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.
84 - **clean-atexit:** :ref:`cfg=clean-atexit`
86 - **contexts/factory:** :ref:`cfg=contexts/factory`
87 - **contexts/guard-size:** :ref:`cfg=contexts/guard-size`
88 - **contexts/nthreads:** :ref:`cfg=contexts/nthreads`
89 - **contexts/parallel-threshold:** :ref:`cfg=contexts/parallel-threshold`
90 - **contexts/stack-size:** :ref:`cfg=contexts/stack-size`
91 - **contexts/synchro:** :ref:`cfg=contexts/synchro`
93 - **cpu/maxmin-selective-update:** :ref:`Cpu Optimization Level <options_model_optim>`
94 - **cpu/model:** :ref:`options_model_select`
95 - **cpu/optim:** :ref:`Cpu Optimization Level <options_model_optim>`
97 - **exception/cutpath:** :ref:`cfg=exception/cutpath`
99 - **host/model:** :ref:`options_model_select`
101 - **maxmin/precision:** :ref:`cfg=maxmin/precision`
102 - **maxmin/concurrency-limit:** :ref:`cfg=maxmin/concurrency-limit`
104 - **msg/debug-multiple-use:** :ref:`cfg=msg/debug-multiple-use`
106 - **model-check:** :ref:`options_modelchecking`
107 - **model-check/checkpoint:** :ref:`cfg=model-check/checkpoint`
108 - **model-check/communications-determinism:** :ref:`cfg=model-check/communications-determinism`
109 - **model-check/dot-output:** :ref:`cfg=model-check/dot-output`
110 - **model-check/hash:** :ref:`cfg=model-checker/hash`
111 - **model-check/max-depth:** :ref:`cfg=model-check/max-depth`
112 - **model-check/property:** :ref:`cfg=model-check/property`
113 - **model-check/record:** :ref:`cfg=model-check/record`
114 - **model-check/reduction:** :ref:`cfg=model-check/reduction`
115 - **model-check/replay:** :ref:`cfg=model-check/replay`
116 - **model-check/send-determinism:** :ref:`cfg=model-check/send-determinism`
117 - **model-check/sparse-checkpoint:** :ref:`cfg=model-check/sparse-checkpoint`
118 - **model-check/termination:** :ref:`cfg=model-check/termination`
119 - **model-check/timeout:** :ref:`cfg=model-check/timeout`
120 - **model-check/visited:** :ref:`cfg=model-check/visited`
122 - **network/bandwidth-factor:** :ref:`cfg=network/bandwidth-factor`
123 - **network/crosstraffic:** :ref:`cfg=network/crosstraffic`
124 - **network/latency-factor:** :ref:`cfg=network/latency-factor`
125 - **network/maxmin-selective-update:** :ref:`Network Optimization Level <options_model_optim>`
126 - **network/model:** :ref:`options_model_select`
127 - **network/optim:** :ref:`Network Optimization Level <options_model_optim>`
128 - **network/TCP-gamma:** :ref:`cfg=network/TCP-gamma`
129 - **network/weight-S:** :ref:`cfg=network/weight-S`
131 - **ns3/TcpModel:** :ref:`options_pls`
132 - **path:** :ref:`cfg=path`
133 - **plugin:** :ref:`cfg=plugin`
135 - **simix/breakpoint:** :ref:`cfg=simix/breakpoint`
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`
172 - **verbose-exit:** :ref:`cfg=verbose-exit`
174 - **vm/model:** :ref:`options_model_select`
178 Configuring the Platform Models
179 -------------------------------
181 .. _options_model_select:
183 Choosing the Platform Models
184 ............................
186 SimGrid comes with several network, CPU and storage models built in,
187 and you can change the used model at runtime by changing the passed
188 configuration. The three main configuration items are given below.
189 For each of these items, passing the special ``help`` value gives you
190 a short description of all possible values (for example,
191 ``--cfg=network/model:help`` will present all provided network
192 models). Also, ``--help-models`` should provide information about all
193 models for all existing resources.
195 - ``network/model``: specify the used network model. Possible values:
197 - **LV08 (default one):** Realistic network analytic model
198 (slow-start modeled by multiplying latency by 13.01, bandwidth by
199 .97; bottleneck sharing uses a payload of S=20537 for evaluating
200 RTT). Described in `Accuracy Study and Improvement of Network
201 Simulation in the SimGrid Framework
202 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
203 - **Constant:** Simplistic network model where all communication
204 take a constant time (one second). This model provides the lowest
205 realism, but is (marginally) faster.
206 - **SMPI:** Realistic network model specifically tailored for HPC
207 settings (accurate modeling of slow start with correction factors on
208 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). This model can be
209 :ref:`further configured <options_model_network>`.
210 - **IB:** Realistic network model specifically tailored for HPC
211 settings with InfiniBand networks (accurate modeling contention
212 behavior, based on the model explained in `this PhD work
213 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
214 This model can be :ref:`further configured <options_model_network>`.
215 - **CM02:** Legacy network analytic model. Very similar to LV08, but
216 without corrective factors. The timings of small messages are thus
217 poorly modeled. This model is described in `A Network Model for
218 Simulation of Grid Application
219 <ftp://ftp.ens-lyon.fr/pub/LIP/Rapports/RR/RR2002/RR2002-40.ps.gz>`_.
220 - **Reno/Reno2/Vegas:** Models from Steven H. Low using lagrange_solve instead of
221 lmm_solve (experts only; check the code for more info).
222 - **NS3** (only available if you compiled SimGrid accordingly):
223 Use the packet-level network
224 simulators as network models (see :ref:`pls_ns3`).
225 This model can be :ref:`further configured <options_pls>`.
227 - ``cpu/model``: specify the used CPU model. We have only one model
230 - **Cas01:** Simplistic CPU model (time=size/power)
232 - ``host/model``: The host concept is the aggregation of a CPU with a
233 network card. Three models exists, but actually, only 2 of them are
234 interesting. The "compound" one is simply due to the way our
235 internal code is organized, and can easily be ignored. So at the
236 end, you have two host models: The default one allows to aggregate
237 an existing CPU model with an existing network model, but does not
238 allow parallel tasks because these beasts need some collaboration
239 between the network and CPU model. That is why, ptask_07 is used by
240 default when using SimDag.
