8 <object id="TOC" data="graphical-toc.svg" type="image/svg+xml"></object>
10 window.onload=function() { // Wait for the SVG to be loaded before changing it
11 var elem=document.querySelector("#TOC").contentDocument.getElementById("ConfigBox")
12 elem.style="opacity:0.93999999;fill:#ff0000;fill-opacity:0.1;stroke:#000000;stroke-width:0.35277778;stroke-linecap:round;stroke-linejoin:round;stroke-miterlimit:4;stroke-dasharray:none;stroke-dashoffset:0;stroke-opacity:1";
18 A number of options can be given at runtime to change the default
19 SimGrid behavior. For a complete list of all configuration options
20 accepted by the SimGrid version used in your simulator, simply pass
21 the --help configuration flag to your program. If some of the options
22 are not documented on this page, this is a bug that you should please
23 report so that we can fix it. Note that some of the options presented
24 here may not be available in your simulators, depending on the
25 :ref:`compile-time options <install_src_config>` that you used.
27 Setting Configuration Items
28 ---------------------------
30 There is several way to pass configuration options to the simulators.
31 The most common way is to use the ``--cfg`` command line argument. For
32 example, to set the item ``Item`` to the value ``Value``, simply
33 type the following on the command-line:
35 .. code-block:: console
37 $ my_simulator --cfg=Item:Value (other arguments)
39 Several ``--cfg`` command line arguments can naturally be used. If you
40 need to include spaces in the argument, don't forget to quote the
41 argument. You can even escape the included quotes (write ``@'`` for ``'`` if
42 you have your argument between simple quotes).
44 Another solution is to use the ``<config>`` tag in the platform file. The
45 only restriction is that this tag must occur before the first
46 platform element (be it ``<zone>``, ``<cluster>``, ``<peer>`` or whatever).
47 The ``<config>`` tag takes an ``id`` attribute, but it is currently
48 ignored so you don't really need to pass it. The important part is that
49 within that tag, you can pass one or several ``<prop>`` tags to specify
50 the configuration to use. For example, setting ``Item`` to ``Value``
51 can be done by adding the following to the beginning of your platform
57 <prop id="Item" value="Value"/>
60 A last solution is to pass your configuration directly in your program
61 with :cpp:func:`simgrid::s4u::Engine::set_config` or :cpp:func:`MSG_config`.
65 #include <simgrid/s4u.hpp>
67 int main(int argc, char *argv[]) {
68 simgrid::s4u::Engine e(&argc, argv);
70 simgrid::s4u::Engine::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/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 - **debug/breakpoint:** :ref:`cfg=debug/breakpoint`
98 - **debug/clean-atexit:** :ref:`cfg=debug/clean-atexit`
99 - **debug/verbose-exit:** :ref:`cfg=debug/verbose-exit`
101 - **exception/cutpath:** :ref:`cfg=exception/cutpath`
103 - **host/model:** :ref:`options_model_select`
105 - **maxmin/precision:** :ref:`cfg=maxmin/precision`
106 - **maxmin/concurrency-limit:** :ref:`cfg=maxmin/concurrency-limit`
108 - **msg/debug-multiple-use:** :ref:`cfg=msg/debug-multiple-use`
110 - **model-check:** :ref:`options_modelchecking`
111 - **model-check/checkpoint:** :ref:`cfg=model-check/checkpoint`
112 - **model-check/communications-determinism:** :ref:`cfg=model-check/communications-determinism`
113 - **model-check/dot-output:** :ref:`cfg=model-check/dot-output`
114 - **model-check/max-depth:** :ref:`cfg=model-check/max-depth`
115 - **model-check/property:** :ref:`cfg=model-check/property`
116 - **model-check/reduction:** :ref:`cfg=model-check/reduction`
117 - **model-check/replay:** :ref:`cfg=model-check/replay`
118 - **model-check/send-determinism:** :ref:`cfg=model-check/send-determinism`
119 - **model-check/termination:** :ref:`cfg=model-check/termination`
120 - **model-check/timeout:** :ref:`cfg=model-check/timeout`
121 - **model-check/visited:** :ref:`cfg=model-check/visited`
123 - **network/bandwidth-factor:** :ref:`cfg=network/bandwidth-factor`
124 - **network/crosstraffic:** :ref:`cfg=network/crosstraffic`
125 - **network/latency-factor:** :ref:`cfg=network/latency-factor`
126 - **network/loopback-lat:** :ref:`cfg=network/loopback`
127 - **network/loopback-bw:** :ref:`cfg=network/loopback`
128 - **network/maxmin-selective-update:** :ref:`Network Optimization Level <options_model_optim>`
129 - **network/model:** :ref:`options_model_select`
130 - **network/optim:** :ref:`Network Optimization Level <options_model_optim>`
131 - **network/TCP-gamma:** :ref:`cfg=network/TCP-gamma`
132 - **network/weight-S:** :ref:`cfg=network/weight-S`
134 - **ns3/TcpModel:** :ref:`options_pls`
135 - **ns3/seed:** :ref:`options_pls`
136 - **path:** :ref:`cfg=path`
137 - **plugin:** :ref:`cfg=plugin`
139 - **storage/max_file_descriptors:** :ref:`cfg=storage/max_file_descriptors`
141 - **surf/precision:** :ref:`cfg=surf/precision`
143 - **For collective operations of SMPI,** please refer to Section :ref:`cfg=smpi/coll-selector`
144 - **smpi/auto-shared-malloc-thresh:** :ref:`cfg=smpi/auto-shared-malloc-thresh`
145 - **smpi/async-small-thresh:** :ref:`cfg=smpi/async-small-thresh`
146 - **smpi/buffering:** :ref:`cfg=smpi/buffering`
147 - **smpi/bw-factor:** :ref:`cfg=smpi/bw-factor`
148 - **smpi/coll-selector:** :ref:`cfg=smpi/coll-selector`
149 - **smpi/comp-adjustment-file:** :ref:`cfg=smpi/comp-adjustment-file`
150 - **smpi/cpu-threshold:** :ref:`cfg=smpi/cpu-threshold`
151 - **smpi/display-allocs:** :ref:`cfg=smpi/display-allocs`
152 - **smpi/display-timing:** :ref:`cfg=smpi/display-timing`
153 - **smpi/errors-are-fatal:** :ref:`cfg=smpi/errors-are-fatal`
154 - **smpi/finalization-barrier:** :ref:`cfg=smpi/finalization-barrier`
155 - **smpi/grow-injected-times:** :ref:`cfg=smpi/grow-injected-times`
156 - **smpi/host-speed:** :ref:`cfg=smpi/host-speed`
157 - **smpi/IB-penalty-factors:** :ref:`cfg=smpi/IB-penalty-factors`
158 - **smpi/iprobe:** :ref:`cfg=smpi/iprobe`
159 - **smpi/iprobe-cpu-usage:** :ref:`cfg=smpi/iprobe-cpu-usage`
160 - **smpi/init:** :ref:`cfg=smpi/init`
161 - **smpi/keep-temps:** :ref:`cfg=smpi/keep-temps`
162 - **smpi/lat-factor:** :ref:`cfg=smpi/lat-factor`
163 - **smpi/ois:** :ref:`cfg=smpi/ois`
164 - **smpi/or:** :ref:`cfg=smpi/or`
165 - **smpi/os:** :ref:`cfg=smpi/os`
166 - **smpi/papi-events:** :ref:`cfg=smpi/papi-events`
167 - **smpi/pedantic:** :ref:`cfg=smpi/pedantic`
168 - **smpi/privatization:** :ref:`cfg=smpi/privatization`
169 - **smpi/privatize-libs:** :ref:`cfg=smpi/privatize-libs`
170 - **smpi/send-is-detached-thresh:** :ref:`cfg=smpi/send-is-detached-thresh`
171 - **smpi/shared-malloc:** :ref:`cfg=smpi/shared-malloc`
172 - **smpi/shared-malloc-hugepage:** :ref:`cfg=smpi/shared-malloc-hugepage`
173 - **smpi/simulate-computation:** :ref:`cfg=smpi/simulate-computation`
174 - **smpi/test:** :ref:`cfg=smpi/test`
175 - **smpi/wtime:** :ref:`cfg=smpi/wtime`
176 - **smpi/list-leaks** :ref:`cfg=smpi/list-leaks`
178 - **Tracing configuration options** can be found in Section :ref:`tracing_tracing_options`
180 - **storage/model:** :ref:`options_model_select`
182 - **vm/model:** :ref:`options_model_select`
186 Configuring the Platform Models
187 -------------------------------
189 .. _options_model_select:
191 Choosing the Platform Models
192 ............................
