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:
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/finalization-barrier:** :ref:`cfg=smpi/finalization-barrier`
154 - **smpi/grow-injected-times:** :ref:`cfg=smpi/grow-injected-times`
155 - **smpi/host-speed:** :ref:`cfg=smpi/host-speed`
156 - **smpi/IB-penalty-factors:** :ref:`cfg=smpi/IB-penalty-factors`
157 - **smpi/iprobe:** :ref:`cfg=smpi/iprobe`
158 - **smpi/iprobe-cpu-usage:** :ref:`cfg=smpi/iprobe-cpu-usage`
159 - **smpi/init:** :ref:`cfg=smpi/init`
160 - **smpi/keep-temps:** :ref:`cfg=smpi/keep-temps`
161 - **smpi/lat-factor:** :ref:`cfg=smpi/lat-factor`
162 - **smpi/ois:** :ref:`cfg=smpi/ois`
163 - **smpi/or:** :ref:`cfg=smpi/or`
164 - **smpi/os:** :ref:`cfg=smpi/os`
165 - **smpi/papi-events:** :ref:`cfg=smpi/papi-events`
166 - **smpi/privatization:** :ref:`cfg=smpi/privatization`
167 - **smpi/privatize-libs:** :ref:`cfg=smpi/privatize-libs`
168 - **smpi/send-is-detached-thresh:** :ref:`cfg=smpi/send-is-detached-thresh`
169 - **smpi/shared-malloc:** :ref:`cfg=smpi/shared-malloc`
170 - **smpi/shared-malloc-hugepage:** :ref:`cfg=smpi/shared-malloc-hugepage`
171 - **smpi/simulate-computation:** :ref:`cfg=smpi/simulate-computation`
172 - **smpi/test:** :ref:`cfg=smpi/test`
173 - **smpi/wtime:** :ref:`cfg=smpi/wtime`
174 - **smpi/list-leaks** :ref:`cfg=smpi/list-leaks`
176 - **Tracing configuration options** can be found in Section :ref:`tracing_tracing_options`
178 - **storage/model:** :ref:`options_model_select`
180 - **vm/model:** :ref:`options_model_select`
184 Configuring the Platform Models
185 -------------------------------
187 .. _options_model_select:
189 Choosing the Platform Models
190 ............................
192 SimGrid comes with several network, CPU and disk models built in,
193 and you can change the used model at runtime by changing the passed
194 configuration. The three main configuration items are given below.
195 For each of these items, passing the special ``help`` value gives you
196 a short description of all possible values (for example,
197 ``--cfg=network/model:help`` will present all provided network
198 models). Also, ``--help-models`` should provide information about all
199 models for all existing resources.
201 - ``network/model``: specify the used network model. Possible values:
203 - **LV08 (default one):** Realistic network analytic model
204 (slow-start modeled by multiplying latency by 13.01, bandwidth by
205 .97; bottleneck sharing uses a payload of S=20537 for evaluating
206 RTT). Described in `Accuracy Study and Improvement of Network
207 Simulation in the SimGrid Framework
208 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
209 - **Constant:** Simplistic network model where all communication
210 take a constant time (one second). This model provides the lowest
211 realism, but is (marginally) faster.
212 - **SMPI:** Realistic network model specifically tailored for HPC
213 settings (accurate modeling of slow start with correction factors on
214 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). This model can be
215 :ref:`further configured <options_model_network>`.
216 - **IB:** Realistic network model specifically tailored for HPC
217 settings with InfiniBand networks (accurate modeling contention
218 behavior, based on the model explained in `this PhD work
219 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
220 This model can be :ref:`further configured <options_model_network>`.
221 - **CM02:** Legacy network analytic model. Very similar to LV08, but
222 without corrective factors. The timings of small messages are thus
223 poorly modeled. This model is described in `A Network Model for
224 Simulation of Grid Application
225 <https://hal.inria.fr/inria-00071989/document>`_.
226 - **ns-3** (only available if you compiled SimGrid accordingly):
227 Use the packet-level network
228 simulators as network models (see :ref:`model_ns3`).
229 This model can be :ref:`further configured <options_pls>`.
231 - ``cpu/model``: specify the used CPU model. We have only one model
234 - **Cas01:** Simplistic CPU model (time=size/speed)
236 - ``host/model``: The host concept is the aggregation of a CPU with a
237 network card. Three models exists, but actually, only 2 of them are
238 interesting. The "compound" one is simply due to the way our
239 internal code is organized, and can easily be ignored. So at the
240 end, you have two host models: The default one allows aggregation of
241 an existing CPU model with an existing network model, but does not
242 allow parallel tasks because these beasts need some collaboration
243 between the network and CPU model. That is why, ptask_07 is used by
244 default when using SimDag.
246 - **default:** Default host model. Currently, CPU:Cas01 and
247 network:LV08 (with cross traffic enabled)
248 - **compound:** Host model that is automatically chosen if
249 you change the network and CPU models
250 - **ptask_L07:** Host model somehow similar to Cas01+CM02 but
251 allowing "parallel tasks", that are intended to model the moldable
252 tasks of the grid scheduling literature.
254 - ``storage/model``: specify the used storage model. Only one model is
256 - ``vm/model``: specify the model for virtual machines. Only one model
259 .. todo: make 'compound' the default host model.
261 .. _options_model_optim:
266 The network and CPU models that are based on lmm_solve (that
267 is, all our analytical models) accept specific optimization
270 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
272 - **Lazy:** Lazy action management (partial invalidation in lmm +
273 heap in action remaining).
274 - **TI:** Trace integration. Highly optimized mode when using
275 availability traces (only available for the Cas01 CPU model for
277 - **Full:** Full update of remaining and variables. Slow but may be
278 useful when debugging.
280 - items ``network/maxmin-selective-update`` and
281 ``cpu/maxmin-selective-update``: configure whether the underlying
282 should be lazily updated or not. It should have no impact on the
283 computed timings, but should speed up the computation. |br| It is
284 still possible to disable this feature because it can reveal
285 counter-productive in very specific scenarios where the
286 interaction level is high. In particular, if all your
287 communication share a given backbone link, you should disable it:
288 without it, a simple regular loop is used to update each
289 communication. With it, each of them is still updated (because of
290 the dependency induced by the backbone), but through a complicated
291 and slow pattern that follows the actual dependencies.
293 .. _cfg=maxmin/precision:
294 .. _cfg=surf/precision:
299 **Option** ``maxmin/precision`` **Default:** 0.00001 (in flops or bytes) |br|
300 **Option** ``surf/precision`` **Default:** 0.00001 (in seconds)
302 The analytical models handle a lot of floating point values. It is
303 possible to change the epsilon used to update and compare them through
304 this configuration item. Changing it may speedup the simulation by
305 discarding very small actions, at the price of a reduced numerical
306 precision. You can modify separately the precision used to manipulate
307 timings (in seconds) and the one used to manipulate amounts of work
310 .. _cfg=maxmin/concurrency-limit:
315 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
317 The maximum number of variables per resource can be tuned through this
318 option. You can have as many simultaneous actions per resources as you
319 want. If your simulation presents a very high level of concurrency, it
320 may help to use e.g. 100 as a value here. It means that at most 100
321 actions can consume a resource at a given time. The extraneous actions
322 are queued and wait until the amount of concurrency of the considered
323 resource lowers under the given boundary.