242 - **default:** Default host model. Currently, CPU:Cas01 and
243 network:LV08 (with cross traffic enabled)
244 - **compound:** Host model that is automatically chosen if
245 you change the network and CPU models
246 - **ptask_L07:** Host model somehow similar to Cas01+CM02 but
247 allowing "parallel tasks", that are intended to model the moldable
248 tasks of the grid scheduling literature.
250 - ``storage/model``: specify the used storage model. Only one model is
252 - ``vm/model``: specify the model for virtual machines. Only one model
255 .. todo: make 'compound' the default host model.
257 .. _options_model_optim:
262 The network and CPU models that are based on lmm_solve (that
263 is, all our analytical models) accept specific optimization
266 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
268 - **Lazy:** Lazy action management (partial invalidation in lmm +
269 heap in action remaining).
270 - **TI:** Trace integration. Highly optimized mode when using
271 availability traces (only available for the Cas01 CPU model for
273 - **Full:** Full update of remaining and variables. Slow but may be
274 useful when debugging.
276 - items ``network/maxmin-selective-update`` and
277 ``cpu/maxmin-selective-update``: configure whether the underlying
278 should be lazily updated or not. It should have no impact on the
279 computed timings, but should speed up the computation. |br| It is
280 still possible to disable this feature because it can reveal
281 counter-productive in very specific scenarios where the
282 interaction level is high. In particular, if all your
283 communication share a given backbone link, you should disable it:
284 without it, a simple regular loop is used to update each
285 communication. With it, each of them is still updated (because of
286 the dependency induced by the backbone), but through a complicated
287 and slow pattern that follows the actual dependencies.
289 .. _cfg=maxmin/precision:
290 .. _cfg=surf/precision:
295 **Option** ``maxmin/precision`` **Default:** 0.00001 (in flops or bytes) |br|
296 **Option** ``surf/precision`` **Default:** 0.00001 (in seconds)
298 The analytical models handle a lot of floating point values. It is
299 possible to change the epsilon used to update and compare them through
300 this configuration item. Changing it may speedup the simulation by
301 discarding very small actions, at the price of a reduced numerical
302 precision. You can modify separately the precision used to manipulate
303 timings (in seconds) and the one used to manipulate amounts of work
306 .. _cfg=maxmin/concurrency-limit:
311 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
313 The maximum number of variables per resource can be tuned through this
314 option. You can have as many simultaneous actions per resources as you
315 want. If your simulation presents a very high level of concurrency, it
316 may help to use e.g. 100 as a value here. It means that at most 100
317 actions can consume a resource at a given time. The extraneous actions
318 are queued and wait until the amount of concurrency of the considered
319 resource lowers under the given boundary.
321 Such limitations help both to the simulation speed and simulation accuracy
322 on highly constrained scenarios, but the simulation speed suffers of this
323 setting on regular (less constrained) scenarios so it is off by default.
325 .. _options_model_network:
327 Configuring the Network Model
328 .............................
330 .. _cfg=network/TCP-gamma:
332 Maximal TCP Window Size
333 ^^^^^^^^^^^^^^^^^^^^^^^
335 **Option** ``network/TCP-gamma`` **Default:** 4194304
337 The analytical models need to know the maximal TCP window size to take
338 the TCP congestion mechanism into account. On Linux, this value can
339 be retrieved using the following commands. Both give a set of values,
340 and you should use the last one, which is the maximal size.
342 .. code-block:: shell
344 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
345 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
347 .. _cfg=smpi/IB-penalty-factors:
348 .. _cfg=network/bandwidth-factor:
349 .. _cfg=network/latency-factor:
350 .. _cfg=network/weight-S:
352 Correcting Important Network Parameters
353 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
355 SimGrid can take network irregularities such as a slow startup or
356 changing behavior depending on the message size into account. You
357 should not change these values unless you really know what you're
358 doing. The corresponding values were computed through data fitting
359 one the timings of packet-level simulators, as described in `Accuracy
360 Study and Improvement of Network Simulation in the SimGrid Framework
361 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
364 If you are using the SMPI model, these correction coefficients are
365 themselves corrected by constant values depending on the size of the
366 exchange. By default SMPI uses factors computed on the Stampede
367 Supercomputer at TACC, with optimal deployment of processes on
368 nodes. Again, only hardcore experts should bother about this fact.
370 InfiniBand network behavior can be modeled through 3 parameters
371 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `this PhD
373 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
375 .. todo:: This section should be rewritten, and actually explain the
376 options network/bandwidth-factor, network/latency-factor,
379 .. _cfg=network/crosstraffic:
381 Simulating Cross-Traffic
382 ^^^^^^^^^^^^^^^^^^^^^^^^
384 Since SimGrid v3.7, cross-traffic effects can be taken into account in
385 analytical simulations. It means that ongoing and incoming
386 communication flows are treated independently. In addition, the LV08
387 model adds 0.05 of usage on the opposite direction for each new
388 created flow. This can be useful to simulate some important TCP
389 phenomena such as ack compression.
391 For that to work, your platform must have two links for each
392 pair of interconnected hosts. An example of usable platform is
393 available in ``examples/platforms/crosstraffic.xml``.
395 This is activated through the ``network/crosstraffic`` item, that
396 can be set to 0 (disable this feature) or 1 (enable it).
398 Note that with the default host model this option is activated by default.
400 .. _cfg=smpi/async-small-thresh:
402 Simulating Asyncronous Send
403 ^^^^^^^^^^^^^^^^^^^^^^^^^^^
405 (this configuration item is experimental and may change or disapear)
407 It is possible to specify that messages below a certain size will be
408 sent as soon as the call to MPI_Send is issued, without waiting for
409 the correspondant receive. This threshold can be configured through
410 the ``smpi/async-small-thresh`` item. The default value is 0. This
411 behavior can also be manually set for mailboxes, by setting the
412 receiving mode of the mailbox with a call to
413 :cpp:func:`MSG_mailbox_set_async`. After this, all messages sent to
414 this mailbox will have this behavior regardless of the message size.