194 SimGrid comes with several network, CPU and disk models built in,
195 and you can change the used model at runtime by changing the passed
196 configuration. The three main configuration items are given below.
197 For each of these items, passing the special ``help`` value gives you
198 a short description of all possible values (for example,
199 ``--cfg=network/model:help`` will present all provided network
200 models). Also, ``--help-models`` should provide information about all
201 models for all existing resources.
203 - ``network/model``: specify the used network model. Possible values:
205 - **LV08 (default one):** Realistic network analytic model
206 (slow-start modeled by multiplying latency by 13.01, bandwidth by
207 .97; bottleneck sharing uses a payload of S=20537 for evaluating
208 RTT). Described in `Accuracy Study and Improvement of Network
209 Simulation in the SimGrid Framework
210 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
211 - **Constant:** Simplistic network model where all communication
212 take a constant time (one second). This model provides the lowest
213 realism, but is (marginally) faster.
214 - **SMPI:** Realistic network model specifically tailored for HPC
215 settings (accurate modeling of slow start with correction factors on
216 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). This model can be
217 :ref:`further configured <options_model_network>`.
218 - **IB:** Realistic network model specifically tailored for HPC
219 settings with InfiniBand networks (accurate modeling contention
220 behavior, based on the model explained in `this PhD work
221 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
222 This model can be :ref:`further configured <options_model_network>`.
223 - **CM02:** Legacy network analytic model. Very similar to LV08, but
224 without corrective factors. The timings of small messages are thus
225 poorly modeled. This model is described in `A Network Model for
226 Simulation of Grid Application
227 <https://hal.inria.fr/inria-00071989/document>`_.
228 - **ns-3** (only available if you compiled SimGrid accordingly):
229 Use the packet-level network
230 simulators as network models (see :ref:`model_ns3`).
231 This model can be :ref:`further configured <options_pls>`.
233 - ``cpu/model``: specify the used CPU model. We have only one model
236 - **Cas01:** Simplistic CPU model (time=size/speed)
238 - ``host/model``: The host concept is the aggregation of a CPU with a
239 network card. Three models exists, but actually, only 2 of them are
240 interesting. The "compound" one is simply due to the way our
241 internal code is organized, and can easily be ignored. So at the
242 end, you have two host models: The default one allows aggregation of
243 an existing CPU model with an existing network model, but does not
244 allow parallel tasks because these beasts need some collaboration
245 between the network and CPU model.
247 - **default:** Default host model. Currently, CPU:Cas01 and
248 network:LV08 (with cross traffic enabled)
249 - **compound:** Host model that is automatically chosen if
250 you change the network and CPU models
251 - **ptask_L07:** Host model somehow similar to Cas01+CM02 but
252 allowing "parallel tasks", that are intended to model the moldable
253 tasks of the grid scheduling literature.
254 - **ptask_BMF:** More realistic model for heterogeneous resource sharing.
255 Implements BMF (Bottleneck max fairness) fairness. To be used with
256 parallel tasks instead of ptask_L07.
258 - ``storage/model``: specify the used storage model. Only one model is
260 - ``vm/model``: specify the model for virtual machines. Only one model
263 .. todo: make 'compound' the default host model.
265 .. _options_model_optim:
270 The network and CPU models that are based on lmm_solve (that
271 is, all our analytical models) accept specific optimization
274 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
276 - **Lazy:** Lazy action management (partial invalidation in lmm +
277 heap in action remaining).
278 - **TI:** Trace integration. Highly optimized mode when using
279 availability traces (only available for the Cas01 CPU model for
281 - **Full:** Full update of remaining and variables. Slow but may be
282 useful when debugging.
284 - items ``network/maxmin-selective-update`` and
285 ``cpu/maxmin-selective-update``: configure whether the underlying
286 should be lazily updated or not. It should have no impact on the
287 computed timings, but should speed up the computation. |br| It is
288 still possible to disable this feature because it can reveal
289 counter-productive in very specific scenarios where the
290 interaction level is high. In particular, if all your
291 communication share a given backbone link, you should disable it:
292 without it, a simple regular loop is used to update each
293 communication. With it, each of them is still updated (because of
294 the dependency induced by the backbone), but through a complicated
295 and slow pattern that follows the actual dependencies.
297 .. _cfg=maxmin/precision:
298 .. _cfg=surf/precision:
303 **Option** ``maxmin/precision`` **Default:** 0.00001 (in flops or bytes) |br|
304 **Option** ``surf/precision`` **Default:** 0.00001 (in seconds)
306 The analytical models handle a lot of floating point values. It is
307 possible to change the epsilon used to update and compare them through
308 this configuration item. Changing it may speedup the simulation by
309 discarding very small actions, at the price of a reduced numerical
310 precision. You can modify separately the precision used to manipulate
311 timings (in seconds) and the one used to manipulate amounts of work
314 .. _cfg=maxmin/concurrency-limit:
319 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
321 The maximum number of variables per resource can be tuned through this
322 option. You can have as many simultaneous actions per resources as you
323 want. If your simulation presents a very high level of concurrency, it
324 may help to use e.g. 100 as a value here. It means that at most 100
325 actions can consume a resource at a given time. The extraneous actions
326 are queued and wait until the amount of concurrency of the considered
327 resource lowers under the given boundary.
329 Such limitations help both to the simulation speed and simulation accuracy
330 on highly constrained scenarios, but the simulation speed suffers of this
331 setting on regular (less constrained) scenarios so it is off by default.
333 .. _options_model_network:
335 Configuring the Network Model
336 .............................
338 .. _cfg=network/TCP-gamma:
340 Maximal TCP Window Size
341 ^^^^^^^^^^^^^^^^^^^^^^^
343 **Option** ``network/TCP-gamma`` **Default:** 4194304
345 The analytical models need to know the maximal TCP window size to take
346 the TCP congestion mechanism into account. On Linux, this value can
347 be retrieved using the following commands. Both give a set of values,
348 and you should use the last one, which is the maximal size.
350 .. code-block:: console
352 $ cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
353 $ cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
355 .. _cfg=network/bandwidth-factor:
356 .. _cfg=network/latency-factor:
357 .. _cfg=network/weight-S:
359 Correcting Important Network Parameters
360 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
362 SimGrid can take network irregularities such as a slow startup or
363 changing behavior depending on the message size into account. You
364 should not change these values unless you really know what you're
365 doing. The corresponding values were computed through data fitting
366 one the timings of packet-level simulators, as described in `Accuracy
367 Study and Improvement of Network Simulation in the SimGrid Framework
368 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
370 - **network/latency-factor**: apply a multiplier to latency.
371 Models the TCP slow-start mechanism.
372 - **network/bandwidth-factor**: actual bandwidth perceived by the
374 - **network/weight-S**: bottleneck sharing constant parameter. Used
377 These parameters are the same for all communications in your simulation,
378 independently of message size or source/destination hosts. A more flexible
379 mechanism based on callbacks was introduced in SimGrid. It provides the user
380 a callback that will be called for each communication, allowing the user
381 to set different latency and bandwidth factors, based on the message size, links used
382 or zones traversed. To more details of how to use it, please look at the
383 `examples/cpp/network-factors/s4u-network-factors.cpp <https://framagit.org/simgrid/simgrid/tree/master/examples/cpp/network-factors/s4u-network-factors.cpp>`_.
386 If you are using the SMPI model, these correction coefficients are
387 themselves corrected by constant values depending on the size of the
388 exchange. By default SMPI uses factors computed on the Stampede
389 Supercomputer at TACC, with optimal deployment of processes on
390 nodes. Again, only hardcore experts should bother about this fact.
391 For more details, see SMPI sections about :ref:`cfg=smpi/bw-factor` and :ref:`cfg=smpi/lat-factor`.
394 .. _cfg=smpi/IB-penalty-factors:
399 InfiniBand network behavior can be modeled through 3 parameters
400 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `this PhD
402 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_ (in French)
403 or more concisely in `this paper <https://hal.inria.fr/hal-00953618/document>`_,
404 even if that paper does only describe models for myrinet and ethernet.
405 You can see in Fig 2 some results for Infiniband, for example. This model
406 may be outdated by now for modern infiniband, anyway, so a new
407 validation would be good.