325 Such limitations help both to the simulation speed and simulation accuracy
326 on highly constrained scenarios, but the simulation speed suffers of this
327 setting on regular (less constrained) scenarios so it is off by default.
329 .. _options_model_network:
331 Configuring the Network Model
332 .............................
334 .. _cfg=network/TCP-gamma:
336 Maximal TCP Window Size
337 ^^^^^^^^^^^^^^^^^^^^^^^
339 **Option** ``network/TCP-gamma`` **Default:** 4194304
341 The analytical models need to know the maximal TCP window size to take
342 the TCP congestion mechanism into account. On Linux, this value can
343 be retrieved using the following commands. Both give a set of values,
344 and you should use the last one, which is the maximal size.
346 .. code-block:: shell
348 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
349 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
351 .. _cfg=network/bandwidth-factor:
352 .. _cfg=network/latency-factor:
353 .. _cfg=network/weight-S:
355 Correcting Important Network Parameters
356 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
358 SimGrid can take network irregularities such as a slow startup or
359 changing behavior depending on the message size into account. You
360 should not change these values unless you really know what you're
361 doing. The corresponding values were computed through data fitting
362 one the timings of packet-level simulators, as described in `Accuracy
363 Study and Improvement of Network Simulation in the SimGrid Framework
364 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
367 If you are using the SMPI model, these correction coefficients are
368 themselves corrected by constant values depending on the size of the
369 exchange. By default SMPI uses factors computed on the Stampede
370 Supercomputer at TACC, with optimal deployment of processes on
371 nodes. Again, only hardcore experts should bother about this fact.
374 .. todo:: This section should be rewritten, and actually explain the
375 options network/bandwidth-factor, network/latency-factor,
378 .. _cfg=smpi/IB-penalty-factors:
383 InfiniBand network behavior can be modeled through 3 parameters
384 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `this PhD
386 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_ (in French)
387 or more concisely in `this paper <https://hal.inria.fr/hal-00953618/document>`_,
388 even if that paper does only describe models for myrinet and ethernet.
389 You can see in Fig 2 some results for Infiniband, for example. This model
390 may be outdated by now for modern infiniband, anyway, so a new
391 validation would be good.
393 The three paramaters are defined as follows:
395 - βs: penalty factor for outgoing messages, computed by running a simple send to
396 two nodes and checking slowdown compared to a single send to one node,
398 - βe: penalty factor for ingoing messages, same computation method but with one
399 node receiving several messages
400 - γr: slowdown factor when communication buffer memory is saturated. It needs a
401 more complicated pattern to run in order to be computed (5.3 in the thesis,
402 page 107), and formula in the end is γr = time(c)/(3×βe×time(ref)), where
403 time(ref) is the time of a single comm with no contention).
405 Once these values are computed, a penalty is assessed for each message (this is
406 the part implemented in the simulator) as shown page 106 of the thesis. Here is
407 a simple translation of this text. First, some notations:
409 - ∆e(e) which corresponds to the incoming degree of node e, that is to say the number of communications having as destination node e.
410 - ∆s (s) which corresponds to the degree outgoing from node s, that is to say the number of communications sent by node s.
411 - Φ (e) which corresponds to the number of communications destined for the node e but coming from a different node.
412 - 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
414 To determine the penalty for a communication, two values need to be calculated. First, the penalty caused by the conflict in transmission, noted ps.
417 - if ∆s (i) = 1 then ps = 1.
418 - if ∆s (i) ≥ 2 and ∆e (i) ≥ 3 then ps = ∆s (i) × βs × γr
419 - else, ps = ∆s (i) × βs
422 Then, the penalty caused by the conflict in reception (noted pe) should be computed as follows:
424 - if ∆e (i) = 1 then pe = 1
425 - else, pe = Φ (e) × βe × Ω (s, e)
427 Finally, the penalty associated with the communication is:
428 p = max (ps ∈ s, pe)
430 .. _cfg=network/crosstraffic:
432 Simulating Cross-Traffic
433 ^^^^^^^^^^^^^^^^^^^^^^^^
435 Since SimGrid v3.7, cross-traffic effects can be taken into account in
436 analytical simulations. It means that ongoing and incoming
437 communication flows are treated independently. In addition, the LV08
438 model adds 0.05 of usage on the opposite direction for each new
439 created flow. This can be useful to simulate some important TCP
440 phenomena such as ack compression.
442 For that to work, your platform must have two links for each
443 pair of interconnected hosts. An example of usable platform is
444 available in ``examples/platforms/crosstraffic.xml``.
446 This is activated through the ``network/crosstraffic`` item, that
447 can be set to 0 (disable this feature) or 1 (enable it).
449 Note that with the default host model this option is activated by default.
451 .. _cfg=network/loopback:
453 Configuring loopback link
454 ^^^^^^^^^^^^^^^^^^^^^^^^^
456 Several network model provide an implicit loopback link to account for local
457 communication on a host. By default it has a 10GBps bandwidth and a null latency.
458 This can be changed with ``network/loopback-lat`` and ``network/loopback-bw``
461 .. _cfg=smpi/async-small-thresh:
463 Simulating Asynchronous Send
464 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
466 (this configuration item is experimental and may change or disappear)
468 It is possible to specify that messages below a certain size (in bytes) will be
469 sent as soon as the call to MPI_Send is issued, without waiting for
470 the correspondent receive. This threshold can be configured through
471 the ``smpi/async-small-thresh`` item. The default value is 0. This
472 behavior can also be manually set for mailboxes, by setting the
473 receiving mode of the mailbox with a call to
474 :cpp:func:`MSG_mailbox_set_async`. After this, all messages sent to
475 this mailbox will have this behavior regardless of the message size.
477 This value needs to be smaller than or equals to the threshold set at
478 :ref:`cfg=smpi/send-is-detached-thresh`, because asynchronous messages
479 are meant to be detached as well.
486 **Option** ``ns3/TcpModel`` **Default:** "default" (ns-3 default)
488 When using ns-3, there is an extra item ``ns3/TcpModel``, corresponding
489 to the ``ns3::TcpL4Protocol::SocketType`` configuration item in
490 ns-3. The only valid values (enforced on the SimGrid side) are
491 'default' (no change to the ns-3 configuration), 'NewReno' or 'Reno' or
494 **Option** ``ns3/seed`` **Default:** "" (don't set the seed in ns-3)
496 This option is the random seed to provide to ns-3 with
497 ``ns3::RngSeedManager::SetSeed`` and ``ns3::RngSeedManager::SetRun``.
499 If left blank, no seed is set in ns-3. If the value 'time' is
500 provided, the current amount of seconds since epoch is used as a seed.