416 This value needs to be smaller than or equals to the threshold set at
417 @ref options_model_smpi_detached , because asynchronous messages are
418 meant to be detached as well.
425 **Option** ``ns3/TcpModel`` **Default:** "default" (NS3 default)
427 When using NS3, there is an extra item ``ns3/TcpModel``, corresponding
428 to the ``ns3::TcpL4Protocol::SocketType`` configuration item in
429 NS3. The only valid values (enforced on the SimGrid side) are
430 'default' (no change to the NS3 configuration), 'NewReno' or 'Reno' or
433 Configuring the Storage model
434 .............................
436 .. _cfg=storage/max_file_descriptors:
438 File Descriptor Cound per Host
439 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
441 **Option** ``storage/max_file_descriptors`` **Default:** 1024
443 Each host maintains a fixed-size array of its file descriptors. You
444 can change its size through this item to either enlarge it if your
445 application requires it or to reduce it to save memory space.
452 SimGrid plugins allow to extend the framework without changing its
453 source code directly. Read the source code of the existing plugins to
454 learn how to do so (in ``src/plugins``), and ask your questions to the
455 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
456 that plugins usually register callbacks to some signals of interest.
457 If they need to store some information about a given object (Link, CPU
458 or Actor), they do so through the use of a dedicated object extension.
460 Some of the existing plugins can be activated from the command line,
461 meaning that you can activate them from the command line without any
462 modification to your simulation code. For example, you can activate
463 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
466 Here is the full list of plugins that can be activated this way:
468 - **host_energy:** keeps track of the energy dissipated by
469 computations. More details in @ref plugin_energy.
470 - **link_energy:** keeps track of the energy dissipated by
471 communications. More details in @ref SURF_plugin_energy.
472 - **host_load:** keeps track of the computational load.
473 More details in @ref plugin_load.
475 .. _options_modelchecking:
477 Configuring the Model-Checking
478 ------------------------------
480 To enable the SimGrid model-checking support the program should
481 be executed using the simgrid-mc wrapper:
483 .. code-block:: shell
485 simgrid-mc ./my_program
487 Safety properties are expressed as assertions using the function
488 :cpp:func:`void MC_assert(int prop)`.
490 .. _cfg=model-check/property:
492 Specifying a liveness property
493 ..............................
495 **Option** ``model-check/property`` **Default:** unset
497 If you want to specify liveness properties, you have to pass them on
498 the command line, specifying the name of the file containing the
499 property, as formatted by the ltl2ba program.
502 .. code-block:: shell
504 simgrid-mc ./my_program --cfg=model-check/property:<filename>
506 .. _cfg=model-check/checkpoint:
508 Going for Stateful Verification
509 ...............................
511 By default, the system is backtracked to its initial state to explore
512 another path instead of backtracking to the exact step before the fork
513 that we want to explore (this is called stateless verification). This
514 is done this way because saving intermediate states can rapidly
515 exhaust the available memory. If you want, you can change the value of
516 the ``model-check/checkpoint`` item. For example,
517 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
518 step. Beware, this will certainly explode your memory. Larger values
519 are probably better, make sure to experiment a bit to find the right
520 setting for your specific system.
522 .. _cfg=model-check/reduction:
524 Specifying the kind of reduction
525 ................................
527 The main issue when using the model-checking is the state space
528 explosion. To counter that problem, you can chose a exploration
529 reduction techniques with
530 ``--cfg=model-check/reduction:<technique>``. For now, this
531 configuration variable can take 2 values:
533 - **none:** Do not apply any kind of reduction (mandatory for now for
535 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if
536 you verify local safety properties (default value for safety
539 There is unfortunately no silver bullet here, and the most efficient
540 reduction techniques cannot be applied to any properties. In
541 particular, the DPOR method cannot be applied on liveness properties
542 since our implementation of DPOR may break some cycles, while cycles
543 are very important to the soundness of the exploration for liveness
546 .. _cfg=model-check/visited:
548 Size of Cycle Detection Set
549 ...........................
551 In order to detect cycles, the model-checker needs to check if a new
552 explored state is in fact the same state than a previous one. For
553 that, the model-checker can take a snapshot of each visited state:
554 this snapshot is then used to compare it with subsequent states in the
557 The ``model-check/visited`` item is the maximum number of states which
558 are stored in memory. If the maximum number of snapshotted state is
559 reached, some states will be removed from the memory and some cycles
560 might be missed. Small values can lead to incorrect verifications, but
561 large value can exhaust your memory, so choose carefully.
563 By default, no state is snapshotted and cycles cannot be detected.
565 .. _cfg=model-check/termination:
567 Non-Termination Detection
568 .........................
570 The ``model-check/termination`` configuration item can be used to
571 report if a non-termination execution path has been found. This is a
572 path with a cycle which means that the program might never terminate.
574 This only works in safety mode, not in liveness mode.
576 This options is disabled by default.
578 .. _cfg=model-check/dot-output:
583 If set, the ``model-check/dot-output`` configuration item is the name
584 of a file in which to write a dot file of the path leading the found
585 property (safety or liveness violation) as well as the cycle for
586 liveness properties. This dot file can then fed to the graphviz dot
587 tool to generate an corresponding graphical representation.
589 .. _cfg=model-check/max-depth:
591 Exploration Depth Limit
592 .......................
594 The ``model-checker/max-depth`` can set the maximum depth of the
595 exploration graph of the model-checker. If this limit is reached, a
596 logging message is sent and the results might not be exact.
598 By default, there is not depth limit.
600 .. _cfg=model-check/timeout:
605 By default, the model-checker does not handle timeout conditions: the `wait`
606 operations never time out. With the ``model-check/timeout`` configuration item
607 set to **yes**, the model-checker will explore timeouts of `wait` operations.
609 .. _cfg=model-check/communications-determinism:
610 .. _cfg=model-check/send-determinism:
612 Communication Determinism
613 .........................
615 The ``model-check/communications-determinism`` and
616 ``model-check/send-determinism`` items can be used to select the
617 communication determinism mode of the model-checker which checks
618 determinism properties of the communications of an application.