409 The three paramaters are defined as follows:
411 - βs: penalty factor for outgoing messages, computed by running a simple send to
412 two nodes and checking slowdown compared to a single send to one node,
414 - βe: penalty factor for ingoing messages, same computation method but with one
415 node receiving several messages
416 - γr: slowdown factor when communication buffer memory is saturated. It needs a
417 more complicated pattern to run in order to be computed (5.3 in the thesis,
418 page 107), and formula in the end is γr = time(c)/(3×βe×time(ref)), where
419 time(ref) is the time of a single comm with no contention).
421 Once these values are computed, a penalty is assessed for each message (this is
422 the part implemented in the simulator) as shown page 106 of the thesis. Here is
423 a simple translation of this text. First, some notations:
425 - ∆e(e) which corresponds to the incoming degree of node e, that is to say the number of communications having as destination node e.
426 - ∆s (s) which corresponds to the degree outgoing from node s, that is to say the number of communications sent by node s.
427 - Φ (e) which corresponds to the number of communications destined for the node e but coming from a different node.
428 - Ω (s, e) which corresponds to the number of messages coming from node s to node e. If node e only receives communications from different nodes then Φ (e) = ∆e (e). On the other hand if, for example, there are three messages coming from node s and going from node e then Φ (e) 6 = ∆e (e) and Ω (s, e) = 3
430 To determine the penalty for a communication, two values need to be calculated. First, the penalty caused by the conflict in transmission, noted ps.
433 - if ∆s (i) = 1 then ps = 1.
434 - if ∆s (i) ≥ 2 and ∆e (i) ≥ 3 then ps = ∆s (i) × βs × γr
435 - else, ps = ∆s (i) × βs
438 Then, the penalty caused by the conflict in reception (noted pe) should be computed as follows:
440 - if ∆e (i) = 1 then pe = 1
441 - else, pe = Φ (e) × βe × Ω (s, e)
443 Finally, the penalty associated with the communication is:
444 p = max (ps ∈ s, pe)
446 .. _cfg=network/crosstraffic:
448 Simulating Cross-Traffic
449 ^^^^^^^^^^^^^^^^^^^^^^^^
451 Since SimGrid v3.7, cross-traffic effects can be taken into account in
452 analytical simulations. It means that ongoing and incoming
453 communication flows are treated independently. In addition, the LV08
454 model adds 0.05 of usage on the opposite direction for each new
455 created flow. This can be useful to simulate some important TCP
456 phenomena such as ack compression.
458 For that to work, your platform must have two links for each
459 pair of interconnected hosts. An example of usable platform is
460 available in ``examples/platforms/crosstraffic.xml``.
462 This is activated through the ``network/crosstraffic`` item, that
463 can be set to 0 (disable this feature) or 1 (enable it).
465 Note that with the default host model this option is activated by default.
467 .. _cfg=network/loopback:
469 Configuring loopback link
470 ^^^^^^^^^^^^^^^^^^^^^^^^^
472 Several network model provide an implicit loopback link to account for local
473 communication on a host. By default it has a 10GBps bandwidth and a null latency.
474 This can be changed with ``network/loopback-lat`` and ``network/loopback-bw``
477 .. _cfg=smpi/async-small-thresh:
479 Simulating Asynchronous Send
480 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
482 (this configuration item is experimental and may change or disappear)
484 It is possible to specify that messages below a certain size (in bytes) will be
485 sent as soon as the call to MPI_Send is issued, without waiting for
486 the correspondent receive. This threshold can be configured through
487 the ``smpi/async-small-thresh`` item. The default value is 0. This
488 behavior can also be manually set for mailboxes, by setting the
489 receiving mode of the mailbox with a call to
490 :cpp:func:`MSG_mailbox_set_async`. After this, all messages sent to
491 this mailbox will have this behavior regardless of the message size.
493 This value needs to be smaller than or equals to the threshold set at
494 :ref:`cfg=smpi/send-is-detached-thresh`, because asynchronous messages
495 are meant to be detached as well.
502 **Option** ``ns3/TcpModel`` **Default:** "default" (ns-3 default)
504 When using ns-3, there is an extra item ``ns3/TcpModel``, corresponding
505 to the ``ns3::TcpL4Protocol::SocketType`` configuration item in
506 ns-3. The only valid values (enforced on the SimGrid side) are
507 'default' (no change to the ns-3 configuration), 'NewReno' or 'Reno' or
510 **Option** ``ns3/seed`` **Default:** "" (don't set the seed in ns-3)
512 This option is the random seed to provide to ns-3 with
513 ``ns3::RngSeedManager::SetSeed`` and ``ns3::RngSeedManager::SetRun``.
515 If left blank, no seed is set in ns-3. If the value 'time' is
516 provided, the current amount of seconds since epoch is used as a seed.
517 Otherwise, the provided value must be a number to use as a seed.
519 Configuring the Storage model
520 .............................
522 .. _cfg=storage/max_file_descriptors:
524 File Descriptor Count per Host
525 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
527 **Option** ``storage/max_file_descriptors`` **Default:** 1024
529 Each host maintains a fixed-size array of its file descriptors. You
530 can change its size through this item to either enlarge it if your
531 application requires it or to reduce it to save memory space.
538 SimGrid plugins allow one to extend the framework without changing its
539 source code directly. Read the source code of the existing plugins to
540 learn how to do so (in ``src/plugins``), and ask your questions to the
541 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
542 that plugins usually register callbacks to some signals of interest.
543 If they need to store some information about a given object (Link, CPU
544 or Actor), they do so through the use of a dedicated object extension.
546 Some of the existing plugins can be activated from the command line,
547 meaning that you can activate them from the command line without any
548 modification to your simulation code. For example, you can activate
549 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
552 Here is a partial list of plugins that can be activated this way. You can get
553 the full list by passing ``--cfg=plugin:help`` to your simulator.
555 - :ref:`Host Energy <plugin_host_energy>`: models the energy dissipation of the compute units.
556 - :ref:`Link Energy <plugin_link_energy>`: models the energy dissipation of the network.
557 - :ref:`Host Load <plugin_host_load>`: monitors the load of the compute units.
559 .. _options_modelchecking:
561 Configuring the Model-Checking
562 ------------------------------
564 To enable SimGrid's model-checking support, the program should
565 be executed using the simgrid-mc wrapper:
567 .. code-block:: console
569 $ simgrid-mc ./my_program
571 Safety properties are expressed as assertions using the function
572 :cpp:func:`void MC_assert(int prop)`.
574 .. _cfg=smpi/buffering:
576 Specifying the MPI buffering behavior
577 .....................................
579 **Option** ``smpi/buffering`` **Default:** infty
581 Buffering in MPI has a huge impact on the communication semantic. For example,
582 standard blocking sends are synchronous calls when the system buffers are full
583 while these calls can complete immediately without even requiring a matching
584 receive call for small messages sent when the system buffers are empty.
586 In SMPI, this depends on the message size, that is compared against two thresholds:
588 - if (size < :ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>`) then
589 MPI_Send returns immediately, even if the corresponding receive has not be issued yet.
590 - if (:ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>` < size < :ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>`) then
591 MPI_Send returns as soon as the corresponding receive has been issued. This is known as the eager mode.
592 - if (:ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>` < size) then
593 MPI_Send returns only when the message has actually been sent over the network. This is known as the rendez-vous mode.
595 The ``smpi/buffering`` (only valid with MC) option gives an easier interface to choose between these semantics. It can take two values:
597 - **zero:** means that buffering should be disabled. All communications are actually blocking.
598 - **infty:** means that buffering should be made infinite. All communications are non-blocking.
600 .. _cfg=model-check/property:
602 Specifying a liveness property
603 ..............................
605 **Option** ``model-check/property`` **Default:** unset
607 If you want to specify liveness properties, you have to pass them on
608 the command line, specifying the name of the file containing the
609 property, as formatted by the `ltl2ba <https://github.com/utwente-fmt/ltl2ba>`_ program.
610 Note that ltl2ba is not part of SimGrid and must be installed separately.
612 .. code-block:: console
614 $ simgrid-mc ./my_program --cfg=model-check/property:<filename>
616 .. _cfg=model-check/checkpoint:
618 Going for Stateful Verification
619 ...............................
621 By default, the system is backtracked to its initial state to explore
622 another path, instead of backtracking to the exact step before the fork
623 that we want to explore (this is called stateless verification). This
624 is done this way because saving intermediate states can rapidly
625 exhaust the available memory. If you want, you can change the value of
626 the ``model-check/checkpoint`` item. For example,
627 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
628 step. Beware, this will certainly explode your memory. Larger values
629 are probably better, make sure to experiment a bit to find the right
630 setting for your specific system.
