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.
255 - ``storage/model``: specify the used storage model. Only one model is
257 - ``vm/model``: specify the model for virtual machines. Only one model
260 .. todo: make 'compound' the default host model.
262 .. _options_model_optim:
267 The network and CPU models that are based on lmm_solve (that
268 is, all our analytical models) accept specific optimization
271 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
273 - **Lazy:** Lazy action management (partial invalidation in lmm +
274 heap in action remaining).
275 - **TI:** Trace integration. Highly optimized mode when using
276 availability traces (only available for the Cas01 CPU model for
278 - **Full:** Full update of remaining and variables. Slow but may be
279 useful when debugging.
281 - items ``network/maxmin-selective-update`` and
282 ``cpu/maxmin-selective-update``: configure whether the underlying
283 should be lazily updated or not. It should have no impact on the
284 computed timings, but should speed up the computation. |br| It is
285 still possible to disable this feature because it can reveal
286 counter-productive in very specific scenarios where the
287 interaction level is high. In particular, if all your
288 communication share a given backbone link, you should disable it:
289 without it, a simple regular loop is used to update each
290 communication. With it, each of them is still updated (because of
291 the dependency induced by the backbone), but through a complicated
292 and slow pattern that follows the actual dependencies.
294 .. _cfg=maxmin/precision:
295 .. _cfg=surf/precision:
300 **Option** ``maxmin/precision`` **Default:** 0.00001 (in flops or bytes) |br|
301 **Option** ``surf/precision`` **Default:** 0.00001 (in seconds)
303 The analytical models handle a lot of floating point values. It is
304 possible to change the epsilon used to update and compare them through
305 this configuration item. Changing it may speedup the simulation by
306 discarding very small actions, at the price of a reduced numerical
307 precision. You can modify separately the precision used to manipulate
308 timings (in seconds) and the one used to manipulate amounts of work
311 .. _cfg=maxmin/concurrency-limit:
316 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
318 The maximum number of variables per resource can be tuned through this
319 option. You can have as many simultaneous actions per resources as you
320 want. If your simulation presents a very high level of concurrency, it
321 may help to use e.g. 100 as a value here. It means that at most 100
322 actions can consume a resource at a given time. The extraneous actions
323 are queued and wait until the amount of concurrency of the considered
324 resource lowers under the given boundary.
326 Such limitations help both to the simulation speed and simulation accuracy
327 on highly constrained scenarios, but the simulation speed suffers of this
328 setting on regular (less constrained) scenarios so it is off by default.
330 .. _options_model_network:
332 Configuring the Network Model
333 .............................
335 .. _cfg=network/TCP-gamma:
337 Maximal TCP Window Size
338 ^^^^^^^^^^^^^^^^^^^^^^^
340 **Option** ``network/TCP-gamma`` **Default:** 4194304
342 The analytical models need to know the maximal TCP window size to take
343 the TCP congestion mechanism into account. On Linux, this value can
344 be retrieved using the following commands. Both give a set of values,
345 and you should use the last one, which is the maximal size.
347 .. code-block:: console
349 $ cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
350 $ cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
352 .. _cfg=network/bandwidth-factor:
353 .. _cfg=network/latency-factor:
354 .. _cfg=network/weight-S:
356 Correcting Important Network Parameters
357 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
359 SimGrid can take network irregularities such as a slow startup or
360 changing behavior depending on the message size into account. You
361 should not change these values unless you really know what you're
362 doing. The corresponding values were computed through data fitting
363 one the timings of packet-level simulators, as described in `Accuracy
364 Study and Improvement of Network Simulation in the SimGrid Framework
365 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
367 - **network/latency-factor**: apply a multiplier to latency.
368 Models the TCP slow-start mechanism.
369 - **network/bandwidth-factor**: actual bandwidth perceived by the
371 - **network/weight-S**: bottleneck sharing constant parameter. Used
374 These parameters are the same for all communications in your simulation,
375 independently of message size or source/destination hosts. A more flexible
376 mechanism based on callbacks was introduced in SimGrid. It provides the user
377 a callback that will be called for each communication, allowing the user
378 to set different latency and bandwidth factors, based on the message size, links used
379 or zones traversed. To more details of how to use it, please look at the
380 `examples/cpp/network-factors/s4u-network-factors.cpp <https://framagit.org/simgrid/simgrid/tree/master/examples/cpp/network-factors/s4u-network-factors.cpp>`_.