620 .. _cfg=model-check/sparse-checkpoint:
622 Incremental Checkpoints
623 .......................
625 When the model-checker is configured to take a snapshot of each
626 explored state (with the ``model-checker/visited`` item), the memory
627 consumption can rapidly reach GiB ou Tib of memory. However, for many
628 workloads, the memory does not change much between different snapshots
629 and taking a complete copy of each snapshot is a waste of memory.
631 The ``model-check/sparse-checkpoint`` option item can be set to
632 **yes** to avoid making a complete copy of each snapshot. Instead,
633 each snapshot will be decomposed in blocks which will be stored
634 separately. If multiple snapshots share the same block (or if the
635 same block is used in the same snapshot), the same copy of the block
636 will be shared leading to a reduction of the memory footprint.
638 For many applications, this option considerably reduces the memory
639 consumption. In somes cases, the model-checker might be slightly
640 slower because of the time taken to manage the metadata about the
641 blocks. In other cases however, this snapshotting strategy will be
642 much faster by reducing the cache consumption. When the memory
643 consumption is important, by avoiding to hit the swap or reducing the
644 swap usage, this option might be much faster than the basic
645 snapshotting strategy.
647 This option is currently disabled by default.
649 Verification Performance Considerations
650 .......................................
652 The size of the stacks can have a huge impact on the memory
653 consumption when using model-checking. By default, each snapshot will
654 save a copy of the whole stacks and not only of the part which is
655 really meaningful: you should expect the contribution of the memory
656 consumption of the snapshots to be @f$ @mbox{number of processes}
657 @times @mbox{stack size} @times @mbox{number of states} @f$.
659 The ``model-check/sparse-checkpoint`` can be used to reduce the memory
660 consumption by trying to share memory between the different snapshots.
662 When compiled against the model checker, the stacks are not
663 protected with guards: if the stack size is too small for your
664 application, the stack will silently overflow on other parts of the
665 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
667 .. _cfg=model-checker/hash:
672 Usually most of the time of the model-checker is spent comparing states. This
673 process is complicated and consumes a lot of bandwidth and cache.
674 In order to speedup the state comparison, the experimental ``model-checker/hash``
675 configuration item enables the computation of a hash summarizing as much
676 information of the state as possible into a single value. This hash can be used
677 to avoid most of the comparisons: the costly comparison is then only used when
678 the hashes are identical.
680 Currently most of the state is not included in the hash because the
681 implementation was found to be buggy and this options is not as useful as
682 it could be. For this reason, it is currently disabled by default.
684 .. _cfg=model-check/record:
685 .. _cfg=model-check/replay:
687 Record/Replay of Verification
688 .............................
690 As the model-checker keeps jumping at different places in the execution graph,
691 it is difficult to understand what happens when trying to debug an application
692 under the model-checker. Event the output of the program is difficult to
693 interpret. Moreover, the model-checker does not behave nicely with advanced
694 debugging tools such as valgrind. For those reason, to identify a trajectory
695 in the execution graph with the model-checker and replay this trajcetory and
696 without the model-checker black-magic but with more standard tools
697 (such as a debugger, valgrind, etc.). For this reason, Simgrid implements an
698 experimental record/replay functionnality in order to record a trajectory with
699 the model-checker and replay it without the model-checker.
701 When the model-checker finds an interesting path in the application
702 execution graph (where a safety or liveness property is violated), it
703 can generate an identifier for this path. To enable this behavious the
704 ``model-check/record`` must be set to **yes**, which is not the case
707 Here is an example of output:
709 .. code-block:: shell
711 [ 0.000000] (0:@) Check a safety property
712 [ 0.000000] (0:@) **************************
713 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
714 [ 0.000000] (0:@) **************************
715 [ 0.000000] (0:@) Counter-example execution trace:
716 [ 0.000000] (0:@) Path = 1/3;1/4
717 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
718 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
719 [ 0.000000] (0:@) Expanded states = 27
720 [ 0.000000] (0:@) Visited states = 68
721 [ 0.000000] (0:@) Executed transitions = 46
723 This path can then be replayed outside of the model-checker (and even
724 in non-MC build of simgrid) by setting the ``model-check/replay`` item
725 to the given path. The other options should be the same (but the
726 model-checker should be disabled).
728 The format and meaning of the path may change between different
729 releases so the same release of Simgrid should be used for the record
730 phase and the replay phase.
732 Configuring the User Code Virtualization
733 ----------------------------------------
735 .. _cfg=contexts/factory:
737 Selecting the Virtualization Factory
738 ....................................
740 **Option** contexts/factory **Default:** "raw"
742 In SimGrid, the user code is virtualized in a specific mechanism that
743 allows the simulation kernel to control its execution: when a user
744 process requires a blocking action (such as sending a message), it is
745 interrupted, and only gets released when the simulated clock reaches
746 the point where the blocking operation is done. This is explained
747 graphically in the `relevant tutorial, available online
748 <http://simgrid.gforge.inria.fr/tutorials/simgrid-simix-101.pdf>`_.
750 In SimGrid, the containers in which user processes are virtualized are
751 called contexts. Several context factory are provided, and you can
752 select the one you want to use with the ``contexts/factory``
753 configuration item. Some of the following may not exist on your
754 machine because of portability issues. In any case, the default one
755 should be the most effcient one (please report bugs if the
756 auto-detection fails for you). They are approximately sorted here from
757 the slowest to the most efficient:
759 - **thread:** very slow factory using full featured threads (either
760 pthreads or windows native threads). They are slow but very
761 standard. Some debuggers or profilers only work with this factory.
762 - **java:** Java applications are virtualized onto java threads (that
763 are regular pthreads registered to the JVM)
764 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
765 - **boost:** This uses the `context
766 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
767 of the boost library for a performance that is comparable to our
769 |br| Install the relevant library (e.g. with the
770 libboost-contexts-dev package on Debian/Ubuntu) and recompile
772 - **raw:** amazingly fast factory using a context switching mechanism
773 of our own, directly implemented in assembly (only available for x86
774 and amd64 platforms for now) and without any unneeded system call.