632 .. _cfg=model-check/reduction:
634 Specifying the kind of reduction
635 ................................
637 The main issue when using the model-checking is the state space
638 explosion. You can activate some reduction technique with
639 ``--cfg=model-check/reduction:<technique>``. For now, this
640 configuration variable can take 2 values:
642 - **none:** Do not apply any kind of reduction (mandatory for
643 liveness properties, as our current DPOR algorithm breaks cycles)
644 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if
645 you verify local safety properties (default value for safety
648 Another way to mitigate the state space explosion is to search for
649 cycles in the exploration with the :ref:`cfg=model-check/visited`
650 configuration. Note that DPOR and state-equality reduction may not
651 play well together. You should choose between them.
653 Our current DPOR implementation could be improved in may ways. We are
654 currently improving its efficiency (both in term of reduction ability
655 and computational speed), and future work could make it compatible
656 with liveness properties.
658 .. _cfg=model-check/visited:
660 Size of Cycle Detection Set (state equality reduction)
661 ......................................................
663 Mc SimGrid can be asked to search for cycles during the exploration,
664 i.e. situations where a new explored state is in fact the same state
665 than a previous one.. This can prove useful to mitigate the state
666 space explosion with safety properties, and this is the crux when
667 searching for counter-examples to the liveness properties.
669 Note that this feature may break the current implementation of the
670 DPOR reduction technique.
672 The ``model-check/visited`` item is the maximum number of states, which
673 are stored in memory. If the maximum number of snapshotted state is
674 reached, some states will be removed from the memory and some cycles
675 might be missed. Small values can lead to incorrect verifications, but
676 large values can exhaust your memory and be CPU intensive as each new
677 state must be compared to that amount of older saved states.
679 The default settings depend on the kind of exploration. With safety
680 checking, no state is snapshotted and cycles cannot be detected. With
681 liveness checking, all states are snapshotted because missing a cycle
682 could hinder the exploration soundness.
684 .. _cfg=model-check/termination:
686 Non-Termination Detection
687 .........................
689 The ``model-check/termination`` configuration item can be used to
690 report if a non-termination execution path has been found. This is a
691 path with a cycle, which means that the program might never terminate.
693 This only works in safety mode, not in liveness mode.
695 This options is disabled by default.
697 .. _cfg=model-check/dot-output:
702 If set, the ``model-check/dot-output`` configuration item is the name
703 of a file in which to write a dot file of the path leading to the
704 property violation discovered (safety or liveness violation), as well
705 as the cycle for liveness properties. This dot file can then be fed to the
706 graphviz dot tool to generate a corresponding graphical representation.
708 .. _cfg=model-check/max-depth:
710 Exploration Depth Limit
711 .......................
713 The ``model-check/max-depth`` can set the maximum depth of the
714 exploration graph of the model checker. If this limit is reached, a
715 logging message is sent and the results might not be exact.
717 By default, the exploration is limited to the depth of 1000.
719 .. _cfg=model-check/timeout:
724 By default, the model checker does not handle timeout conditions: the `wait`
725 operations never time out. With the ``model-check/timeout`` configuration item
726 set to **yes**, the model checker will explore timeouts of `wait` operations.
728 .. _cfg=model-check/communications-determinism:
729 .. _cfg=model-check/send-determinism:
731 Communication Determinism
732 .........................
734 The ``model-check/communications-determinism`` and
735 ``model-check/send-determinism`` items can be used to select the
736 communication determinism mode of the model checker, which checks
737 determinism properties of the communications of an application.
741 Verification Performance Considerations
742 .......................................
744 The size of the stacks can have a huge impact on the memory
745 consumption when using model-checking. By default, each snapshot will
746 save a copy of the whole stacks and not only of the part that is
747 really meaningful: you should expect the contribution of the memory
748 consumption of the snapshots to be:
749 :math:`\text{number of processes} \times \text{stack size} \times \text{number of states}`.
751 When compiled against the model checker, the stacks are not
752 protected with guards: if the stack size is too small for your
753 application, the stack will silently overflow into other parts of the
754 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
756 .. _cfg=model-check/replay:
758 Replaying buggy execution paths from the model checker
759 ......................................................
761 Debugging the problems reported by the model checker is challenging:
762 First, the application under verification cannot be debugged with gdb
763 because the model checker already traces it. Then, the model checker may
764 explore several execution paths before encountering the issue, making it
765 very difficult to understand the output. Fortunately, SimGrid provides
766 the execution path leading to any reported issue so that you can replay
767 this path reported by the model checker, enabling the usage of classical
770 When the model checker finds an interesting path in the application
771 execution graph (where a safety or liveness property is violated), it
772 generates an identifier for this path. Here is an example of the output:
774 .. code-block:: console
776 [ 0.000000] (0:@) Check a safety property
777 [ 0.000000] (0:@) **************************
778 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
779 [ 0.000000] (0:@) **************************
780 [ 0.000000] (0:@) Counter-example execution trace:
781 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
782 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
783 [ 0.000000] (0:@) Path = 1/3;1/4
784 [ 0.000000] (0:@) Expanded states = 27
785 [ 0.000000] (0:@) Visited states = 68
786 [ 0.000000] (0:@) Executed transitions = 46
788 The interesting line is ``Path = 1/3;1/4``, which means that you should use
789 ``--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
790 execution path. All options (but the model checker related ones) must
791 remain the same. In particular, if you ran your application with
792 ``smpirun -wrapper simgrid-mc``, then do it again. Remove all
793 MC-related options, keep non-MC-related ones and add
794 ``--cfg=model-check/replay:???``.
796 Currently, if the path is of the form ``X;Y;Z``, each number denotes
797 the actor's pid that is selected at each indecision point. If it's of
798 the form ``X/a;Y/b``, the X and Y are the selected pids while the a
799 and b are the return values of their simcalls. In the previous
800 example, ``1/3;1/4``, you can see from the full output that the actor
801 1 is doing MC_RANDOM simcalls, so the 3 and 4 simply denote the values
802 that these simcall return on the execution branch leading to the
805 Configuring the User Code Virtualization
806 ----------------------------------------
808 .. _cfg=contexts/factory:
810 Selecting the Virtualization Factory
811 ....................................
813 **Option** contexts/factory **Default:** "raw"
815 In SimGrid, the user code is virtualized in a specific mechanism that
816 allows the simulation kernel to control its execution: when a user
817 process requires a blocking action (such as sending a message), it is
818 interrupted, and only gets released when the simulated clock reaches
819 the point where the blocking operation is done. This is explained
820 graphically in the `relevant tutorial, available online
821 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
823 In SimGrid, the containers in which user processes are virtualized are
824 called contexts. Several context factory are provided, and you can
825 select the one you want to use with the ``contexts/factory``
826 configuration item. Some of the following may not exist on your
827 machine because of portability issues. In any case, the default one
828 should be the most effcient one (please report bugs if the
829 auto-detection fails for you). They are approximately sorted here from
830 the slowest to the most efficient:
832 - **thread:** very slow factory using full featured threads (either
833 pthreads or windows native threads). They are slow but very
834 standard. Some debuggers or profilers only work with this factory.
835 - **java:** Java applications are virtualized onto java threads (that
836 are regular pthreads registered to the JVM)
837 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
838 - **boost:** This uses the `context
839 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
840 of the boost library for a performance that is comparable to our
842 |br| Install the relevant library (e.g. with the
843 libboost-contexts-dev package on Debian/Ubuntu) and recompile
845 - **raw:** amazingly fast factory using a context switching mechanism
846 of our own, directly implemented in assembly (only available for x86
847 and amd64 platforms for now) and without any unneeded system call.
849 The main reason to change this setting is when the debugging tools become
850 fooled by the optimized context factories. Threads are the most
851 debugging-friendly contexts, as they allow one to set breakpoints
852 anywhere with gdb and visualize backtraces for all processes, in order
853 to debug concurrency issues. Valgrind is also more comfortable with
854 threads, but it should be usable with all factories (Exception: the
855 callgrind tool really dislikes raw and ucontext factories).
857 .. _cfg=contexts/stack-size:
859 Adapting the Stack Size
860 .......................
862 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
864 Each virtualized used process is executed using a specific system
865 stack. The size of this stack has a huge impact on the simulation
866 scalability, but its default value is rather large. This is because
867 the error messages that you get when the stack size is too small are
868 rather disturbing: this leads to stack overflow (overwriting other
869 stacks), leading to segfaults with corrupted stack traces.