383 If you are using the SMPI model, these correction coefficients are
384 themselves corrected by constant values depending on the size of the
385 exchange. By default SMPI uses factors computed on the Stampede
386 Supercomputer at TACC, with optimal deployment of processes on
387 nodes. Again, only hardcore experts should bother about this fact.
388 For more details, see SMPI sections about :ref:`cfg=smpi/bw-factor` and :ref:`cfg=smpi/lat-factor`.
391 .. _cfg=smpi/IB-penalty-factors:
396 InfiniBand network behavior can be modeled through 3 parameters
397 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `this PhD
399 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_ (in French)
400 or more concisely in `this paper <https://hal.inria.fr/hal-00953618/document>`_,
401 even if that paper does only describe models for myrinet and ethernet.
402 You can see in Fig 2 some results for Infiniband, for example. This model
403 may be outdated by now for modern infiniband, anyway, so a new
404 validation would be good.
406 The three paramaters are defined as follows:
408 - βs: penalty factor for outgoing messages, computed by running a simple send to
409 two nodes and checking slowdown compared to a single send to one node,
411 - βe: penalty factor for ingoing messages, same computation method but with one
412 node receiving several messages
413 - γr: slowdown factor when communication buffer memory is saturated. It needs a
414 more complicated pattern to run in order to be computed (5.3 in the thesis,
415 page 107), and formula in the end is γr = time(c)/(3×βe×time(ref)), where
416 time(ref) is the time of a single comm with no contention).
418 Once these values are computed, a penalty is assessed for each message (this is
419 the part implemented in the simulator) as shown page 106 of the thesis. Here is
420 a simple translation of this text. First, some notations:
422 - ∆e(e) which corresponds to the incoming degree of node e, that is to say the number of communications having as destination node e.
423 - ∆s (s) which corresponds to the degree outgoing from node s, that is to say the number of communications sent by node s.
424 - Φ (e) which corresponds to the number of communications destined for the node e but coming from a different node.
425 - 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
427 To determine the penalty for a communication, two values need to be calculated. First, the penalty caused by the conflict in transmission, noted ps.
430 - if ∆s (i) = 1 then ps = 1.
431 - if ∆s (i) ≥ 2 and ∆e (i) ≥ 3 then ps = ∆s (i) × βs × γr
432 - else, ps = ∆s (i) × βs
435 Then, the penalty caused by the conflict in reception (noted pe) should be computed as follows:
437 - if ∆e (i) = 1 then pe = 1
438 - else, pe = Φ (e) × βe × Ω (s, e)
440 Finally, the penalty associated with the communication is:
441 p = max (ps ∈ s, pe)
443 .. _cfg=network/crosstraffic:
445 Simulating Cross-Traffic
446 ^^^^^^^^^^^^^^^^^^^^^^^^
448 Since SimGrid v3.7, cross-traffic effects can be taken into account in
449 analytical simulations. It means that ongoing and incoming
450 communication flows are treated independently. In addition, the LV08
451 model adds 0.05 of usage on the opposite direction for each new
452 created flow. This can be useful to simulate some important TCP
453 phenomena such as ack compression.
455 For that to work, your platform must have two links for each
456 pair of interconnected hosts. An example of usable platform is
457 available in ``examples/platforms/crosstraffic.xml``.
459 This is activated through the ``network/crosstraffic`` item, that
460 can be set to 0 (disable this feature) or 1 (enable it).
462 Note that with the default host model this option is activated by default.
464 .. _cfg=network/loopback:
466 Configuring loopback link
467 ^^^^^^^^^^^^^^^^^^^^^^^^^
469 Several network model provide an implicit loopback link to account for local
470 communication on a host. By default it has a 10GBps bandwidth and a null latency.
471 This can be changed with ``network/loopback-lat`` and ``network/loopback-bw``
474 .. _cfg=smpi/async-small-thresh:
476 Simulating Asynchronous Send
477 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
479 (this configuration item is experimental and may change or disappear)
481 It is possible to specify that messages below a certain size (in bytes) will be
482 sent as soon as the call to MPI_Send is issued, without waiting for
483 the correspondent receive. This threshold can be configured through
484 the ``smpi/async-small-thresh`` item. The default value is 0. This
485 behavior can also be manually set for mailboxes, by setting the
486 receiving mode of the mailbox with a call to
487 :cpp:func:`MSG_mailbox_set_async`. After this, all messages sent to
488 this mailbox will have this behavior regardless of the message size.
490 This value needs to be smaller than or equals to the threshold set at
491 :ref:`cfg=smpi/send-is-detached-thresh`, because asynchronous messages
492 are meant to be detached as well.