776 The main reason to change this setting is when the debugging tools get
777 fooled by the optimized context factories. Threads are the most
778 debugging-friendly contextes, as they allow to set breakpoints
779 anywhere with gdb and visualize backtraces for all processes, in order
780 to debug concurrency issues. Valgrind is also more comfortable with
781 threads, but it should be usable with all factories (Exception: the
782 callgrind tool really dislikes raw and ucontext factories).
784 .. _cfg=contexts/stack-size:
786 Adapting the Stack Size
787 .......................
789 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
791 Each virtualized used process is executed using a specific system
792 stack. The size of this stack has a huge impact on the simulation
793 scalability, but its default value is rather large. This is because
794 the error messages that you get when the stack size is too small are
795 rather disturbing: this leads to stack overflow (overwriting other
796 stacks), leading to segfaults with corrupted stack traces.
798 If you want to push the scalability limits of your code, you might
799 want to reduce the ``contexts/stack-size`` item. Its default value is
800 8192 (in KiB), while our Chord simulation works with stacks as small
801 as 16 KiB, for example. For the thread factory, the default value is
802 the one of the system but you can still change it with this parameter.
804 The operating system should only allocate memory for the pages of the
805 stack which are actually used and you might not need to use this in
806 most cases. However, this setting is very important when using the
807 model checker (see :ref:`options_mc_perf`).
809 .. _cfg=contexts/guard-size:
811 Disabling Stack Guard Pages
812 ...........................
814 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages on Windows or with MC)
816 A stack guard page is usually used which prevents the stack of a given
817 actor from overflowing on another stack. But the performance impact
818 may become prohibitive when the amount of actors increases. The
819 option ``contexts/guard-size`` is the number of stack guard pages
820 used. By setting it to 0, no guard pages will be used: in this case,
821 you should avoid using small stacks (with :ref:`contexts/stack-size
822 <cfg=contexts/stack-size>`) as the stack will silently overflow on
823 other parts of the memory.
825 When no stack guard page is created, stacks may then silently overflow
826 on other parts of the memory if their size is too small for the
829 .. _cfg=contexts/nthreads:
830 .. _cfg=contexts/parallel-threshold:
831 .. _cfg=contexts/synchro:
833 Running User Code in Parallel
834 .............................
836 Parallel execution of the user code is only considered stable in
837 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
838 simulations may well fail in parallel mode. It is described in
839 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
841 If you are using the **ucontext** or **raw** context factories, you can
842 request to execute the user code in parallel. Several threads are
843 launched, each of them handling as much user contexts at each run. To
844 actiave this, set the ``contexts/nthreads`` item to the amount of
845 cores that you have in your computer (or lower than 1 to have
846 the amount of cores auto-detected).
848 Even if you asked several worker threads using the previous option,
849 you can request to start the parallel execution (and pay the
850 associated synchronization costs) only if the potential parallelism is
851 large enough. For that, set the ``contexts/parallel-threshold``
852 item to the minimal amount of user contexts needed to start the
853 parallel execution. In any given simulation round, if that amount is
854 not reached, the contexts will be run sequentially directly by the
855 main thread (thus saving the synchronization costs). Note that this
856 option is mainly useful when the grain of the user code is very fine,
857 because our synchronization is now very efficient.
859 When parallel execution is activated, you can choose the
860 synchronization schema used with the ``contexts/synchro`` item,
861 which value is either:
863 - **futex:** ultra optimized synchronisation schema, based on futexes
864 (fast user-mode mutexes), and thus only available on Linux systems.
865 This is the default mode when available.
866 - **posix:** slow but portable synchronisation using only POSIX
868 - **busy_wait:** not really a synchronisation: the worker threads
869 constantly request new contexts to execute. It should be the most
870 efficient synchronisation schema, but it loads all the cores of
871 your machine for no good reason. You probably prefer the other less
875 Configuring the Tracing
876 -----------------------
878 The :ref:`tracing subsystem <outcomes_vizu>` can be configured in
879 several different ways depending on the nature of the simulator (MSG,
880 SimDag, SMPI) and the kind of traces that need to be obtained. See the
881 :ref:`Tracing Configuration Options subsection
882 <tracing_tracing_options>` to get a detailed description of each
883 configuration option.
885 We detail here a simple way to get the traces working for you, even if
886 you never used the tracing API.
889 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
891 .. code-block:: shell
893 --cfg=tracing:yes --cfg=tracing/uncategorized:yes --cfg=triva/uncategorized:uncat.plist
895 The first parameter activates the tracing subsystem, the second
896 tells it to trace host and link utilization (without any
897 categorization) and the third creates a graph configuration file to
898 configure Triva when analysing the resulting trace file.
900 - MSG or SimDag-based simulator and categorized traces (you need to
901 declare categories and classify your tasks according to them)
903 .. code-block:: shell
905 --cfg=tracing:yes --cfg=tracing/categorized:yes --cfg=triva/categorized:cat.plist
907 The first parameter activates the tracing subsystem, the second
908 tells it to trace host and link categorized utilization and the
909 third creates a graph configuration file to configure Triva when
910 analysing the resulting trace file.
912 - SMPI simulator and traces for a space/time view:
914 .. code-block:: shell
918 The `-trace` parameter for the smpirun script runs the simulation
919 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
920 smpirun's `-help` parameter for additional tracing options.
922 Sometimes you might want to put additional information on the trace to
923 correctly identify them later, or to provide data that can be used to
924 reproduce an experiment. You have two ways to do that:
926 - Add a string on top of the trace file as comment:
928 .. code-block:: shell
930 --cfg=tracing/comment:my_simulation_identifier
932 - Add the contents of a textual file on top of the trace file as comment:
934 .. code-block:: shell
936 --cfg=tracing/comment-file:my_file_with_additional_information.txt
938 Please, use these two parameters (for comments) to make reproducible
939 simulations. For additional details about this and all tracing
940 options, check See the :ref:`tracing_tracing_options`.
945 .. _cfg=msg/debug-multiple-use:
950 **Option** ``msg/debug-multiple-use`` **Default:** off
952 Sometimes your application may try to send a task that is still being
953 executed somewhere else, making it impossible to send this task. However,
954 for debugging purposes, one may want to know what the other host is/was
955 doing. This option shows a backtrace of the other process.
