debug/clean-atexit:
1421 Cleanup at Termination
1422 ......................
1424 **Option** ``debug/clean-atexit`` **default:** on
1426 If your code is segfaulting during its finalization, it may help to
1427 disable this option to request SimGrid to not attempt any cleanups at
1428 the end of the simulation. Since the Unix process is ending anyway,
1429 the operating system will wipe it all.
1436 **Option** ``path`` **default:** . (current dir)
1438 It is possible to specify a list of directories to search into for the
1439 trace files (see :ref:`pf_trace`) by using this configuration
1440 item. To add several directory to the path, set the configuration
1441 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1443 .. _cfg=debug/breakpoint:
1448 **Option** ``debug/breakpoint`` **default:** unset
1450 This configuration option sets a breakpoint: when the simulated clock
1451 reaches the given time, a SIGTRAP is raised. This can be used to stop
1452 the execution and get a backtrace with a debugger.
1454 It is also possible to set the breakpoint from inside the debugger, by
1455 writing in global variable simgrid::simix::breakpoint. For example,
1458 .. code-block:: shell
1460 set variable simgrid::simix::breakpoint = 3.1416
1462 .. _cfg=debug/verbose-exit:
1467 **Option** ``debug/verbose-exit`` **default:** on
1469 By default, when Ctrl-C is pressed, the status of all existing actors
1470 is displayed before exiting the simulation. This is very useful to
1471 debug your code, but it can reveal troublesome if you have many
1472 actors. Set this configuration item to **off** to disable this
1475 .. _cfg=exception/cutpath:
1477 Truncate local path from exception backtrace
1478 ............................................
1480 **Option** ``exception/cutpath`` **default:** off
1482 This configuration option is used to remove the path from the
1483 backtrace shown when an exception is thrown. This is mainly useful for
1484 the tests: the full file path makes the tests not reproducible because
1485 the path of source files depend of the build settings. That would
1486 break most of our tests as we keep comparing output.
1488 Logging Configuration
1489 ---------------------
1491 It can be done by using XBT. Go to :ref:`XBT_log` for more details.