499 **Option** ``ns3/TcpModel`` **Default:** "default" (ns-3 default)
501 When using ns-3, there is an extra item ``ns3/TcpModel``, corresponding
502 to the ``ns3::TcpL4Protocol::SocketType`` configuration item in
503 ns-3. The only valid values (enforced on the SimGrid side) are
504 'default' (no change to the ns-3 configuration), 'NewReno' or 'Reno' or
507 **Option** ``ns3/seed`` **Default:** "" (don't set the seed in ns-3)
509 This option is the random seed to provide to ns-3 with
510 ``ns3::RngSeedManager::SetSeed`` and ``ns3::RngSeedManager::SetRun``.
512 If left blank, no seed is set in ns-3. If the value 'time' is
513 provided, the current amount of seconds since epoch is used as a seed.
514 Otherwise, the provided value must be a number to use as a seed.
516 Configuring the Storage model
517 .............................
519 .. _cfg=storage/max_file_descriptors:
521 File Descriptor Count per Host
522 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
524 **Option** ``storage/max_file_descriptors`` **Default:** 1024
526 Each host maintains a fixed-size array of its file descriptors. You
527 can change its size through this item to either enlarge it if your
528 application requires it or to reduce it to save memory space.
535 SimGrid plugins allow one to extend the framework without changing its
536 source code directly. Read the source code of the existing plugins to
537 learn how to do so (in ``src/plugins``), and ask your questions to the
538 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
539 that plugins usually register callbacks to some signals of interest.
540 If they need to store some information about a given object (Link, CPU
541 or Actor), they do so through the use of a dedicated object extension.
543 Some of the existing plugins can be activated from the command line,
544 meaning that you can activate them from the command line without any
545 modification to your simulation code. For example, you can activate
546 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
549 Here is a partial list of plugins that can be activated this way. You can get
550 the full list by passing ``--cfg=plugin:help`` to your simulator.
552 - :ref:`Host Energy <plugin_host_energy>`: models the energy dissipation of the compute units.
553 - :ref:`Link Energy <plugin_link_energy>`: models the energy dissipation of the network.
554 - :ref:`Host Load <plugin_host_load>`: monitors the load of the compute units.
556 .. _options_modelchecking:
558 Configuring the Model-Checking
559 ------------------------------
561 To enable SimGrid's model-checking support, the program should
562 be executed using the simgrid-mc wrapper:
564 .. code-block:: console
566 $ simgrid-mc ./my_program
568 Safety properties are expressed as assertions using the function
569 :cpp:func:`void MC_assert(int prop)`.
571 .. _cfg=smpi/buffering:
573 Specifying the MPI buffering behavior
574 .....................................
576 **Option** ``smpi/buffering`` **Default:** infty
578 Buffering in MPI has a huge impact on the communication semantic. For example,
579 standard blocking sends are synchronous calls when the system buffers are full
580 while these calls can complete immediately without even requiring a matching
581 receive call for small messages sent when the system buffers are empty.
583 In SMPI, this depends on the message size, that is compared against two thresholds:
585 - if (size < :ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>`) then
586 MPI_Send returns immediately, even if the corresponding receive has not be issued yet.
587 - 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
588 MPI_Send returns as soon as the corresponding receive has been issued. This is known as the eager mode.
589 - if (:ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>` < size) then
590 MPI_Send returns only when the message has actually been sent over the network. This is known as the rendez-vous mode.
592 The ``smpi/buffering`` (only valid with MC) option gives an easier interface to choose between these semantics. It can take two values:
594 - **zero:** means that buffering should be disabled. All communications are actually blocking.
595 - **infty:** means that buffering should be made infinite. All communications are non-blocking.
597 .. _cfg=model-check/property:
599 Specifying a liveness property
600 ..............................
602 **Option** ``model-check/property`` **Default:** unset
604 If you want to specify liveness properties, you have to pass them on
605 the command line, specifying the name of the file containing the
606 property, as formatted by the `ltl2ba <https://github.com/utwente-fmt/ltl2ba>`_ program.
607 Note that ltl2ba is not part of SimGrid and must be installed separately.
609 .. code-block:: console
611 $ simgrid-mc ./my_program --cfg=model-check/property:<filename>
613 .. _cfg=model-check/checkpoint:
615 Going for Stateful Verification
616 ...............................