960 The SMPI interface provides several specific configuration items.
961 These are uneasy to see since the code is usually launched through the
962 ``smiprun`` script directly.
964 .. _cfg=smpi/host-speed:
965 .. _cfg=smpi/cpu-threshold:
966 .. _cfg=smpi/simulate-computation:
968 Automatic Benchmarking of SMPI Code
969 ...................................
971 In SMPI, the sequential code is automatically benchmarked, and these
972 computations are automatically reported to the simulator. That is to
973 say that if you have a large computation between a ``MPI_Recv()`` and
974 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
975 this code, and create an execution task within the simulator to take
976 this into account. For that, the actual duration is measured on the
977 host machine and then scaled to the power of the corresponding
978 simulated machine. The variable ``smpi/host-speed`` allows to specify
979 the computational speed of the host machine (in flop/s) to use when
980 scaling the execution times. It defaults to 20000, but you really want
981 to update it to get accurate simulation results.
983 When the code is constituted of numerous consecutive MPI calls, the
984 previous mechanism feeds the simulation kernel with numerous tiny
985 computations. The ``smpi/cpu-threshold`` item becomes handy when this
986 impacts badly the simulation performance. It specifies a threshold (in
987 seconds) below which the execution chunks are not reported to the
988 simulation kernel (default value: 1e-6).
990 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
991 time spent below this threshold. SMPI does not consider the
992 `amount` of these computations; there is no offset for this. Hence,
993 a value that is too small, may lead to unreliable simulation
996 In some cases, however, one may wish to disable simulation of
997 application computation. This is the case when SMPI is used not to
998 simulate an MPI applications, but instead an MPI code that performs
999 "live replay" of another MPI app (e.g., ScalaTrace's replay tool,
1000 various on-line simulators that run an app at scale). In this case the
1001 computation of the replay/simulation logic should not be simulated by
1002 SMPI. Instead, the replay tool or on-line simulator will issue
1003 "computation events", which correspond to the actual MPI simulation
1004 being replayed/simulated. At the moment, these computation events can
1005 be simulated using SMPI by calling internal smpi_execute*() functions.
1007 To disable the benchmarking/simulation of computation in the simulated
1008 application, the variable ``smpi/simulate-computation`` should be set
1009 to no. This option just ignores the timings in your simulation; it
1010 still executes the computations itself. If you want to stop SMPI from
1011 doing that, you should check the SMPI_SAMPLE macros, documented in
1012 Section :ref:`SMPI_adapting_speed`.
1014 +------------------------------------+-------------------------+-----------------------------+
1015 | Solution | Computations executed? | Computations simulated? |
1016 +====================================+=========================+=============================+
1017 | --cfg=smpi/simulate-computation:no | Yes | Never |
1018 +------------------------------------+-------------------------+-----------------------------+
1019 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
1020 +------------------------------------+-------------------------+-----------------------------+
1021 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
1022 +------------------------------------+-------------------------+-----------------------------+
1024 .. _cfg=smpi/comp-adjustment-file:
1026 Slow-down or speed-up parts of your code
1027 ........................................
1029 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
1031 This option allows you to pass a file that contains two columns: The
1032 first column defines the section that will be subject to a speedup;
1033 the second column is the speedup. For instance:
1035 .. code-block:: shell
1037 "start:stop","ratio"
1038 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1040 The first line is the header - you must include it. The following
1041 line means that the code between two consecutive MPI calls on line 30
1042 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1043 of 1.18244559422142. The value for the second column is therefore a
1044 speedup, if it is larger than 1 and a slow-down if it is smaller
1045 than 1. Nothing will be changed if it is equal to 1.
1047 Of course, you can set any arbitrary filenames you want (so the start
1048 and end don't have to be in the same file), but be aware that this
1049 mechanism only supports `consecutive calls!`
1051 Please note that you must pass the ``-trace-call-location`` flag to
1052 smpicc or smpiff, respectively. This flag activates some internal
1053 macro definitions that help with obtaining the call location.
1055 .. _cfg=smpi/bw-factor:
1060 **Option** ``smpi/bw-factor``
1061 |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
1063 The possible throughput of network links is often dependent on the
1064 message sizes, as protocols may adapt to different message sizes. With
1065 this option, a series of message sizes and factors are given, helping
1066 the simulation to be more realistic. For instance, the current default
1067 value means that messages with size 65472 and more will get a total of
1068 MAX_BANDWIDTH*0.940694, messages of size 15424 to 65471 will get
1069 MAX_BANDWIDTH*0.697866 and so on (where MAX_BANDWIDTH denotes the
1070 bandwidth of the link).
1072 An experimental script to compute these factors is available online. See
1073 http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
1074 http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
1076 .. _cfg=smpi/display-timing:
1078 Reporting Simulation Time
1079 .........................
1081 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1083 Most of the time, you run MPI code with SMPI to compute the time it
1084 would take to run it on a platform. But since the code is run through
1085 the ``smpirun`` script, you don't have any control on the launcher
1086 code, making it difficult to report the simulated time when the
1087 simulation ends. If you enable the ``smpi/display-timing`` item,
1088 ``smpirun`` will display this information when the simulation
1091 .. _cfg=smpi/keep-temps:
1093 Keeping temporary files after simulation
1094 ........................................
1096 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1098 SMPI usually generates a lot of temporary files that are cleaned after
1099 use. This option request to preserve them, for example to debug or
1100 profile your code. Indeed, the binary files are removed very early
1101 under the dlopen privatization schema, which tend to fool the
1104 .. _cfg=smpi/lat-factor:
1109 **Option** ``smpi/lat-factor`` |br|
1110 **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
1112 The motivation and syntax for this option is identical to the motivation/syntax
1113 of :ref:`cfg=smpi/bw-factor`.
1115 There is an important difference, though: While smpi/bw-factor `reduces` the
1116 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
1117 increase the latency, i.e., values larger than or equal to 1 are valid here.
1119 .. _cfg=smpi/papi-events:
1121 Trace hardware counters with PAPI
1122 .................................