501 Otherwise, the provided value must be a number to use as a seed.
503 Configuring the Storage model
504 .............................
506 .. _cfg=storage/max_file_descriptors:
508 File Descriptor Count per Host
509 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
511 **Option** ``storage/max_file_descriptors`` **Default:** 1024
513 Each host maintains a fixed-size array of its file descriptors. You
514 can change its size through this item to either enlarge it if your
515 application requires it or to reduce it to save memory space.
522 SimGrid plugins allow one to extend the framework without changing its
523 source code directly. Read the source code of the existing plugins to
524 learn how to do so (in ``src/plugins``), and ask your questions to the
525 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
526 that plugins usually register callbacks to some signals of interest.
527 If they need to store some information about a given object (Link, CPU
528 or Actor), they do so through the use of a dedicated object extension.
530 Some of the existing plugins can be activated from the command line,
531 meaning that you can activate them from the command line without any
532 modification to your simulation code. For example, you can activate
533 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
536 Here is a partial list of plugins that can be activated this way. You can get
537 the full list by passing ``--cfg=plugin:help`` to your simulator.
539 - :ref:`Host Energy <plugin_host_energy>`: models the energy dissipation of the compute units.
540 - :ref:`Link Energy <plugin_link_energy>`: models the energy dissipation of the network.
541 - :ref:`Host Load <plugin_host_load>`: monitors the load of the compute units.
543 .. _options_modelchecking:
545 Configuring the Model-Checking
546 ------------------------------
548 To enable SimGrid's model-checking support, the program should
549 be executed using the simgrid-mc wrapper:
551 .. code-block:: shell
553 simgrid-mc ./my_program
555 Safety properties are expressed as assertions using the function
556 :cpp:func:`void MC_assert(int prop)`.
558 .. _cfg=smpi/buffering:
560 Specifying the MPI buffering behavior
561 .....................................
563 **Option** ``smpi/buffering`` **Default:** infty
565 Buffering in MPI has a huge impact on the communication semantic. For example,
566 standard blocking sends are synchronous calls when the system buffers are full
567 while these calls can complete immediately without even requiring a matching
568 receive call for small messages sent when the system buffers are empty.
570 In SMPI, this depends on the message size, that is compared against two thresholds:
572 - if (size < :ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>`) then
573 MPI_Send returns immediately, even if the corresponding receive has not be issued yet.
574 - 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
575 MPI_Send returns as soon as the corresponding receive has been issued. This is known as the eager mode.
576 - if (:ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>` < size) then
577 MPI_Send returns only when the message has actually been sent over the network. This is known as the rendez-vous mode.
579 The ``smpi/buffering`` (only valid with MC) option gives an easier interface to choose between these semantics. It can take two values:
581 - **zero:** means that buffering should be disabled. All communications are actually blocking.
582 - **infty:** means that buffering should be made infinite. All communications are non-blocking.
584 .. _cfg=model-check/property:
586 Specifying a liveness property
587 ..............................
589 **Option** ``model-check/property`` **Default:** unset
591 If you want to specify liveness properties, you have to pass them on
592 the command line, specifying the name of the file containing the
593 property, as formatted by the `ltl2ba <https://github.com/utwente-fmt/ltl2ba>`_ program.
594 Note that ltl2ba is not part of SimGrid and must be installed separately.
596 .. code-block:: shell
598 simgrid-mc ./my_program --cfg=model-check/property:<filename>
600 .. _cfg=model-check/checkpoint:
602 Going for Stateful Verification
603 ...............................
605 By default, the system is backtracked to its initial state to explore
606 another path, instead of backtracking to the exact step before the fork
607 that we want to explore (this is called stateless verification). This
608 is done this way because saving intermediate states can rapidly
609 exhaust the available memory. If you want, you can change the value of
610 the ``model-check/checkpoint`` item. For example,
611 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
612 step. Beware, this will certainly explode your memory. Larger values
613 are probably better, make sure to experiment a bit to find the right
614 setting for your specific system.
616 .. _cfg=model-check/reduction:
618 Specifying the kind of reduction
619 ................................
621 The main issue when using the model-checking is the state space
622 explosion. You can activate some reduction technique with
623 ``--cfg=model-check/reduction:<technique>``. For now, this
624 configuration variable can take 2 values:
626 - **none:** Do not apply any kind of reduction (mandatory for
627 liveness properties, as our current DPOR algorithm breaks cycles)
628 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if
629 you verify local safety properties (default value for safety
632 Another way to mitigate the state space explosion is to search for
633 cycles in the exploration with the :ref:`cfg=model-check/visited`
634 configuration. Note that DPOR and state-equality reduction may not
635 play well together. You should choose between them.
637 Our current DPOR implementation could be improved in may ways. We are
638 currently improving its efficiency (both in term of reduction ability
639 and computational speed), and future work could make it compatible
640 with liveness properties.
642 .. _cfg=model-check/visited:
644 Size of Cycle Detection Set (state equality reduction)
645 ......................................................
647 Mc SimGrid can be asked to search for cycles during the exploration,
648 i.e. situations where a new explored state is in fact the same state
649 than a previous one.. This can prove useful to mitigate the state
650 space explosion with safety properties, and this is the crux when
651 searching for counter-examples to the liveness properties.
653 Note that this feature may break the current implementation of the
654 DPOR reduction technique.
656 The ``model-check/visited`` item is the maximum number of states, which
657 are stored in memory. If the maximum number of snapshotted state is
658 reached, some states will be removed from the memory and some cycles
659 might be missed. Small values can lead to incorrect verifications, but
660 large values can exhaust your memory and be CPU intensive as each new
661 state must be compared to that amount of older saved states.
663 The default settings depend on the kind of exploration. With safety
664 checking, no state is snapshotted and cycles cannot be detected. With
665 liveness checking, all states are snapshotted because missing a cycle
666 could hinder the exploration soundness.
668 .. _cfg=model-check/termination:
670 Non-Termination Detection
671 .........................
673 The ``model-check/termination`` configuration item can be used to
674 report if a non-termination execution path has been found. This is a
675 path with a cycle, which means that the program might never terminate.
677 This only works in safety mode, not in liveness mode.
679 This options is disabled by default.
681 .. _cfg=model-check/dot-output:
686 If set, the ``model-check/dot-output`` configuration item is the name
687 of a file in which to write a dot file of the path leading to the
688 property violation discovered (safety or liveness violation), as well
689 as the cycle for liveness properties. This dot file can then be fed to the
690 graphviz dot tool to generate a corresponding graphical representation.
692 .. _cfg=model-check/max-depth:
694 Exploration Depth Limit
695 .......................
697 The ``model-check/max-depth`` can set the maximum depth of the
698 exploration graph of the model checker. If this limit is reached, a
699 logging message is sent and the results might not be exact.
701 By default, the exploration is limited to the depth of 1000.
703 .. _cfg=model-check/timeout:
708 By default, the model checker does not handle timeout conditions: the `wait`
709 operations never time out. With the ``model-check/timeout`` configuration item
710 set to **yes**, the model checker will explore timeouts of `wait` operations.