871 If you want to push the scalability limits of your code, you might
872 want to reduce the ``contexts/stack-size`` item. Its default value is
873 8192 (in KiB), while our Chord simulation works with stacks as small
874 as 16 KiB, for example. You can ensure that some actors have a specific
875 size by simply changing the value of this configuration item before
876 creating these actors. The :cpp:func:`simgrid::s4u::Engine::set_config`
877 functions are handy for that.
879 This *setting is ignored* when using the thread factory (because there
880 is no way to modify the stack size with C++ system threads). Instead,
881 you should compile SimGrid and your application with
882 ``-fsplit-stack``. Note that this compilation flag is not compatible
883 with the model checker right now.
885 The operating system should only allocate memory for the pages of the
886 stack which are actually used and you might not need to use this in
887 most cases. However, this setting is very important when using the
888 model checker (see :ref:`options_mc_perf`).
890 .. _cfg=contexts/guard-size:
892 Disabling Stack Guard Pages
893 ...........................
895 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages on Windows or with MC)
897 Unless you use the threads context factory (see
898 :ref:`cfg=contexts/factory`), a stack guard page is usually used
899 which prevents the stack of a given actor from overflowing on another
900 stack. But the performance impact may become prohibitive when the
901 amount of actors increases. The option ``contexts/guard-size`` is the
902 number of stack guard pages used. By setting it to 0, no guard pages
903 will be used: in this case, you should avoid using small stacks (with
904 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
905 will silently overflow on other parts of the memory.
907 When no stack guard page is created, stacks may then silently overflow
908 on other parts of the memory if their size is too small for the
911 .. _cfg=contexts/nthreads:
912 .. _cfg=contexts/synchro:
914 Running User Code in Parallel
915 .............................
917 Parallel execution of the user code is only considered stable in
918 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
919 simulations may well fail in parallel mode. It is described in
920 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
922 If you are using the **ucontext** or **raw** context factories, you can
923 request to execute the user code in parallel. Several threads are
924 launched, each of them handling the same number of user contexts at each
925 run. To activate this, set the ``contexts/nthreads`` item to the amount
926 of cores that you have in your computer (or lower than 1 to have the
927 amount of cores auto-detected).
929 When parallel execution is activated, you can choose the
930 synchronization schema used with the ``contexts/synchro`` item,
931 which value is either:
933 - **futex:** ultra optimized synchronisation schema, based on futexes
934 (fast user-mode mutexes), and thus only available on Linux systems.
935 This is the default mode when available.
936 - **posix:** slow but portable synchronisation using only POSIX
938 - **busy_wait:** not really a synchronisation: the worker threads
939 constantly request new contexts to execute. It should be the most
940 efficient synchronisation schema, but it loads all the cores of
941 your machine for no good reason. You probably prefer the other less
944 Configuring the Tracing
945 -----------------------
947 The :ref:`tracing subsystem <outcome_vizu>` can be configured in
948 several different ways depending on the used interface (S4U, SMPI)
949 and the kind of traces that needs to be obtained. See the
950 :ref:`Tracing Configuration Options subsection
951 <tracing_tracing_options>` for a full description of each
952 configuration option.
954 We detail here a simple way to get the traces working for you, even if
955 you never used the tracing API.
958 - Any SimGrid-based simulator (MSG, SMPI, ...) and raw traces:
962 --cfg=tracing:yes --cfg=tracing/uncategorized:yes
964 The first parameter activates the tracing subsystem, and the second
965 tells it to trace host and link utilization (without any
968 - MSG-based simulator and categorized traces (you need to
969 declare categories and classify your tasks according to them)
973 --cfg=tracing:yes --cfg=tracing/categorized:yes
975 The first parameter activates the tracing subsystem, and the second
976 tells it to trace host and link categorized utilization.
978 - SMPI simulator and traces for a space/time view:
980 .. code-block:: console
984 The `-trace` parameter for the smpirun script runs the simulation
985 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
986 smpirun's `-help` parameter for additional tracing options.
988 Sometimes you might want to put additional information on the trace to
989 correctly identify them later, or to provide data that can be used to
990 reproduce an experiment. You have two ways to do that:
992 - Add a string on top of the trace file as comment:
996 --cfg=tracing/comment:my_simulation_identifier
998 - Add the contents of a textual file on top of the trace file as comment:
1000 .. code-block:: none
1002 --cfg=tracing/comment-file:my_file_with_additional_information.txt
1004 Please, use these two parameters (for comments) to make reproducible
1005 simulations. For additional details about this and all tracing
1006 options, check See the :ref:`tracing_tracing_options`.
1011 .. _cfg=msg/debug-multiple-use:
1016 **Option** ``msg/debug-multiple-use`` **Default:** off
1018 Sometimes your application may try to send a task that is still being
1019 executed somewhere else, making it impossible to send this task. However,
1020 for debugging purposes, one may want to know what the other host is/was
1021 doing. This option shows a backtrace of the other process.
1026 The SMPI interface provides several specific configuration items.
1027 These are not easy to see, since the code is usually launched through the
1028 ``smiprun`` script directly.
1030 .. _cfg=smpi/host-speed:
1031 .. _cfg=smpi/cpu-threshold:
1032 .. _cfg=smpi/simulate-computation:
1034 Automatic Benchmarking of SMPI Code
1035 ...................................
1037 In SMPI, the sequential code is automatically benchmarked, and these
1038 computations are automatically reported to the simulator. That is to
1039 say that if you have a large computation between a ``MPI_Recv()`` and
1040 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
1041 this code, and create an execution task within the simulator to take
1042 this into account. For that, the actual duration is measured on the
1043 host machine and then scaled to the power of the corresponding
1044 simulated machine. The variable ``smpi/host-speed`` allows one to
1045 specify the computational speed of the host machine (in flop/s by
1046 default) to use when scaling the execution times.
1048 The default value is ``smpi/host-speed=20kf`` (= 20,000 flop/s). This
1049 is probably underestimated for most machines, leading SimGrid to
1050 overestimate the amount of flops in the execution blocks that are
1051 automatically injected in the simulator. As a result, the execution
1052 time of the whole application will probably be overestimated until you
1053 use a realistic value.
1055 When the code consists of numerous consecutive MPI calls, the
1056 previous mechanism feeds the simulation kernel with numerous tiny
1057 computations. The ``smpi/cpu-threshold`` item becomes handy when this
1058 impacts badly on the simulation performance. It specifies a threshold (in
1059 seconds) below which the execution chunks are not reported to the
1060 simulation kernel (default value: 1e-6).
1062 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
1063 time spent below this threshold. SMPI does not consider the
1064 `amount of time` of these computations; there is no offset for
1065 this. Hence, a value that is too small, may lead to unreliable
1068 In some cases, however, one may wish to disable simulation of
1069 the computation of an application. This is the case when SMPI is used not to
1070 simulate an MPI application, but instead an MPI code that performs
1071 "live replay" of another MPI app (e.g., ScalaTrace's replay tool, or
1072 various on-line simulators that run an app at scale). In this case the
1073 computation of the replay/simulation logic should not be simulated by
1074 SMPI. Instead, the replay tool or on-line simulator will issue
1075 "computation events", which correspond to the actual MPI simulation
1076 being replayed/simulated. At the moment, these computation events can
1077 be simulated using SMPI by calling internal smpi_execute*() functions.
1079 To disable the benchmarking/simulation of a computation in the simulated
1080 application, the variable ``smpi/simulate-computation`` should be set
1081 to **no**. This option just ignores the timings in your simulation; it
1082 still executes the computations itself. If you want to stop SMPI from
1083 doing that, you should check the SMPI_SAMPLE macros, documented in
1084 Section :ref:`SMPI_use_faster`.
1086 +------------------------------------+-------------------------+-----------------------------+
1087 | Solution | Computations executed? | Computations simulated? |
1088 +====================================+=========================+=============================+
1089 | --cfg=smpi/simulate-computation:no | Yes | Never |
1090 +------------------------------------+-------------------------+-----------------------------+
1091 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
1092 +------------------------------------+-------------------------+-----------------------------+
1093 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
1094 +------------------------------------+-------------------------+-----------------------------+
1096 .. _cfg=smpi/comp-adjustment-file:
1098 Slow-down or speed-up parts of your code
1099 ........................................