1124 **Option** ``smpi/papi-events`` **default:** unset
1126 When the PAPI support was compiled in SimGrid, this option takes the
1127 names of PAPI counters and adds their respective values to the trace
1128 files (See Section :ref:`tracing_tracing_options`).
1132 This feature currently requires superuser privileges, as registers
1133 are queried. Only use this feature with code you trust! Call
1134 smpirun for instance via ``smpirun -wrapper "sudo "
1135 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1136 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1137 will not be required.
1139 It is planned to make this feature available on a per-process (or per-thread?) basis.
1140 The first draft, however, just implements a "global" (i.e., for all processes) set
1141 of counters, the "default" set.
1143 .. code-block:: shell
1145 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1147 .. _cfg=smpi/privatization:
1149 Automatic Privatization of Global Variables
1150 ...........................................
1152 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1154 MPI executables are usually meant to be executed in separated
1155 processes, but SMPI is executed in only one process. Global variables
1156 from executables will be placed in the same memory zone and shared
1157 between processes, causing intricate bugs. Several options are
1158 possible to avoid this, as described in the main `SMPI publication
1159 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1160 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1161 automatically privatizing the globals, and this option allows to
1162 choose between them.
1164 - **no** (default when not using smpirun): Do not automatically
1165 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1167 - **dlopen** or **yes** (default when using smpirun): Link multiple
1168 times against the binary.
1169 - **mmap** (slower, but maybe somewhat more stable):
1170 Runtime automatic switching of the data segments.
1173 This configuration option cannot be set in your platform file. You can only
1174 pass it as an argument to smpirun.
1176 .. _cfg=smpi/privatize-libs:
1178 Automatic privatization of global variables inside external libraries
1179 .....................................................................
1181 **Option** ``smpi/privatize-libs`` **default:** unset
1183 **Linux/BSD only:** When using dlopen (default) privatization,
1184 privatize specific shared libraries with internal global variables, if
1185 they can't be linked statically. For example libgfortran is usually
1186 used for Fortran I/O and indexes in files can be mixed up.
1188 Multiple libraries can be given, semicolon separated.
1190 This configuration option can only use either full paths to libraries,
1191 or full names. Check with ldd the name of the library you want to
1194 .. code-block:: shell
1198 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1201 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1202 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1203 but not ``libgfortran`` nor ``libgfortran.so``.
1205 .. _cfg=smpi/send-is-detached-thresh:
1207 Simulating MPI detached send
1208 ............................
1210 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1212 This threshold specifies the size in bytes under which the send will
1213 return immediately. This is different from the threshold detailed in
1214 :ref:`options_model_network_asyncsend` because the message is not
1215 effectively sent when the send is posted. SMPI still waits for the
1216 correspondant receive to be posted to perform the communication
1219 .. _cfg=smpi/coll-selector:
1221 Simulating MPI collective algorithms
1222 ....................................
1224 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1226 SMPI implements more than 100 different algorithms for MPI collective
1227 communication, to accurately simulate the behavior of most of the
1228 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1229 use the decision logic of either OpenMPI or MPICH libraries (by
1230 default SMPI uses naive version of collective operations).
1232 Each collective operation can be manually selected with a
1233 ``smpi/collective_name:algo_name``. Available algorithms are listed in
1234 :ref:`SMPI_use_colls`.
1236 .. TODO:: All available collective algorithms will be made available
1237 via the ``smpirun --help-coll`` command.
1239 .. _cfg=smpi/iprobe:
1241 Inject constant times for MPI_Iprobe
1242 ....................................
1244 **Option** ``smpi/iprobe`` **default:** 0.0001
1246 The behavior and motivation for this configuration option is identical
1247 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1250 .. _cfg=smpi/iprobe-cpu-usage:
1252 Reduce speed for iprobe calls
1253 .............................
1255 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1257 MPI_Iprobe calls can be heavily used in applications. To account
1258 correctly for the energy cores spend probing, it is necessary to
1259 reduce the load that these calls cause inside SimGrid.
1261 For instance, we measured a max power consumption of 220 W for a
1262 particular application but only 180 W while this application was
1263 probing. Hence, the correct factor that should be passed to this
1264 option would be 180/220 = 0.81.
1268 Inject constant times for MPI_Init
1269 ..................................
1271 **Option** ``smpi/init`` **default:** 0
1273 The behavior and motivation for this configuration option is identical
1274 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1278 Inject constant times for MPI_Isend()
1279 .....................................
1281 **Option** ``smpi/ois``
1283 The behavior and motivation for this configuration option is identical
1284 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1288 Inject constant times for MPI_send()
1289 ....................................
1291 **Option** ``smpi/os``
1293 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1294 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1295 time). SMPI can factor these costs in as well, but the user has to
1296 configure SMPI accordingly as these values may vary by machine. This
1297 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1298 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1299 exactly as ``smpi/ois``.
1301 This item can consist of multiple sections; each section takes three
1302 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1303 so this example contains two sections. Furthermore, each section
1304 consists of three values.
1306 1. The first value denotes the minimum size for this section to take effect;
1307 read it as "if message size is greater than this value (and other section has a larger
1308 first value that is also smaller than the message size), use this".
1309 In the first section above, this value is "1".
1311 2. The second value is the startup time; this is a constant value that will always
1312 be charged, no matter what the size of the message. In the first section above,
1315 3. The third value is the `per-byte` cost. That is, it is charged for every
1316 byte of the message (incurring cost messageSize*cost_per_byte)
1317 and hence accounts also for larger messages. In the first
1318 section of the example above, this value is "2".
1320 Now, SMPI always checks which section it should take for a given
1321 message; that is, if a message of size 11 is sent with the
1322 configuration of the example above, only the second section will be
1323 used, not the first, as the first value of the second section is
1324 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1325 message of size 11 incurs the following cost inside MPI_Send:
1326 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1328 Note that the order of sections can be arbitrary; they will be ordered internally.
1332 Inject constant times for MPI_Recv()
1333 ....................................
1335 **Option** ``smpi/or``
1337 The behavior and motivation for this configuration option is identical
1338 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1341 .. _cfg=smpi/grow-injected-times:
1343 Inject constant times for MPI_Test
1344 ..................................