712 .. _cfg=model-check/communications-determinism:
713 .. _cfg=model-check/send-determinism:
715 Communication Determinism
716 .........................
718 The ``model-check/communications-determinism`` and
719 ``model-check/send-determinism`` items can be used to select the
720 communication determinism mode of the model checker, which checks
721 determinism properties of the communications of an application.
725 Verification Performance Considerations
726 .......................................
728 The size of the stacks can have a huge impact on the memory
729 consumption when using model-checking. By default, each snapshot will
730 save a copy of the whole stacks and not only of the part that is
731 really meaningful: you should expect the contribution of the memory
732 consumption of the snapshots to be:
733 :math:`\text{number of processes} \times \text{stack size} \times \text{number of states}`.
735 When compiled against the model checker, the stacks are not
736 protected with guards: if the stack size is too small for your
737 application, the stack will silently overflow into other parts of the
738 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
740 .. _cfg=model-check/replay:
742 Replaying buggy execution paths from the model checker
743 ......................................................
745 Debugging the problems reported by the model checker is challenging:
746 First, the application under verification cannot be debugged with gdb
747 because the model checker already traces it. Then, the model checker may
748 explore several execution paths before encountering the issue, making it
749 very difficult to understand the output. Fortunately, SimGrid provides
750 the execution path leading to any reported issue so that you can replay
751 this path reported by the model checker, enabling the usage of classical
754 When the model checker finds an interesting path in the application
755 execution graph (where a safety or liveness property is violated), it
756 generates an identifier for this path. Here is an example of the output:
758 .. code-block:: shell
760 [ 0.000000] (0:@) Check a safety property
761 [ 0.000000] (0:@) **************************
762 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
763 [ 0.000000] (0:@) **************************
764 [ 0.000000] (0:@) Counter-example execution trace:
765 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
766 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
767 [ 0.000000] (0:@) Path = 1/3;1/4
768 [ 0.000000] (0:@) Expanded states = 27
769 [ 0.000000] (0:@) Visited states = 68
770 [ 0.000000] (0:@) Executed transitions = 46
772 The interesting line is ``Path = 1/3;1/4``, which means that you should use
773 ``--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
774 execution path. All options (but the model checker related ones) must
775 remain the same. In particular, if you ran your application with
776 ``smpirun -wrapper simgrid-mc``, then do it again. Remove all
777 MC-related options, keep non-MC-related ones and add
778 ``--cfg=model-check/replay:???``.
780 Currently, if the path is of the form ``X;Y;Z``, each number denotes
781 the actor's pid that is selected at each indecision point. If it's of
782 the form ``X/a;Y/b``, the X and Y are the selected pids while the a
783 and b are the return values of their simcalls. In the previous
784 example, ``1/3;1/4``, you can see from the full output that the actor
785 1 is doing MC_RANDOM simcalls, so the 3 and 4 simply denote the values
786 that these simcall return on the execution branch leading to the
789 Configuring the User Code Virtualization
790 ----------------------------------------
792 .. _cfg=contexts/factory:
794 Selecting the Virtualization Factory
795 ....................................
797 **Option** contexts/factory **Default:** "raw"
799 In SimGrid, the user code is virtualized in a specific mechanism that
800 allows the simulation kernel to control its execution: when a user
801 process requires a blocking action (such as sending a message), it is
802 interrupted, and only gets released when the simulated clock reaches
803 the point where the blocking operation is done. This is explained
804 graphically in the `relevant tutorial, available online
805 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
807 In SimGrid, the containers in which user processes are virtualized are
808 called contexts. Several context factory are provided, and you can
809 select the one you want to use with the ``contexts/factory``
810 configuration item. Some of the following may not exist on your
811 machine because of portability issues. In any case, the default one
812 should be the most effcient one (please report bugs if the
813 auto-detection fails for you). They are approximately sorted here from
814 the slowest to the most efficient:
816 - **thread:** very slow factory using full featured threads (either
817 pthreads or windows native threads). They are slow but very
818 standard. Some debuggers or profilers only work with this factory.
819 - **java:** Java applications are virtualized onto java threads (that
820 are regular pthreads registered to the JVM)
821 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
822 - **boost:** This uses the `context
823 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
824 of the boost library for a performance that is comparable to our
826 |br| Install the relevant library (e.g. with the
827 libboost-contexts-dev package on Debian/Ubuntu) and recompile
829 - **raw:** amazingly fast factory using a context switching mechanism
830 of our own, directly implemented in assembly (only available for x86
831 and amd64 platforms for now) and without any unneeded system call.
833 The main reason to change this setting is when the debugging tools become
834 fooled by the optimized context factories. Threads are the most
835 debugging-friendly contexts, as they allow one to set breakpoints
836 anywhere with gdb and visualize backtraces for all processes, in order
837 to debug concurrency issues. Valgrind is also more comfortable with
838 threads, but it should be usable with all factories (Exception: the
839 callgrind tool really dislikes raw and ucontext factories).
841 .. _cfg=contexts/stack-size:
843 Adapting the Stack Size
844 .......................
846 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
848 Each virtualized used process is executed using a specific system
849 stack. The size of this stack has a huge impact on the simulation
850 scalability, but its default value is rather large. This is because
851 the error messages that you get when the stack size is too small are
852 rather disturbing: this leads to stack overflow (overwriting other
853 stacks), leading to segfaults with corrupted stack traces.
855 If you want to push the scalability limits of your code, you might
856 want to reduce the ``contexts/stack-size`` item. Its default value is
857 8192 (in KiB), while our Chord simulation works with stacks as small
858 as 16 KiB, for example. You can ensure that some actors have a specific
859 size by simply changing the value of this configuration item before
860 creating these actors. The :cpp:func:`simgrid::s4u::Engine::set_config`
861 functions are handy for that.
863 This *setting is ignored* when using the thread factory (because there
864 is no way to modify the stack size with C++ system threads). Instead,
865 you should compile SimGrid and your application with
866 ``-fsplit-stack``. Note that this compilation flag is not compatible
867 with the model checker right now.
869 The operating system should only allocate memory for the pages of the
870 stack which are actually used and you might not need to use this in
871 most cases. However, this setting is very important when using the
872 model checker (see :ref:`options_mc_perf`).
874 .. _cfg=contexts/guard-size:
876 Disabling Stack Guard Pages
877 ...........................
879 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages on Windows or with MC)
881 Unless you use the threads context factory (see
882 :ref:`cfg=contexts/factory`), a stack guard page is usually used
883 which prevents the stack of a given actor from overflowing on another
884 stack. But the performance impact may become prohibitive when the
885 amount of actors increases. The option ``contexts/guard-size`` is the
886 number of stack guard pages used. By setting it to 0, no guard pages
887 will be used: in this case, you should avoid using small stacks (with
888 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
889 will silently overflow on other parts of the memory.
891 When no stack guard page is created, stacks may then silently overflow
892 on other parts of the memory if their size is too small for the
895 .. _cfg=contexts/nthreads:
896 .. _cfg=contexts/synchro:
898 Running User Code in Parallel
899 .............................