1101 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
1103 This option allows you to pass a file that contains two columns: The
1104 first column defines the section that will be subject to a speedup;
1105 the second column is the speedup. For instance:
1107 .. code-block:: none
1109 "start:stop","ratio"
1110 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1112 The first line is the header - you must include it. The following
1113 line means that the code between two consecutive MPI calls on line 30
1114 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1115 of 1.18244559422142. The value for the second column is therefore a
1116 speedup, if it is larger than 1 and a slowdown if it is smaller
1117 than 1. Nothing will be changed if it is equal to 1.
1119 Of course, you can set any arbitrary filenames you want (so the start
1120 and end don't have to be in the same file), but be aware that this
1121 mechanism only supports `consecutive calls!`
1123 Please note that you must pass the ``-trace-call-location`` flag to
1124 smpicc or smpiff, respectively. This flag activates some internal
1125 macro definitions that help with obtaining the call location.
1127 .. _cfg=smpi/bw-factor:
1132 **Option** ``smpi/bw-factor``
1133 |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
1135 The possible throughput of network links is often dependent on the
1136 message sizes, as protocols may adapt to different message sizes. With
1137 this option, a series of message sizes and factors are given, helping
1138 the simulation to be more realistic. For instance, the current default
1139 value means that messages with size 65472 bytes and more will get a total of
1140 MAX_BANDWIDTH*0.940694, messages of size 15424 to 65471 will get
1141 MAX_BANDWIDTH*0.697866, and so on (where MAX_BANDWIDTH denotes the
1142 bandwidth of the link).
1144 An experimental script to compute these factors is available online. See
1145 https://framagit.org/simgrid/platform-calibration/
1146 https://simgrid.org/contrib/smpi-saturation-doc.html
1148 .. _cfg=smpi/display-timing:
1150 Reporting Simulation Time
1151 .........................
1153 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1155 Most of the time, you run MPI code with SMPI to compute the time it
1156 would take to run it on a platform. But since the code is run through
1157 the ``smpirun`` script, you don't have any control on the launcher
1158 code, making it difficult to report the simulated time when the
1159 simulation ends. If you enable the ``smpi/display-timing`` item,
1160 ``smpirun`` will display this information when the simulation
1162 SMPI will also display information about the amout of real time spent
1163 in application code and in SMPI internals, to provide hints about the
1164 need to use sampling to reduce simulation time.
1166 .. _cfg=smpi/display-allocs:
1168 Reporting memory allocations
1169 ............................
1171 **Option** ``smpi/display-allocs`` **Default:** 0 (false)
1173 SMPI intercepts malloc and calloc calls performed inside the running
1174 application, if it wasn't compiled with SMPI_NO_OVERRIDE_MALLOC.
1175 With this option, SMPI will show at the end of execution the amount of
1176 memory allocated through these calls, and locate the most expensive one.
1177 This helps finding the targets for manual memory sharing, or the threshold
1178 to use for smpi/auto-shared-malloc-thresh option (see :ref:`cfg=smpi/auto-shared-malloc-thresh`).
1180 .. _cfg=smpi/keep-temps:
1182 Keeping temporary files after simulation
1183 ........................................
1185 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1187 SMPI usually generates a lot of temporary files that are cleaned after
1188 use. This option requests to preserve them, for example to debug or
1189 profile your code. Indeed, the binary files are removed very early
1190 under the dlopen privatization schema, which tends to fool the
1193 .. _cfg=smpi/lat-factor:
1198 **Option** ``smpi/lat-factor`` |br|
1199 **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
1201 The motivation and syntax for this option is identical to the motivation/syntax
1202 of :ref:`cfg=smpi/bw-factor`.
1204 There is an important difference, though: While smpi/bw-factor `reduces` the
1205 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
1206 increase the latency, i.e., values larger than or equal to 1 are valid here.
1208 .. _cfg=smpi/papi-events:
1210 Trace hardware counters with PAPI
1211 .................................
1213 **Option** ``smpi/papi-events`` **default:** unset
1215 When the PAPI support is compiled into SimGrid, this option takes the
1216 names of PAPI counters and adds their respective values to the trace
1217 files (See Section :ref:`tracing_tracing_options`).
1221 This feature currently requires superuser privileges, as registers
1222 are queried. Only use this feature with code you trust! Call
1223 smpirun for instance via ``smpirun -wrapper "sudo "
1224 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1225 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1226 will not be required.
1228 It is planned to make this feature available on a per-process (or per-thread?) basis.
1229 The first draft, however, just implements a "global" (i.e., for all processes) set
1230 of counters, the "default" set.
1232 .. code-block:: none
1234 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1236 .. _cfg=smpi/privatization:
1238 Automatic Privatization of Global Variables
1239 ...........................................
1241 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1243 MPI executables are usually meant to be executed in separate
1244 processes, but SMPI is executed in only one process. Global variables
1245 from executables will be placed in the same memory region and shared
1246 between processes, causing intricate bugs. Several options are
1247 possible to avoid this, as described in the main `SMPI publication
1248 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1249 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1250 automatically privatizing the globals, and this option allows one to
1251 choose between them.
1253 - **no** (default when not using smpirun): Do not automatically
1254 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1256 - **dlopen** or **yes** (default when using smpirun): Link multiple
1257 times against the binary.
1258 - **mmap** (slower, but maybe somewhat more stable):
1259 Runtime automatic switching of the data segments.
1262 This configuration option cannot be set in your platform file. You can only
1263 pass it as an argument to smpirun.
1265 .. _cfg=smpi/privatize-libs:
1267 Automatic privatization of global variables inside external libraries
1268 .....................................................................
1270 **Option** ``smpi/privatize-libs`` **default:** unset
1272 **Linux/BSD only:** When using dlopen (default) privatization,
1273 privatize specific shared libraries with internal global variables, if
1274 they can't be linked statically. For example libgfortran is usually
1275 used for Fortran I/O and indexes in files can be mixed up.
1277 Multiple libraries can be given, semicolon separated.
1279 This configuration option can only use either full paths to libraries,
1280 or full names. Check with ldd the name of the library you want to
1283 .. code-block:: console
1287 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1290 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1291 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1292 but not ``libgfortran`` nor ``libgfortran.so``.
1294 .. _cfg=smpi/send-is-detached-thresh:
1296 Simulating MPI detached send
1297 ............................
1299 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1301 This threshold specifies the size in bytes under which the send will
1302 return immediately. This is different from the threshold detailed in
1303 :ref:`cfg=smpi/async-small-thresh` because the message is not
1304 really sent when the send is posted. SMPI still waits for the
1305 corresponding receive to be posted, in order to perform the communication
1308 .. _cfg=smpi/coll-selector:
1310 Simulating MPI collective algorithms
1311 ....................................
1313 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1315 SMPI implements more than 100 different algorithms for MPI collective
1316 communication, to accurately simulate the behavior of most of the
1317 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1318 select the decision logic either of the OpenMPI or the MPICH libraries. (By
1319 default SMPI uses naive version of collective operations.)
1321 Each collective operation can be manually selected with a
1322 ``smpi/collective_name:algo_name``. Available algorithms are listed in
1323 :ref:`SMPI_use_colls`.
1325 .. TODO:: All available collective algorithms will be made available
1326 via the ``smpirun --help-coll`` command.
1328 .. _cfg=smpi/finalization-barrier:
1330 Add a barrier in MPI_Finalize
1331 .............................
1333 **Option** ``smpi/finalization-barrier`` **default:** off
1335 By default, SMPI processes are destroyed as soon as soon as their code ends,
1336 so after a successful MPI_Finalize call returns. In some rare cases, some data
1337 might have been attached to MPI objects still active in the remaining processes,
1338 and can be destroyed eagerly by the finished process.
1339 If your code shows issues at finalization, such as segmentation fault, triggering
1340 this option will add an explicit MPI_Barrier(MPI_COMM_WORLD) call inside the
1341 MPI_Finalize, so that all processes will terminate at almost the same point.
1342 It might affect the total timing by the cost of a barrier.
1344 .. _cfg=smpi/errors-are-fatal:
1346 Disable MPI fatal errors
1347 ........................
1349 **Option** ``smpi/errors-are-fatal`` **default:** on
1351 By default, SMPI processes will crash if a MPI error code is returned. MPI allows
1352 to explicitely set MPI_ERRORS_RETURN errhandler to avoid this behaviour. This flag
1353 will turn on this behaviour by default (for all concerned types and errhandlers).
1354 This can ease debugging by going after the first reported error.
1356 .. _cfg=smpi/pedantic:
1358 Disable pedantic MPI errors
1359 ...........................
1361 **Option** ``smpi/pedantic`` **default:** on
1363 By default, SMPI will report all errors it finds in MPI codes. Some of these errors
1364 may not be considered as errors by all developers. This flag can be turned off to
1365 avoid reporting some usually harmless mistakes.