618 By default, the system is backtracked to its initial state to explore
619 another path, instead of backtracking to the exact step before the fork
620 that we want to explore (this is called stateless verification). This
621 is done this way because saving intermediate states can rapidly
622 exhaust the available memory. If you want, you can change the value of
623 the ``model-check/checkpoint`` item. For example,
624 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
625 step. Beware, this will certainly explode your memory. Larger values
626 are probably better, make sure to experiment a bit to find the right
627 setting for your specific system.
629 .. _cfg=model-check/reduction:
631 Specifying the kind of reduction
632 ................................
634 The main issue when using the model-checking is the state space
635 explosion. You can activate some reduction technique with
636 ``--cfg=model-check/reduction:<technique>``. For now, this
637 configuration variable can take 2 values:
639 - **none:** Do not apply any kind of reduction (mandatory for
640 liveness properties, as our current DPOR algorithm breaks cycles)
641 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if
642 you verify local safety properties (default value for safety
645 Another way to mitigate the state space explosion is to search for
646 cycles in the exploration with the :ref:`cfg=model-check/visited`
647 configuration. Note that DPOR and state-equality reduction may not
648 play well together. You should choose between them.
650 Our current DPOR implementation could be improved in may ways. We are
651 currently improving its efficiency (both in term of reduction ability
652 and computational speed), and future work could make it compatible
653 with liveness properties.
655 .. _cfg=model-check/visited:
657 Size of Cycle Detection Set (state equality reduction)
658 ......................................................
660 Mc SimGrid can be asked to search for cycles during the exploration,
661 i.e. situations where a new explored state is in fact the same state
662 than a previous one.. This can prove useful to mitigate the state
663 space explosion with safety properties, and this is the crux when
664 searching for counter-examples to the liveness properties.
666 Note that this feature may break the current implementation of the
667 DPOR reduction technique.
669 The ``model-check/visited`` item is the maximum number of states, which
670 are stored in memory. If the maximum number of snapshotted state is
671 reached, some states will be removed from the memory and some cycles
672 might be missed. Small values can lead to incorrect verifications, but
673 large values can exhaust your memory and be CPU intensive as each new
674 state must be compared to that amount of older saved states.
676 The default settings depend on the kind of exploration. With safety
677 checking, no state is snapshotted and cycles cannot be detected. With
678 liveness checking, all states are snapshotted because missing a cycle
679 could hinder the exploration soundness.
681 .. _cfg=model-check/termination:
683 Non-Termination Detection
684 .........................
686 The ``model-check/termination`` configuration item can be used to
687 report if a non-termination execution path has been found. This is a
688 path with a cycle, which means that the program might never terminate.
690 This only works in safety mode, not in liveness mode.
692 This options is disabled by default.
694 .. _cfg=model-check/dot-output:
699 If set, the ``model-check/dot-output`` configuration item is the name
700 of a file in which to write a dot file of the path leading to the
701 property violation discovered (safety or liveness violation), as well
702 as the cycle for liveness properties. This dot file can then be fed to the
703 graphviz dot tool to generate a corresponding graphical representation.
705 .. _cfg=model-check/max-depth:
707 Exploration Depth Limit
708 .......................
710 The ``model-check/max-depth`` can set the maximum depth of the
711 exploration graph of the model checker. If this limit is reached, a
712 logging message is sent and the results might not be exact.
714 By default, the exploration is limited to the depth of 1000.
716 .. _cfg=model-check/timeout:
721 By default, the model checker does not handle timeout conditions: the `wait`
722 operations never time out. With the ``model-check/timeout`` configuration item
723 set to **yes**, the model checker will explore timeouts of `wait` operations.
725 .. _cfg=model-check/communications-determinism:
726 .. _cfg=model-check/send-determinism:
728 Communication Determinism
729 .........................
731 The ``model-check/communications-determinism`` and
732 ``model-check/send-determinism`` items can be used to select the
733 communication determinism mode of the model checker, which checks
734 determinism properties of the communications of an application.
738 Verification Performance Considerations
739 .......................................
741 The size of the stacks can have a huge impact on the memory
742 consumption when using model-checking. By default, each snapshot will
743 save a copy of the whole stacks and not only of the part that is
744 really meaningful: you should expect the contribution of the memory
745 consumption of the snapshots to be:
746 :math:`\text{number of processes} \times \text{stack size} \times \text{number of states}`.
748 When compiled against the model checker, the stacks are not
749 protected with guards: if the stack size is too small for your
750 application, the stack will silently overflow into other parts of the
751 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
753 .. _cfg=model-check/replay:
755 Replaying buggy execution paths from the model checker
756 ......................................................