901 Parallel execution of the user code is only considered stable in
902 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
903 simulations may well fail in parallel mode. It is described in
904 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
906 If you are using the **ucontext** or **raw** context factories, you can
907 request to execute the user code in parallel. Several threads are
908 launched, each of them handling the same number of user contexts at each
909 run. To activate this, set the ``contexts/nthreads`` item to the amount
910 of cores that you have in your computer (or lower than 1 to have the
911 amount of cores auto-detected).
913 When parallel execution is activated, you can choose the
914 synchronization schema used with the ``contexts/synchro`` item,
915 which value is either:
917 - **futex:** ultra optimized synchronisation schema, based on futexes
918 (fast user-mode mutexes), and thus only available on Linux systems.
919 This is the default mode when available.
920 - **posix:** slow but portable synchronisation using only POSIX
922 - **busy_wait:** not really a synchronisation: the worker threads
923 constantly request new contexts to execute. It should be the most
924 efficient synchronisation schema, but it loads all the cores of
925 your machine for no good reason. You probably prefer the other less
928 Configuring the Tracing
929 -----------------------
931 The :ref:`tracing subsystem <outcomes_vizu>` can be configured in
932 several different ways depending on the used interface (S4U, SMPI, SimDag)
933 and the kind of traces that needs to be obtained. See the
934 :ref:`Tracing Configuration Options subsection
935 <tracing_tracing_options>` for a full description of each
936 configuration option.
938 We detail here a simple way to get the traces working for you, even if
939 you never used the tracing API.
942 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
944 .. code-block:: shell
946 --cfg=tracing:yes --cfg=tracing/uncategorized:yes
948 The first parameter activates the tracing subsystem, and the second
949 tells it to trace host and link utilization (without any
952 - MSG or SimDag-based simulator and categorized traces (you need to
953 declare categories and classify your tasks according to them)
955 .. code-block:: shell
957 --cfg=tracing:yes --cfg=tracing/categorized:yes
959 The first parameter activates the tracing subsystem, and the second
960 tells it to trace host and link categorized utilization.
962 - SMPI simulator and traces for a space/time view:
964 .. code-block:: shell
968 The `-trace` parameter for the smpirun script runs the simulation
969 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
970 smpirun's `-help` parameter for additional tracing options.
972 Sometimes you might want to put additional information on the trace to
973 correctly identify them later, or to provide data that can be used to
974 reproduce an experiment. You have two ways to do that:
976 - Add a string on top of the trace file as comment:
978 .. code-block:: shell
980 --cfg=tracing/comment:my_simulation_identifier
982 - Add the contents of a textual file on top of the trace file as comment:
984 .. code-block:: shell
986 --cfg=tracing/comment-file:my_file_with_additional_information.txt
988 Please, use these two parameters (for comments) to make reproducible
989 simulations. For additional details about this and all tracing
990 options, check See the :ref:`tracing_tracing_options`.
995 .. _cfg=msg/debug-multiple-use:
1000 **Option** ``msg/debug-multiple-use`` **Default:** off
1002 Sometimes your application may try to send a task that is still being
1003 executed somewhere else, making it impossible to send this task. However,
1004 for debugging purposes, one may want to know what the other host is/was
1005 doing. This option shows a backtrace of the other process.
1010 The SMPI interface provides several specific configuration items.
1011 These are not easy to see, since the code is usually launched through the
1012 ``smiprun`` script directly.
1014 .. _cfg=smpi/host-speed:
1015 .. _cfg=smpi/cpu-threshold:
1016 .. _cfg=smpi/simulate-computation:
1018 Automatic Benchmarking of SMPI Code
1019 ...................................
1021 In SMPI, the sequential code is automatically benchmarked, and these
1022 computations are automatically reported to the simulator. That is to
1023 say that if you have a large computation between a ``MPI_Recv()`` and
1024 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
1025 this code, and create an execution task within the simulator to take
1026 this into account. For that, the actual duration is measured on the
1027 host machine and then scaled to the power of the corresponding
1028 simulated machine. The variable ``smpi/host-speed`` allows one to
1029 specify the computational speed of the host machine (in flop/s by
1030 default) to use when scaling the execution times.
1032 The default value is ``smpi/host-speed=20kf`` (= 20,000 flop/s). This
1033 is probably underestimated for most machines, leading SimGrid to
1034 overestimate the amount of flops in the execution blocks that are
1035 automatically injected in the simulator. As a result, the execution
1036 time of the whole application will probably be overestimated until you
1037 use a realistic value.
1039 When the code consists of numerous consecutive MPI calls, the
1040 previous mechanism feeds the simulation kernel with numerous tiny
1041 computations. The ``smpi/cpu-threshold`` item becomes handy when this
1042 impacts badly on the simulation performance. It specifies a threshold (in
1043 seconds) below which the execution chunks are not reported to the
1044 simulation kernel (default value: 1e-6).
1046 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
1047 time spent below this threshold. SMPI does not consider the
1048 `amount of time` of these computations; there is no offset for
1049 this. Hence, a value that is too small, may lead to unreliable
1052 In some cases, however, one may wish to disable simulation of
1053 the computation of an application. This is the case when SMPI is used not to
1054 simulate an MPI application, but instead an MPI code that performs
1055 "live replay" of another MPI app (e.g., ScalaTrace's replay tool, or
1056 various on-line simulators that run an app at scale). In this case the
1057 computation of the replay/simulation logic should not be simulated by
1058 SMPI. Instead, the replay tool or on-line simulator will issue
1059 "computation events", which correspond to the actual MPI simulation
1060 being replayed/simulated. At the moment, these computation events can
1061 be simulated using SMPI by calling internal smpi_execute*() functions.
1063 To disable the benchmarking/simulation of a computation in the simulated
1064 application, the variable ``smpi/simulate-computation`` should be set
1065 to **no**. This option just ignores the timings in your simulation; it
1066 still executes the computations itself. If you want to stop SMPI from
1067 doing that, you should check the SMPI_SAMPLE macros, documented in
1068 Section :ref:`SMPI_use_faster`.
1070 +------------------------------------+-------------------------+-----------------------------+
1071 | Solution | Computations executed? | Computations simulated? |
1072 +====================================+=========================+=============================+
1073 | --cfg=smpi/simulate-computation:no | Yes | Never |
1074 +------------------------------------+-------------------------+-----------------------------+
1075 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
1076 +------------------------------------+-------------------------+-----------------------------+
1077 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
1078 +------------------------------------+-------------------------+-----------------------------+
1080 .. _cfg=smpi/comp-adjustment-file:
1082 Slow-down or speed-up parts of your code
1083 ........................................
1085 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
1087 This option allows you to pass a file that contains two columns: The
1088 first column defines the section that will be subject to a speedup;
1089 the second column is the speedup. For instance:
1091 .. code-block:: shell
1093 "start:stop","ratio"
1094 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1096 The first line is the header - you must include it. The following
1097 line means that the code between two consecutive MPI calls on line 30
1098 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1099 of 1.18244559422142. The value for the second column is therefore a
1100 speedup, if it is larger than 1 and a slowdown if it is smaller
1101 than 1. Nothing will be changed if it is equal to 1.