1366 Concerned errors list (will be expanded in the future):
1368 - Calling MPI_Win_fence only once in a program, hence just opening an epoch without
1371 .. _cfg=smpi/iprobe:
1373 Inject constant times for MPI_Iprobe
1374 ....................................
1376 **Option** ``smpi/iprobe`` **default:** 0.0001
1378 The behavior and motivation for this configuration option is identical
1379 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1382 .. _cfg=smpi/iprobe-cpu-usage:
1384 Reduce speed for iprobe calls
1385 .............................
1387 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1389 MPI_Iprobe calls can be heavily used in applications. To account
1390 correctly for the energy that cores spend probing, it is necessary to
1391 reduce the load that these calls cause inside SimGrid.
1393 For instance, we measured a maximum power consumption of 220 W for a
1394 particular application but only 180 W while this application was
1395 probing. Hence, the correct factor that should be passed to this
1396 option would be 180/220 = 0.81.
1400 Inject constant times for MPI_Init
1401 ..................................
1403 **Option** ``smpi/init`` **default:** 0
1405 The behavior and motivation for this configuration option is identical
1406 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1410 Inject constant times for MPI_Isend()
1411 .....................................
1413 **Option** ``smpi/ois``
1415 The behavior and motivation for this configuration option is identical
1416 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1420 Inject constant times for MPI_send()
1421 ....................................
1423 **Option** ``smpi/os``
1425 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1426 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1427 time). SMPI can factor these costs in as well, but the user has to
1428 configure SMPI accordingly as these values may vary by machine. This
1429 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1430 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1431 exactly as ``smpi/ois``.
1433 This item can consist of multiple sections; each section takes three
1434 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1435 so this example contains two sections. Furthermore, each section
1436 consists of three values.
1438 1. The first value denotes the minimum size in bytes for this section to take effect;
1439 read it as "if message size is greater than this value (and other section has a larger
1440 first value that is also smaller than the message size), use this".
1441 In the first section above, this value is "1".
1443 2. The second value is the startup time; this is a constant value that will always
1444 be charged, no matter what the size of the message. In the first section above,
1447 3. The third value is the `per-byte` cost. That is, it is charged for every
1448 byte of the message (incurring cost messageSize*cost_per_byte)
1449 and hence accounts also for larger messages. In the first
1450 section of the example above, this value is "2".
1452 Now, SMPI always checks which section it should use for a given
1453 message; that is, if a message of size 11 is sent with the
1454 configuration of the example above, only the second section will be
1455 used, not the first, as the first value of the second section is
1456 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1457 message of size 11 incurs the following cost inside MPI_Send:
1458 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1460 Note that the order of sections can be arbitrary; they will be ordered internally.
1464 Inject constant times for MPI_Recv()
1465 ....................................
1467 **Option** ``smpi/or``
1469 The behavior and motivation for this configuration option is identical
1470 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1473 .. _cfg=smpi/grow-injected-times:
1475 Inject constant times for MPI_Test
1476 ..................................
1478 **Option** ``smpi/test`` **default:** 0.0001
1480 By setting this option, you can control the amount of time a process
1481 sleeps when MPI_Test() is called; this is important, because SimGrid
1482 normally only advances the time while communication is happening and
1483 thus, MPI_Test will not add to the time, resulting in deadlock if it is
1484 used as a break-condition as in the following example:
1489 MPI_Test(request, flag, status);
1493 To speed up execution, we use a counter to keep track of how often we
1494 checked if the handle is now valid or not. Hence, we actually
1495 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1496 process to sleep increases linearly with the number of previously
1497 failed tests. This behavior can be disabled by setting
1498 ``smpi/grow-injected-times`` to **no**. This will also disable this
1499 behavior for MPI_Iprobe.
1501 .. _cfg=smpi/shared-malloc:
1502 .. _cfg=smpi/shared-malloc-hugepage:
1507 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1509 If your simulation consumes too much memory, you may want to modify
1510 your code so that the working areas are shared by all MPI ranks. For
1511 example, in a block-cyclic matrix multiplication, you will only
1512 allocate one set of blocks, and all processes will share them.
1513 Naturally, this will lead to very wrong results, but this will save a
1514 lot of memory. So this is still desirable for some studies. For more on
1515 the motivation for that feature, please refer to the `relevant section
1516 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1517 of the SMPI CourseWare (see Activity #2.2 of the pointed
1518 assignment). In practice, change the calls for malloc() and free() into
1519 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1521 SMPI provides two algorithms for this feature. The first one, called
1522 ``local``, allocates one block per call to SMPI_SHARED_MALLOC()
1523 (each call site gets its own block) ,and this block is shared
1524 among all MPI ranks. This is implemented with the shm_* functions
1525 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1526 for each shared block.
1528 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1529 returns a new address, but it only points to a shadow block: its memory
1530 area is mapped on a 1 MiB file on disk. If the returned block is of size
1531 N MiB, then the same file is mapped N times to cover the whole block.
1532 At the end, no matter how many times you call SMPI_SHARED_MALLOC, this will
1533 only consume 1 MiB in memory.
1535 You can disable this behavior and come back to regular mallocs (for
1536 example for debugging purposes) using ``no`` as a value.
1538 If you want to keep private some parts of the buffer, for instance if these
1539 parts are used by the application logic and should not be corrupted, you
1540 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). For example:
1544 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1546 This will allocate 500 bytes to mem, such that mem[27..41] and
1547 mem[100..199] are shared while other area remain private.
1549 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1551 When smpi/shared-malloc:global is used, the memory consumption problem
1552 is solved, but it may induce too much load on the kernel's pages table.
1553 In this case, you should use huge pages so that the kernel creates only one
1554 entry per MB of malloced data instead of one entry per 4 kB.
1555 To activate this, you must mount a hugetlbfs on your system and allocate
1556 at least one huge page:
1558 .. code-block:: console
1561 $ sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1562 $ sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1564 Then, you can pass the option
1565 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1566 actually activate the huge page support in shared mallocs.
1568 .. _cfg=smpi/auto-shared-malloc-thresh:
1570 Automatically share allocations
1571 ...............................
1573 **Option** ``smpi/auto-shared-malloc-thresh:`` **Default:** 0 (false)
1574 This value in bytes represents the size above which all allocations
1575 will be "shared" by default (as if they were performed through
1576 SMPI_SHARED_MALLOC macros). Default = 0 = disabled feature.
1577 The value must be carefully chosen to only select data buffers which
1578 will not modify execution path or cause crash if their content is false.
1579 Option :ref:`cfg=smpi/display-allocs` can be used to locate the largest
1580 allocation detected in a run, and provide a good starting threshold.
1581 Note : malloc, calloc and free are overridden by smpicc/cxx by default.
1582 This can cause some troubles if codes are already overriding these. If this
1583 is the case, defining SMPI_NO_OVERRIDE_MALLOC in the compilation flags can
1584 help, but will make this feature unusable.
1588 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1589 ...................................................................
1591 **Option** ``smpi/wtime`` **default:** 10 ns
1593 This option controls the amount of (simulated) time spent in calls to
1594 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1595 to 0, the simulated clock is not advanced in these calls, which leads
1596 to issues if your application contains such a loop:
1600 while(MPI_Wtime() < some_time_bound) {
1601 /* some tests, with no communication nor computation */
1604 When the option smpi/wtime is set to 0, the time advances only on
1605 communications and computations. So the previous code results in an
1606 infinite loop: the current [simulated] time will never reach
1607 ``some_time_bound``. This infinite loop is avoided when that option
1608 is set to a small value, as it is by default since SimGrid v3.21.
1610 Note that if your application does not contain any loop depending on
1611 the current time only, then setting this option to a non-zero value
1612 will slow down your simulations by a tiny bit: the simulation loop has
1613 to be broken out of and reset each time your code asks for the current time.
1614 If the simulation speed really matters to you, you can avoid this
1615 extra delay by setting smpi/wtime to 0.
1617 .. _cfg=smpi/list-leaks:
1619 Report leaked MPI objects
1620 .........................
1622 **Option** ``smpi/list-leaks`` **default:** 0
1624 This option controls whether to report leaked MPI objects.
1625 The parameter is the number of leaks to report.
1627 Other Configurations
1628 --------------------
1630 .. _cfg=debug/clean-atexit:
1632 Cleanup at Termination
1633 ......................