758 Debugging the problems reported by the model checker is challenging:
759 First, the application under verification cannot be debugged with gdb
760 because the model checker already traces it. Then, the model checker may
761 explore several execution paths before encountering the issue, making it
762 very difficult to understand the output. Fortunately, SimGrid provides
763 the execution path leading to any reported issue so that you can replay
764 this path reported by the model checker, enabling the usage of classical
767 When the model checker finds an interesting path in the application
768 execution graph (where a safety or liveness property is violated), it
769 generates an identifier for this path. Here is an example of the output:
771 .. code-block:: console
773 [ 0.000000] (0:@) Check a safety property
774 [ 0.000000] (0:@) **************************
775 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
776 [ 0.000000] (0:@) **************************
777 [ 0.000000] (0:@) Counter-example execution trace:
778 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
779 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
780 [ 0.000000] (0:@) Path = 1/3;1/4
781 [ 0.000000] (0:@) Expanded states = 27
782 [ 0.000000] (0:@) Visited states = 68
783 [ 0.000000] (0:@) Executed transitions = 46
785 The interesting line is ``Path = 1/3;1/4``, which means that you should use
786 ``--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
787 execution path. All options (but the model checker related ones) must
788 remain the same. In particular, if you ran your application with
789 ``smpirun -wrapper simgrid-mc``, then do it again. Remove all
790 MC-related options, keep non-MC-related ones and add
791 ``--cfg=model-check/replay:???``.
793 Currently, if the path is of the form ``X;Y;Z``, each number denotes
794 the actor's pid that is selected at each indecision point. If it's of
795 the form ``X/a;Y/b``, the X and Y are the selected pids while the a
796 and b are the return values of their simcalls. In the previous
797 example, ``1/3;1/4``, you can see from the full output that the actor
798 1 is doing MC_RANDOM simcalls, so the 3 and 4 simply denote the values
799 that these simcall return on the execution branch leading to the
802 Configuring the User Code Virtualization
803 ----------------------------------------
805 .. _cfg=contexts/factory:
807 Selecting the Virtualization Factory
808 ....................................
810 **Option** contexts/factory **Default:** "raw"
812 In SimGrid, the user code is virtualized in a specific mechanism that
813 allows the simulation kernel to control its execution: when a user
814 process requires a blocking action (such as sending a message), it is
815 interrupted, and only gets released when the simulated clock reaches
816 the point where the blocking operation is done. This is explained
817 graphically in the `relevant tutorial, available online
818 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
820 In SimGrid, the containers in which user processes are virtualized are
821 called contexts. Several context factory are provided, and you can
822 select the one you want to use with the ``contexts/factory``
823 configuration item. Some of the following may not exist on your
824 machine because of portability issues. In any case, the default one
825 should be the most effcient one (please report bugs if the
826 auto-detection fails for you). They are approximately sorted here from
827 the slowest to the most efficient:
829 - **thread:** very slow factory using full featured threads (either
830 pthreads or windows native threads). They are slow but very
831 standard. Some debuggers or profilers only work with this factory.
832 - **java:** Java applications are virtualized onto java threads (that
833 are regular pthreads registered to the JVM)
834 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
835 - **boost:** This uses the `context
836 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
837 of the boost library for a performance that is comparable to our
839 |br| Install the relevant library (e.g. with the
840 libboost-contexts-dev package on Debian/Ubuntu) and recompile
842 - **raw:** amazingly fast factory using a context switching mechanism
843 of our own, directly implemented in assembly (only available for x86
844 and amd64 platforms for now) and without any unneeded system call.
846 The main reason to change this setting is when the debugging tools become
847 fooled by the optimized context factories. Threads are the most
848 debugging-friendly contexts, as they allow one to set breakpoints
849 anywhere with gdb and visualize backtraces for all processes, in order
850 to debug concurrency issues. Valgrind is also more comfortable with
851 threads, but it should be usable with all factories (Exception: the
852 callgrind tool really dislikes raw and ucontext factories).
854 .. _cfg=contexts/stack-size:
856 Adapting the Stack Size
857 .......................
859 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
861 Each virtualized used process is executed using a specific system
862 stack. The size of this stack has a huge impact on the simulation
863 scalability, but its default value is rather large. This is because
864 the error messages that you get when the stack size is too small are
865 rather disturbing: this leads to stack overflow (overwriting other
866 stacks), leading to segfaults with corrupted stack traces.