1346 **Option** ``smpi/test`` **default:** 0.0001
1348 By setting this option, you can control the amount of time a process
1349 sleeps when MPI_Test() is called; this is important, because SimGrid
1350 normally only advances the time while communication is happening and
1351 thus, MPI_Test will not add to the time, resulting in a deadlock if
1352 used as a break-condition as in the following example:
1357 MPI_Test(request, flag, status);
1361 To speed up execution, we use a counter to keep track on how often we
1362 already checked if the handle is now valid or not. Hence, we actually
1363 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1364 process to sleep increases linearly with the number of previously
1365 failed tests. This behavior can be disabled by setting
1366 ``smpi/grow-injected-times`` to **no**. This will also disable this
1367 behavior for MPI_Iprobe.
1369 .. _cfg=smpi/shared-malloc:
1370 .. _cfg=smpi/shared-malloc-hugepage:
1375 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1377 If your simulation consumes too much memory, you may want to modify
1378 your code so that the working areas are shared by all MPI ranks. For
1379 example, in a bloc-cyclic matrix multiplication, you will only
1380 allocate one set of blocs, and every processes will share them.
1381 Naturally, this will lead to very wrong results, but this will save a
1382 lot of memory so this is still desirable for some studies. For more on
1383 the motivation for that feature, please refer to the `relevant section
1384 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1385 of the SMPI CourseWare (see Activity #2.2 of the pointed
1386 assignment). In practice, change the call to malloc() and free() into
1387 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1389 SMPI provides two algorithms for this feature. The first one, called
1390 ``local``, allocates one bloc per call to SMPI_SHARED_MALLOC() in your
1391 code (each call location gets its own bloc) and this bloc is shared
1392 amongst all MPI ranks. This is implemented with the shm_* functions
1393 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1394 for each shared bloc.
1396 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1397 returns a new adress, but it only points to a shadow bloc: its memory
1398 area is mapped on a 1MiB file on disk. If the returned bloc is of size
1399 N MiB, then the same file is mapped N times to cover the whole bloc.
1400 At the end, no matter how many SMPI_SHARED_MALLOC you do, this will
1401 only consume 1 MiB in memory.
1403 You can disable this behavior and come back to regular mallocs (for
1404 example for debugging purposes) using @c "no" as a value.
1406 If you want to keep private some parts of the buffer, for instance if these
1407 parts are used by the application logic and should not be corrupted, you
1408 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). Example:
1412 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1414 This will allocate 500 bytes to mem, such that mem[27..41] and
1415 mem[100..199] are shared while other area remain private.
1417 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1419 When smpi/shared-malloc:global is used, the memory consumption problem
1420 is solved, but it may induce too much load on the kernel's pages table.
1421 In this case, you should use huge pages so that we create only one
1422 entry per Mb of malloced data instead of one entry per 4k.
1423 To activate this, you must mount a hugetlbfs on your system and allocate
1424 at least one huge page:
1426 .. code-block:: shell
1429 sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1430 sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1432 Then, you can pass the option
1433 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1434 actually activate the huge page support in shared mallocs.
1438 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1439 ...................................................................
1441 **Option** ``smpi/wtime`` **default:** 10 ns
1443 This option controls the amount of (simulated) time spent in calls to
1444 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1445 to 0, the simulated clock is not advanced in these calls, which leads
1446 to issue if your application contains such a loop:
1450 while(MPI_Wtime() < some_time_bound) {
1451 /* some tests, with no communication nor computation */
1454 When the option smpi/wtime is set to 0, the time advances only on
1455 communications and computations, so the previous code results in an
1456 infinite loop: the current [simulated] time will never reach
1457 ``some_time_bound``. This infinite loop is avoided when that option
1458 is set to a small amount, as it is by default since SimGrid v3.21.
1460 Note that if your application does not contain any loop depending on
1461 the current time only, then setting this option to a non-zero value
1462 will slow down your simulations by a tiny bit: the simulation loop has
1463 to be broken and reset each time your code ask for the current time.
1464 If the simulation speed really matters to you, you can avoid this
1465 extra delay by setting smpi/wtime to 0.
1467 Other Configurations
1468 --------------------
1470 .. _cfg=clean-atexit:
1472 Cleanup at Termination
1473 ......................
1475 **Option** ``clean-atexit`` **default:** on
1477 If your code is segfaulting during its finalization, it may help to
1478 disable this option to request SimGrid to not attempt any cleanups at
1479 the end of the simulation. Since the Unix process is ending anyway,
1480 the operating system will wipe it all.
1487 **Option** ``path`` **default:** . (current dir)
1489 It is possible to specify a list of directories to search into for the
1490 trace files (see :ref:`pf_trace`) by using this configuration
1491 item. To add several directory to the path, set the configuration
1492 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1494 .. _cfg=simix/breakpoint:
1499 **Option** ``simix/breakpoint`` **default:** unset
1501 This configuration option sets a breakpoint: when the simulated clock
1502 reaches the given time, a SIGTRAP is raised. This can be used to stop
1503 the execution and get a backtrace with a debugger.
1505 It is also possible to set the breakpoint from inside the debugger, by
1506 writing in global variable simgrid::simix::breakpoint. For example,
1509 .. code-block:: shell
1511 set variable simgrid::simix::breakpoint = 3.1416
1513 .. _cfg=verbose-exit:
1518 **Option** ``verbose-exit`` **default:** on
1520 By default, when Ctrl-C is pressed, the status of all existing actors
1521 is displayed before exiting the simulation. This is very useful to
1522 debug your code, but it can reveal troublesome if you have many
1523 actors. Set this configuration item to **off** to disable this
1526 .. _cfg=exception/cutpath:
1528 Truncate local path from exception backtrace
1529 ............................................
1531 **Option** ``exception/cutpath`` **default:** off
1533 This configuration option is used to remove the path from the
1534 backtrace shown when an exception is thrown. This is mainly useful for
1535 the tests: the full file path makes the tests not reproducible because
1536 the path of source files depend of the build settings. That would
1537 break most of our tests as we keep comparing output.
1539 Logging Configuration
1540 ---------------------
1542 It can be done by using XBT. Go to :ref:`XBT_log` for more details.