1103 Of course, you can set any arbitrary filenames you want (so the start
1104 and end don't have to be in the same file), but be aware that this
1105 mechanism only supports `consecutive calls!`
1107 Please note that you must pass the ``-trace-call-location`` flag to
1108 smpicc or smpiff, respectively. This flag activates some internal
1109 macro definitions that help with obtaining the call location.
1111 .. _cfg=smpi/bw-factor:
1116 **Option** ``smpi/bw-factor``
1117 |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
1119 The possible throughput of network links is often dependent on the
1120 message sizes, as protocols may adapt to different message sizes. With
1121 this option, a series of message sizes and factors are given, helping
1122 the simulation to be more realistic. For instance, the current default
1123 value means that messages with size 65472 bytes and more will get a total of
1124 MAX_BANDWIDTH*0.940694, messages of size 15424 to 65471 will get
1125 MAX_BANDWIDTH*0.697866, and so on (where MAX_BANDWIDTH denotes the
1126 bandwidth of the link).
1128 An experimental script to compute these factors is available online. See
1129 https://framagit.org/simgrid/platform-calibration/
1130 https://simgrid.org/contrib/smpi-saturation-doc.html
1132 .. _cfg=smpi/display-timing:
1134 Reporting Simulation Time
1135 .........................
1137 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1139 Most of the time, you run MPI code with SMPI to compute the time it
1140 would take to run it on a platform. But since the code is run through
1141 the ``smpirun`` script, you don't have any control on the launcher
1142 code, making it difficult to report the simulated time when the
1143 simulation ends. If you enable the ``smpi/display-timing`` item,
1144 ``smpirun`` will display this information when the simulation
1146 SMPI will also display information about the amout of real time spent
1147 in application code and in SMPI internals, to provide hints about the
1148 need to use sampling to reduce simulation time.
1150 .. _cfg=smpi/display-allocs:
1152 Reporting memory allocations
1153 ............................
1155 **Option** ``smpi/display-allocs`` **Default:** 0 (false)
1157 SMPI intercepts malloc and calloc calls performed inside the running
1158 application, if it wasn't compiled with SMPI_NO_OVERRIDE_MALLOC.
1159 With this option, SMPI will show at the end of execution the amount of
1160 memory allocated through these calls, and locate the most expensive one.
1161 This helps finding the targets for manual memory sharing, or the threshold
1162 to use for smpi/auto-shared-malloc-thresh option (see :ref:`cfg=smpi/auto-shared-malloc-thresh`).
1164 .. _cfg=smpi/keep-temps:
1166 Keeping temporary files after simulation
1167 ........................................
1169 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1171 SMPI usually generates a lot of temporary files that are cleaned after
1172 use. This option requests to preserve them, for example to debug or
1173 profile your code. Indeed, the binary files are removed very early
1174 under the dlopen privatization schema, which tends to fool the
1177 .. _cfg=smpi/lat-factor:
1182 **Option** ``smpi/lat-factor`` |br|
1183 **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
1185 The motivation and syntax for this option is identical to the motivation/syntax
1186 of :ref:`cfg=smpi/bw-factor`.
1188 There is an important difference, though: While smpi/bw-factor `reduces` the
1189 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
1190 increase the latency, i.e., values larger than or equal to 1 are valid here.
1192 .. _cfg=smpi/papi-events:
1194 Trace hardware counters with PAPI
1195 .................................
1197 **Option** ``smpi/papi-events`` **default:** unset
1199 When the PAPI support is compiled into SimGrid, this option takes the
1200 names of PAPI counters and adds their respective values to the trace
1201 files (See Section :ref:`tracing_tracing_options`).
1205 This feature currently requires superuser privileges, as registers
1206 are queried. Only use this feature with code you trust! Call
1207 smpirun for instance via ``smpirun -wrapper "sudo "
1208 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1209 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1210 will not be required.
1212 It is planned to make this feature available on a per-process (or per-thread?) basis.
1213 The first draft, however, just implements a "global" (i.e., for all processes) set
1214 of counters, the "default" set.
1216 .. code-block:: shell
1218 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1220 .. _cfg=smpi/privatization:
1222 Automatic Privatization of Global Variables
1223 ...........................................
1225 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1227 MPI executables are usually meant to be executed in separate
1228 processes, but SMPI is executed in only one process. Global variables
1229 from executables will be placed in the same memory region and shared
1230 between processes, causing intricate bugs. Several options are
1231 possible to avoid this, as described in the main `SMPI publication
1232 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1233 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1234 automatically privatizing the globals, and this option allows one to
1235 choose between them.
1237 - **no** (default when not using smpirun): Do not automatically
1238 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1240 - **dlopen** or **yes** (default when using smpirun): Link multiple
1241 times against the binary.
1242 - **mmap** (slower, but maybe somewhat more stable):
1243 Runtime automatic switching of the data segments.
1246 This configuration option cannot be set in your platform file. You can only
1247 pass it as an argument to smpirun.
1249 .. _cfg=smpi/privatize-libs:
1251 Automatic privatization of global variables inside external libraries
1252 .....................................................................
1254 **Option** ``smpi/privatize-libs`` **default:** unset
1256 **Linux/BSD only:** When using dlopen (default) privatization,
1257 privatize specific shared libraries with internal global variables, if
1258 they can't be linked statically. For example libgfortran is usually
1259 used for Fortran I/O and indexes in files can be mixed up.
1261 Multiple libraries can be given, semicolon separated.
1263 This configuration option can only use either full paths to libraries,
1264 or full names. Check with ldd the name of the library you want to
1267 .. code-block:: shell
1271 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1274 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1275 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1276 but not ``libgfortran`` nor ``libgfortran.so``.
1278 .. _cfg=smpi/send-is-detached-thresh:
1280 Simulating MPI detached send
1281 ............................
1283 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1285 This threshold specifies the size in bytes under which the send will
1286 return immediately. This is different from the threshold detailed in
1287 :ref:`cfg=smpi/async-small-thresh` because the message is not
1288 really sent when the send is posted. SMPI still waits for the
1289 corresponding receive to be posted, in order to perform the communication
1292 .. _cfg=smpi/coll-selector:
1294 Simulating MPI collective algorithms
1295 ....................................
1297 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1299 SMPI implements more than 100 different algorithms for MPI collective
1300 communication, to accurately simulate the behavior of most of the
1301 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1302 select the decision logic either of the OpenMPI or the MPICH libraries. (By
1303 default SMPI uses naive version of collective operations.)
1305 Each collective operation can be manually selected with a
1306 ``smpi/collective_name:algo_name``. Available algorithms are listed in
1307 :ref:`SMPI_use_colls`.
1309 .. TODO:: All available collective algorithms will be made available
1310 via the ``smpirun --help-coll`` command.