868 If you want to push the scalability limits of your code, you might
869 want to reduce the ``contexts/stack-size`` item. Its default value is
870 8192 (in KiB), while our Chord simulation works with stacks as small
871 as 16 KiB, for example. You can ensure that some actors have a specific
872 size by simply changing the value of this configuration item before
873 creating these actors. The :cpp:func:`simgrid::s4u::Engine::set_config`
874 functions are handy for that.
876 This *setting is ignored* when using the thread factory (because there
877 is no way to modify the stack size with C++ system threads). Instead,
878 you should compile SimGrid and your application with
879 ``-fsplit-stack``. Note that this compilation flag is not compatible
880 with the model checker right now.
882 The operating system should only allocate memory for the pages of the
883 stack which are actually used and you might not need to use this in
884 most cases. However, this setting is very important when using the
885 model checker (see :ref:`options_mc_perf`).
887 .. _cfg=contexts/guard-size:
889 Disabling Stack Guard Pages
890 ...........................
892 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages on Windows or with MC)
894 Unless you use the threads context factory (see
895 :ref:`cfg=contexts/factory`), a stack guard page is usually used
896 which prevents the stack of a given actor from overflowing on another
897 stack. But the performance impact may become prohibitive when the
898 amount of actors increases. The option ``contexts/guard-size`` is the
899 number of stack guard pages used. By setting it to 0, no guard pages
900 will be used: in this case, you should avoid using small stacks (with
901 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
902 will silently overflow on other parts of the memory.
904 When no stack guard page is created, stacks may then silently overflow
905 on other parts of the memory if their size is too small for the
908 .. _cfg=contexts/nthreads:
909 .. _cfg=contexts/synchro:
911 Running User Code in Parallel
912 .............................
914 Parallel execution of the user code is only considered stable in
915 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
916 simulations may well fail in parallel mode. It is described in
917 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
919 If you are using the **ucontext** or **raw** context factories, you can
920 request to execute the user code in parallel. Several threads are
921 launched, each of them handling the same number of user contexts at each
922 run. To activate this, set the ``contexts/nthreads`` item to the amount
923 of cores that you have in your computer (or lower than 1 to have the
924 amount of cores auto-detected).
926 When parallel execution is activated, you can choose the
927 synchronization schema used with the ``contexts/synchro`` item,
928 which value is either:
930 - **futex:** ultra optimized synchronisation schema, based on futexes
931 (fast user-mode mutexes), and thus only available on Linux systems.
932 This is the default mode when available.
933 - **posix:** slow but portable synchronisation using only POSIX
935 - **busy_wait:** not really a synchronisation: the worker threads
936 constantly request new contexts to execute. It should be the most
937 efficient synchronisation schema, but it loads all the cores of
938 your machine for no good reason. You probably prefer the other less
941 Configuring the Tracing
942 -----------------------
944 The :ref:`tracing subsystem <outcome_vizu>` can be configured in
945 several different ways depending on the used interface (S4U, SMPI)
946 and the kind of traces that needs to be obtained. See the
947 :ref:`Tracing Configuration Options subsection
948 <tracing_tracing_options>` for a full description of each
949 configuration option.
951 We detail here a simple way to get the traces working for you, even if
952 you never used the tracing API.
955 - Any SimGrid-based simulator (MSG, SMPI, ...) and raw traces:
959 --cfg=tracing:yes --cfg=tracing/uncategorized:yes
961 The first parameter activates the tracing subsystem, and the second
962 tells it to trace host and link utilization (without any
965 - MSG-based simulator and categorized traces (you need to
966 declare categories and classify your tasks according to them)
970 --cfg=tracing:yes --cfg=tracing/categorized:yes
972 The first parameter activates the tracing subsystem, and the second
973 tells it to trace host and link categorized utilization.
975 - SMPI simulator and traces for a space/time view:
977 .. code-block:: console
981 The `-trace` parameter for the smpirun script runs the simulation
982 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
983 smpirun's `-help` parameter for additional tracing options.
985 Sometimes you might want to put additional information on the trace to
986 correctly identify them later, or to provide data that can be used to
987 reproduce an experiment. You have two ways to do that:
989 - Add a string on top of the trace file as comment:
993 --cfg=tracing/comment:my_simulation_identifier
995 - Add the contents of a textual file on top of the trace file as comment:
999 --cfg=tracing/comment-file:my_file_with_additional_information.txt
1001 Please, use these two parameters (for comments) to make reproducible
1002 simulations. For additional details about this and all tracing
1003 options, check See the :ref:`tracing_tracing_options`.