1635 **Option** ``debug/clean-atexit`` **default:** on
1637 If your code is segfaulting during its finalization, it may help to
1638 disable this option to request that SimGrid not attempt any cleanups at
1639 the end of the simulation. Since the Unix process is ending anyway,
1640 the operating system will wipe it all.
1647 **Option** ``path`` **default:** . (current dir)
1649 It is possible to specify a list of directories to search in for the
1650 trace files (see :ref:`pf_trace`) by using this configuration
1651 item. To add several directory to the path, set the configuration
1652 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1654 .. _cfg=debug/breakpoint:
1659 **Option** ``debug/breakpoint`` **default:** unset
1661 This configuration option sets a breakpoint: when the simulated clock
1662 reaches the given time, a SIGTRAP is raised. This can be used to stop
1663 the execution and get a backtrace with a debugger.
1665 It is also possible to set the breakpoint from inside the debugger, by
1666 writing in global variable simgrid::simix::breakpoint. For example,
1669 .. code-block:: none
1671 set variable simgrid::simix::breakpoint = 3.1416
1673 .. _cfg=debug/verbose-exit:
1678 **Option** ``debug/verbose-exit`` **default:** on
1680 By default, when Ctrl-C is pressed, the status of all existing actors
1681 is displayed before exiting the simulation. This is very useful to
1682 debug your code, but it can become troublesome if you have many
1683 actors. Set this configuration item to **off** to disable this
1686 .. _cfg=exception/cutpath:
1688 Truncate local path from exception backtrace
1689 ............................................
1691 **Option** ``exception/cutpath`` **default:** off
1693 This configuration option is used to remove the path from the
1694 backtrace shown when an exception is thrown. This is mainly useful for
1695 the tests: the full file path would makes the tests non-reproducible because
1696 the paths of source files depend of the build settings. That would
1697 break most of the tests since their output is continually compared.
1701 Logging configuration
1702 ---------------------
1704 As introduced in :ref:`outcome_logs`, the SimGrid logging mechanism allows to configure at runtime the messages that should be displayed and those that should be omitted. Each
1705 message produced in the code is given a category (denoting its topic) and a priority. Then at runtime, each category is given a threshold (only messages of priority higher than
1706 that threshold are displayed), a layout (deciding how the messages in this category are formatted), and an appender (deciding what to do with the message: either print on stderr or
1709 This section explains how to configure this logging features. You can also refer to the documentation of the :ref:`programmer's interface <logging_prog>`, that allows to produce
1710 messages from your code.
1712 Most of the time, the logging mechanism is configured at runtime using the ``--log`` command-line argument, even if you can also use :c:func:`xbt_log_control_set()` to control it from
1713 your program. To pass configure more than one setting, you can either pass several ``--log`` arguments, or separate your settings with spaces, that must be quoted accordingly. In
1714 practice, the following is equivalent to the above settings: ``--log=root.thresh:error --log=s4u_host.thresh:debug``.
1716 If you want to specify more than one setting, you can either pass several ``--log`` argument to your program as above, or separate them with spaces. In this case, you want to quote
1717 your settings, as in ``--log="root.thresh:error s4u_host.thresh:debug"``. The parameters are interpreted in order, from left to right.
1720 Threshold configuration
1721 .......................
1723 The keyword ``threshold`` controls which logging event will get displayed in a given category. For example, ``--log=root.threshold:debug`` displays *every* message produced in the
1724 ``root`` category and its subcategories (i.e., every message produced -- this is *extremely* verbose), while ``--log=root.thres:critical`` turns almost everything off. As you can
1725 see, ``threshold`` can be abbreviated here.
1727 Existing thresholds:
1729 - ``trace`` some functions display a message at this level when entering or returning
1730 - ``debug`` output that is mostly useful when debugging the corresponding module.
1731 - ``verbose`` verbose output that is only mildly interesting and can easily be ignored
1732 - ``info`` usual output (this is the default threshold of all categories)
1733 - ``warning`` minor issue encountered
1734 - ``error`` issue encountered
1735 - ``critical`` major issue encountered, such as assertions failures
1739 Format configuration
1740 ....................
1742 The keyword ``fmt`` controls the layout (the format) of a logging category. For example, ``--log=root.fmt:%m`` reduces the output to the user-message only, removing any decoration such
1743 as the date, or the actor ID, everything. Existing format directives:
1746 - %n: line separator (LOG4J compatible)
1747 - %e: plain old space (SimGrid extension)
1749 - %m: user-provided message
1751 - %c: Category name (LOG4J compatible)
1752 - %p: Priority name (LOG4J compatible)
1754 - %h: Hostname (SimGrid extension)
1755 - %a: Actor name (SimGrid extension -- note that with SMPI this is the integer value of the process rank)
1756 - %i: Actor PID (SimGrid extension -- this is a 'i' as in 'i'dea)
1757 - %t: Thread "name" (LOG4J compatible -- actually the address of the thread in memory)
1759 - %F: file name where the log event was raised (LOG4J compatible)
1760 - %l: location where the log event was raised (LOG4J compatible, like '%%F:%%L' -- this is a l as in 'l'etter)
1761 - %L: line number where the log event was raised (LOG4J compatible)
1762 - %M: function name (LOG4J compatible -- called method name here of course).
1764 - %d: date (UNIX-like epoch)
1765 - %r: application age (time elapsed since the beginning of the application)
1768 ``--log=root.fmt:'[%h:%a:(%i) %r] %l: %m%n'`` gives you the default layout used for info messages while ``--log=root.fmt:'[%h:%a:(%i) %r] %l: [%c/%p] %m%n'`` gives you the default
1769 layout for the other priorities (it adds the source code location). Also, the actor identification is omitted by the default layout for the messages coming directly from the
1770 SimGrid kernel, so info messages are formatted with ``[%r] [%c/%p] %m%n`` in this case. When specifying the layout manually, such distinctions are currently impossible, and the
1771 provided layout is used for every messages.
1773 As with printf, you can specify the precision and width of the fields. For example, ``%.4r`` limits the date precision to four digits while ``%15h`` limits the host name to at most
1777 If you want to have spaces in your log format, you should protect it. Otherwise, SimGrid will consider that this is a space-separated list of several parameters. But you should
1778 also protect it from the shell that also splits command line arguments on spaces. At the end, you should use something such as ``--log="'root.fmt:%l: [%p/%c]: %m%n'"``.
1779 Another option is to use the ``%e`` directive for spaces, as in ``--log=root.fmt:%l:%e[%p/%c]:%e%m%n``.
1784 The keyword ``app`` controls the appended of a logging category. For example ``--log=root.app:file:mylogfile`` redirects every output to the file ``mylogfile``.
1786 With the ``splitfile`` appender, a new file is created when the size of the output reaches the specified size. The format is ``--log=root.app:splitfile:<size>:<file name>``. For
1787 example, ``--log=root.app:splitfile:500:mylog_%`` creates log files of at most 500 bytes, using the names ``mylog_0``, ``mylog_1``, ``mylog_2``, etc.
1789 The ``rollfile`` appender uses one file only, but the file is emptied and recreated when its size reaches the specified maximum. For example, ``--log=root.app:rollfile:500:mylog``
1790 ensures that the log file ``mylog`` will never overpass 500 bytes in size.
1792 Any appender setup this way have its own layout format, that you may change afterward. When specifying a new appender, its additivity is set to false to prevent log event displayed
1793 by this appender to "leak" to any other appender higher in the hierarchy. You can naturally change that if you want your messages to be displayed twice.
1798 The keyword ``add`` controls the additivity of a logging category. By default, the messages are only passed one appender only: the more specific, i.e. the first one found when
1799 climbing the tree from the category in which they were produced. In Log4J parlance, it is said that the default additivity of appenders is false. If you change this setting to
1800 ``on`` (or ``yes`` or ``1``), the produced messages will also be passed to the upper appender.
1802 Let's consider a more complex example: ``--log="root.app:file:all.log s4u.app:file:iface.log xbt.app:file:xbt.log xbt.add:yes``. Here, the logging of s4u will be sent to the
1803 ``iface.log`` file; the logging of the xbt toolbox will be sent to both the ``xbt.log`` file and the ``all.log`` file (because xbt additivity was enabled); and every other loggings
1804 will only be sent to ``all.log``.
1809 ``--help-logs`` displays a complete help message about logging in SimGrid.
1811 ``--help-log-categories`` displays the actual hierarchy of log categories for this binary.
1813 ``--log=no_loc`` hides the source locations (file names and line numbers) from the messages. This is useful to make tests reproducible.