1312 Add a barrier in MPI_Finalize
1313 .............................
1315 .. _cfg=smpi/finalization-barrier:
1317 **Option** ``smpi/finalization-barrier`` **default:** off
1319 By default, SMPI processes are destroyed as soon as soon as their code ends,
1320 so after a successful MPI_Finalize call returns. In some rare cases, some data
1321 might have been attached to MPI objects still active in the remaining processes,
1322 and can be destroyed eagerly by the finished process.
1323 If your code shows issues at finalization, such as segmentation fault, triggering
1324 this option will add an explicit MPI_Barrier(MPI_COMM_WORLD) call inside the
1325 MPI_Finalize, so that all processes will terminate at almost the same point.
1326 It might affect the total timing by the cost of a barrier.
1328 .. _cfg=smpi/iprobe:
1330 Inject constant times for MPI_Iprobe
1331 ....................................
1333 **Option** ``smpi/iprobe`` **default:** 0.0001
1335 The behavior and motivation for this configuration option is identical
1336 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1339 .. _cfg=smpi/iprobe-cpu-usage:
1341 Reduce speed for iprobe calls
1342 .............................
1344 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1346 MPI_Iprobe calls can be heavily used in applications. To account
1347 correctly for the energy that cores spend probing, it is necessary to
1348 reduce the load that these calls cause inside SimGrid.
1350 For instance, we measured a maximum power consumption of 220 W for a
1351 particular application but only 180 W while this application was
1352 probing. Hence, the correct factor that should be passed to this
1353 option would be 180/220 = 0.81.
1357 Inject constant times for MPI_Init
1358 ..................................
1360 **Option** ``smpi/init`` **default:** 0
1362 The behavior and motivation for this configuration option is identical
1363 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1367 Inject constant times for MPI_Isend()
1368 .....................................
1370 **Option** ``smpi/ois``
1372 The behavior and motivation for this configuration option is identical
1373 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1377 Inject constant times for MPI_send()
1378 ....................................
1380 **Option** ``smpi/os``
1382 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1383 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1384 time). SMPI can factor these costs in as well, but the user has to
1385 configure SMPI accordingly as these values may vary by machine. This
1386 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1387 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1388 exactly as ``smpi/ois``.
1390 This item can consist of multiple sections; each section takes three
1391 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1392 so this example contains two sections. Furthermore, each section
1393 consists of three values.
1395 1. The first value denotes the minimum size in bytes for this section to take effect;
1396 read it as "if message size is greater than this value (and other section has a larger
1397 first value that is also smaller than the message size), use this".
1398 In the first section above, this value is "1".
1400 2. The second value is the startup time; this is a constant value that will always
1401 be charged, no matter what the size of the message. In the first section above,
1404 3. The third value is the `per-byte` cost. That is, it is charged for every
1405 byte of the message (incurring cost messageSize*cost_per_byte)
1406 and hence accounts also for larger messages. In the first
1407 section of the example above, this value is "2".
1409 Now, SMPI always checks which section it should use for a given
1410 message; that is, if a message of size 11 is sent with the
1411 configuration of the example above, only the second section will be
1412 used, not the first, as the first value of the second section is
1413 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1414 message of size 11 incurs the following cost inside MPI_Send:
1415 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1417 Note that the order of sections can be arbitrary; they will be ordered internally.
1421 Inject constant times for MPI_Recv()
1422 ....................................
1424 **Option** ``smpi/or``
1426 The behavior and motivation for this configuration option is identical
1427 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1430 .. _cfg=smpi/grow-injected-times:
1432 Inject constant times for MPI_Test
1433 ..................................
1435 **Option** ``smpi/test`` **default:** 0.0001
1437 By setting this option, you can control the amount of time a process
1438 sleeps when MPI_Test() is called; this is important, because SimGrid
1439 normally only advances the time while communication is happening and
1440 thus, MPI_Test will not add to the time, resulting in deadlock if it is
1441 used as a break-condition as in the following example:
1446 MPI_Test(request, flag, status);
1450 To speed up execution, we use a counter to keep track of how often we
1451 checked if the handle is now valid or not. Hence, we actually
1452 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1453 process to sleep increases linearly with the number of previously
1454 failed tests. This behavior can be disabled by setting
1455 ``smpi/grow-injected-times`` to **no**. This will also disable this
1456 behavior for MPI_Iprobe.
1458 .. _cfg=smpi/shared-malloc:
1459 .. _cfg=smpi/shared-malloc-hugepage:
1464 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1466 If your simulation consumes too much memory, you may want to modify
1467 your code so that the working areas are shared by all MPI ranks. For
1468 example, in a block-cyclic matrix multiplication, you will only
1469 allocate one set of blocks, and all processes will share them.
1470 Naturally, this will lead to very wrong results, but this will save a
1471 lot of memory. So this is still desirable for some studies. For more on
1472 the motivation for that feature, please refer to the `relevant section
1473 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1474 of the SMPI CourseWare (see Activity #2.2 of the pointed
1475 assignment). In practice, change the calls for malloc() and free() into
1476 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1478 SMPI provides two algorithms for this feature. The first one, called
1479 ``local``, allocates one block per call to SMPI_SHARED_MALLOC()
1480 (each call site gets its own block) ,and this block is shared
1481 among all MPI ranks. This is implemented with the shm_* functions
1482 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1483 for each shared block.
1485 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1486 returns a new address, but it only points to a shadow block: its memory
1487 area is mapped on a 1 MiB file on disk. If the returned block is of size
1488 N MiB, then the same file is mapped N times to cover the whole block.
1489 At the end, no matter how many times you call SMPI_SHARED_MALLOC, this will
1490 only consume 1 MiB in memory.
1492 You can disable this behavior and come back to regular mallocs (for
1493 example for debugging purposes) using ``no`` as a value.
1495 If you want to keep private some parts of the buffer, for instance if these
1496 parts are used by the application logic and should not be corrupted, you
1497 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). For example:
1501 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1503 This will allocate 500 bytes to mem, such that mem[27..41] and
1504 mem[100..199] are shared while other area remain private.
1506 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1508 When smpi/shared-malloc:global is used, the memory consumption problem
1509 is solved, but it may induce too much load on the kernel's pages table.
1510 In this case, you should use huge pages so that the kernel creates only one
1511 entry per MB of malloced data instead of one entry per 4 kB.
1512 To activate this, you must mount a hugetlbfs on your system and allocate
1513 at least one huge page:
1515 .. code-block:: shell
1518 sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1519 sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1521 Then, you can pass the option
1522 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1523 actually activate the huge page support in shared mallocs.
1525 .. _cfg=smpi/auto-shared-malloc-thresh:
1527 Automatically share allocations
1528 ...............................
1530 **Option** ``smpi/auto-shared-malloc-thresh:`` **Default:** 0 (false)
1531 This value in bytes represents the size above which all allocations
1532 will be "shared" by default (as if they were performed through
1533 SMPI_SHARED_MALLOC macros). Default = 0 = disabled feature.
1534 The value must be carefully chosen to only select data buffers which
1535 will not modify execution path or cause crash if their content is false.