1008 .. _cfg=msg/debug-multiple-use:
1013 **Option** ``msg/debug-multiple-use`` **Default:** off
1015 Sometimes your application may try to send a task that is still being
1016 executed somewhere else, making it impossible to send this task. However,
1017 for debugging purposes, one may want to know what the other host is/was
1018 doing. This option shows a backtrace of the other process.
1023 The SMPI interface provides several specific configuration items.
1024 These are not easy to see, since the code is usually launched through the
1025 ``smiprun`` script directly.
1027 .. _cfg=smpi/host-speed:
1028 .. _cfg=smpi/cpu-threshold:
1029 .. _cfg=smpi/simulate-computation:
1031 Automatic Benchmarking of SMPI Code
1032 ...................................
1034 In SMPI, the sequential code is automatically benchmarked, and these
1035 computations are automatically reported to the simulator. That is to
1036 say that if you have a large computation between a ``MPI_Recv()`` and
1037 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
1038 this code, and create an execution task within the simulator to take
1039 this into account. For that, the actual duration is measured on the
1040 host machine and then scaled to the power of the corresponding
1041 simulated machine. The variable ``smpi/host-speed`` allows one to
1042 specify the computational speed of the host machine (in flop/s by
1043 default) to use when scaling the execution times.
1045 The default value is ``smpi/host-speed=20kf`` (= 20,000 flop/s). This
1046 is probably underestimated for most machines, leading SimGrid to
1047 overestimate the amount of flops in the execution blocks that are
1048 automatically injected in the simulator. As a result, the execution
1049 time of the whole application will probably be overestimated until you
1050 use a realistic value.
1052 When the code consists of numerous consecutive MPI calls, the
1053 previous mechanism feeds the simulation kernel with numerous tiny
1054 computations. The ``smpi/cpu-threshold`` item becomes handy when this
1055 impacts badly on the simulation performance. It specifies a threshold (in
1056 seconds) below which the execution chunks are not reported to the
1057 simulation kernel (default value: 1e-6).
1059 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
1060 time spent below this threshold. SMPI does not consider the
1061 `amount of time` of these computations; there is no offset for
1062 this. Hence, a value that is too small, may lead to unreliable
1065 In some cases, however, one may wish to disable simulation of
1066 the computation of an application. This is the case when SMPI is used not to
1067 simulate an MPI application, but instead an MPI code that performs
1068 "live replay" of another MPI app (e.g., ScalaTrace's replay tool, or
1069 various on-line simulators that run an app at scale). In this case the
1070 computation of the replay/simulation logic should not be simulated by
1071 SMPI. Instead, the replay tool or on-line simulator will issue
1072 "computation events", which correspond to the actual MPI simulation
1073 being replayed/simulated. At the moment, these computation events can
1074 be simulated using SMPI by calling internal smpi_execute*() functions.
1076 To disable the benchmarking/simulation of a computation in the simulated
1077 application, the variable ``smpi/simulate-computation`` should be set
1078 to **no**. This option just ignores the timings in your simulation; it
1079 still executes the computations itself. If you want to stop SMPI from
1080 doing that, you should check the SMPI_SAMPLE macros, documented in
1081 Section :ref:`SMPI_use_faster`.
1083 +------------------------------------+-------------------------+-----------------------------+
1084 | Solution | Computations executed? | Computations simulated? |
1085 +====================================+=========================+=============================+
1086 | --cfg=smpi/simulate-computation:no | Yes | Never |
1087 +------------------------------------+-------------------------+-----------------------------+
1088 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
1089 +------------------------------------+-------------------------+-----------------------------+
1090 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
1091 +------------------------------------+-------------------------+-----------------------------+
1093 .. _cfg=smpi/comp-adjustment-file:
1095 Slow-down or speed-up parts of your code
1096 ........................................
1098 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
1100 This option allows you to pass a file that contains two columns: The
1101 first column defines the section that will be subject to a speedup;
1102 the second column is the speedup. For instance:
1104 .. code-block:: none
1106 "start:stop","ratio"
1107 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1109 The first line is the header - you must include it. The following
1110 line means that the code between two consecutive MPI calls on line 30
1111 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1112 of 1.18244559422142. The value for the second column is therefore a
1113 speedup, if it is larger than 1 and a slowdown if it is smaller
1114 than 1. Nothing will be changed if it is equal to 1.