1536 Option :ref:`cfg=smpi/display-allocs` can be used to locate the largest
1537 allocation detected in a run, and provide a good starting threshold.
1538 Note : malloc, calloc and free are overridden by smpicc/cxx by default.
1539 This can cause some troubles if codes are already overriding these. If this
1540 is the case, defining SMPI_NO_OVERRIDE_MALLOC in the compilation flags can
1541 help, but will make this feature unusable.
1545 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1546 ...................................................................
1548 **Option** ``smpi/wtime`` **default:** 10 ns
1550 This option controls the amount of (simulated) time spent in calls to
1551 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1552 to 0, the simulated clock is not advanced in these calls, which leads
1553 to issues if your application contains such a loop:
1557 while(MPI_Wtime() < some_time_bound) {
1558 /* some tests, with no communication nor computation */
1561 When the option smpi/wtime is set to 0, the time advances only on
1562 communications and computations. So the previous code results in an
1563 infinite loop: the current [simulated] time will never reach
1564 ``some_time_bound``. This infinite loop is avoided when that option
1565 is set to a small value, as it is by default since SimGrid v3.21.
1567 Note that if your application does not contain any loop depending on
1568 the current time only, then setting this option to a non-zero value
1569 will slow down your simulations by a tiny bit: the simulation loop has
1570 to be broken out of and reset each time your code asks for the current time.
1571 If the simulation speed really matters to you, you can avoid this
1572 extra delay by setting smpi/wtime to 0.
1574 .. _cfg=smpi/list-leaks:
1576 Report leaked MPI objects
1577 .........................
1579 **Option** ``smpi/list-leaks`` **default:** 0
1581 This option controls whether to report leaked MPI objects.
1582 The parameter is the number of leaks to report.
1584 Other Configurations
1585 --------------------
1587 .. _cfg=debug/clean-atexit:
1589 Cleanup at Termination
1590 ......................
1592 **Option** ``debug/clean-atexit`` **default:** on
1594 If your code is segfaulting during its finalization, it may help to
1595 disable this option to request that SimGrid not attempt any cleanups at
1596 the end of the simulation. Since the Unix process is ending anyway,
1597 the operating system will wipe it all.
1604 **Option** ``path`` **default:** . (current dir)
1606 It is possible to specify a list of directories to search in for the
1607 trace files (see :ref:`pf_trace`) by using this configuration
1608 item. To add several directory to the path, set the configuration
1609 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1611 .. _cfg=debug/breakpoint:
1616 **Option** ``debug/breakpoint`` **default:** unset
1618 This configuration option sets a breakpoint: when the simulated clock
1619 reaches the given time, a SIGTRAP is raised. This can be used to stop
1620 the execution and get a backtrace with a debugger.
1622 It is also possible to set the breakpoint from inside the debugger, by
1623 writing in global variable simgrid::simix::breakpoint. For example,
1626 .. code-block:: shell
1628 set variable simgrid::simix::breakpoint = 3.1416
1630 .. _cfg=debug/verbose-exit:
1635 **Option** ``debug/verbose-exit`` **default:** on
1637 By default, when Ctrl-C is pressed, the status of all existing actors
1638 is displayed before exiting the simulation. This is very useful to
1639 debug your code, but it can become troublesome if you have many
1640 actors. Set this configuration item to **off** to disable this
1643 .. _cfg=exception/cutpath:
1645 Truncate local path from exception backtrace
1646 ............................................
1648 **Option** ``exception/cutpath`` **default:** off
1650 This configuration option is used to remove the path from the
1651 backtrace shown when an exception is thrown. This is mainly useful for
1652 the tests: the full file path would makes the tests non-reproducible because
1653 the paths of source files depend of the build settings. That would
1654 break most of the tests since their output is continually compared.
1658 Logging configuration
1659 ---------------------
1661 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
1662 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
1663 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
1666 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
1667 messages from your code.
1669 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
1670 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
1671 practice, the following is equivalent to the above settings: ``--log=root.thresh:error --log=s4u_host.thresh:debug``.
1673 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
1674 your settings, as in ``--log="root.thresh:error s4u_host.thresh:debug"``. The parameters are interpreted in order, from left to right.
1677 Threshold configuration
1678 .......................
1680 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
1681 ``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
1682 see, ``threshold`` can be abbreviated here.
1684 Existing thresholds:
1686 - ``trace`` some functions display a message at this level when entering or returning
1687 - ``debug`` output that is mostly useful when debugging the corresponding module.
1688 - ``verbose`` verbose output that is only mildly interesting and can easily be ignored
1689 - ``info`` usual output (this is the default threshold of all categories)
1690 - ``warning`` minor issue encountered
1691 - ``error`` issue encountered
1692 - ``critical`` major issue encountered, such as assertions failures
1696 Format configuration
1697 ....................
1699 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
1700 as the date, or the actor ID, everything. Existing format directives:
1703 - %n: line separator (LOG4J compatible)
1704 - %e: plain old space (SimGrid extension)
1706 - %m: user-provided message
1708 - %c: Category name (LOG4J compatible)
1709 - %p: Priority name (LOG4J compatible)
1711 - %h: Hostname (SimGrid extension)
1712 - %a: Actor name (SimGrid extension -- note that with SMPI this is the integer value of the process rank)
1713 - %i: Actor PID (SimGrid extension -- this is a 'i' as in 'i'dea)
1714 - %t: Thread "name" (LOG4J compatible -- actually the address of the thread in memory)
1716 - %F: file name where the log event was raised (LOG4J compatible)
1717 - %l: location where the log event was raised (LOG4J compatible, like '%%F:%%L' -- this is a l as in 'l'etter)
1718 - %L: line number where the log event was raised (LOG4J compatible)
1719 - %M: function name (LOG4J compatible -- called method name here of course).
1721 - %d: date (UNIX-like epoch)
1722 - %r: application age (time elapsed since the beginning of the application)
1725 ``--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
1726 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
1727 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
1728 provided layout is used for every messages.
1730 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
1734 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
1735 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'"``.
1736 Another option is to use the ``%e`` directive for spaces, as in ``--log=root.fmt:%l:%e[%p/%c]:%e%m%n``.
1741 The keyword ``app`` controls the appended of a logging category. For example ``--log=root.app:file:mylogfile`` redirects every output to the file ``mylogfile``.
1743 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
1744 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.
1746 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``
1747 ensures that the log file ``mylog`` will never overpass 500 bytes in size.
1749 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
1750 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.
1755 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
1756 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
1757 ``on`` (or ``yes`` or ``1``), the produced messages will also be passed to the upper appender.
1759 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
1760 ``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
1761 will only be sent to ``all.log``.
1766 ``--help-logs`` displays a complete help message about logging in SimGrid.
1768 ``--help-log-categories`` displays the actual hierarchy of log categories for this binary.
1770 ``--log=no_loc`` hides the source locations (file names and line numbers) from the messages. This is useful to make tests reproducible.