1116 Of course, you can set any arbitrary filenames you want (so the start
1117 and end don't have to be in the same file), but be aware that this
1118 mechanism only supports `consecutive calls!`
1120 Please note that you must pass the ``-trace-call-location`` flag to
1121 smpicc or smpiff, respectively. This flag activates some internal
1122 macro definitions that help with obtaining the call location.
1124 .. _cfg=smpi/bw-factor:
1129 **Option** ``smpi/bw-factor``
1130 |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
1132 The possible throughput of network links is often dependent on the
1133 message sizes, as protocols may adapt to different message sizes. With
1134 this option, a series of message sizes and factors are given, helping
1135 the simulation to be more realistic. For instance, the current default
1136 value means that messages with size 65472 bytes and more will get a total of
1137 MAX_BANDWIDTH*0.940694, messages of size 15424 to 65471 will get
1138 MAX_BANDWIDTH*0.697866, and so on (where MAX_BANDWIDTH denotes the
1139 bandwidth of the link).
1141 An experimental script to compute these factors is available online. See
1142 https://framagit.org/simgrid/platform-calibration/
1143 https://simgrid.org/contrib/smpi-saturation-doc.html
1145 .. _cfg=smpi/display-timing:
1147 Reporting Simulation Time
1148 .........................
1150 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1152 Most of the time, you run MPI code with SMPI to compute the time it
1153 would take to run it on a platform. But since the code is run through
1154 the ``smpirun`` script, you don't have any control on the launcher
1155 code, making it difficult to report the simulated time when the
1156 simulation ends. If you enable the ``smpi/display-timing`` item,
1157 ``smpirun`` will display this information when the simulation
1159 SMPI will also display information about the amout of real time spent
1160 in application code and in SMPI internals, to provide hints about the
1161 need to use sampling to reduce simulation time.
1163 .. _cfg=smpi/display-allocs:
1165 Reporting memory allocations
1166 ............................
1168 **Option** ``smpi/display-allocs`` **Default:** 0 (false)
1170 SMPI intercepts malloc and calloc calls performed inside the running
1171 application, if it wasn't compiled with SMPI_NO_OVERRIDE_MALLOC.
1172 With this option, SMPI will show at the end of execution the amount of
1173 memory allocated through these calls, and locate the most expensive one.
1174 This helps finding the targets for manual memory sharing, or the threshold
1175 to use for smpi/auto-shared-malloc-thresh option (see :ref:`cfg=smpi/auto-shared-malloc-thresh`).
1177 .. _cfg=smpi/keep-temps:
1179 Keeping temporary files after simulation
1180 ........................................
1182 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1184 SMPI usually generates a lot of temporary files that are cleaned after
1185 use. This option requests to preserve them, for example to debug or
1186 profile your code. Indeed, the binary files are removed very early
1187 under the dlopen privatization schema, which tends to fool the
1190 .. _cfg=smpi/lat-factor:
1195 **Option** ``smpi/lat-factor`` |br|
1196 **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
1198 The motivation and syntax for this option is identical to the motivation/syntax
1199 of :ref:`cfg=smpi/bw-factor`.
1201 There is an important difference, though: While smpi/bw-factor `reduces` the
1202 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
1203 increase the latency, i.e., values larger than or equal to 1 are valid here.
1205 .. _cfg=smpi/papi-events:
1207 Trace hardware counters with PAPI
1208 .................................
1210 **Option** ``smpi/papi-events`` **default:** unset
1212 When the PAPI support is compiled into SimGrid, this option takes the
1213 names of PAPI counters and adds their respective values to the trace
1214 files (See Section :ref:`tracing_tracing_options`).
1218 This feature currently requires superuser privileges, as registers
1219 are queried. Only use this feature with code you trust! Call
1220 smpirun for instance via ``smpirun -wrapper "sudo "
1221 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1222 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1223 will not be required.
1225 It is planned to make this feature available on a per-process (or per-thread?) basis.
1226 The first draft, however, just implements a "global" (i.e., for all processes) set
1227 of counters, the "default" set.
1229 .. code-block:: none
1231 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1233 .. _cfg=smpi/privatization:
1235 Automatic Privatization of Global Variables
1236 ...........................................
1238 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1240 MPI executables are usually meant to be executed in separate
1241 processes, but SMPI is executed in only one process. Global variables
1242 from executables will be placed in the same memory region and shared
1243 between processes, causing intricate bugs. Several options are
1244 possible to avoid this, as described in the main `SMPI publication
1245 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1246 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1247 automatically privatizing the globals, and this option allows one to
1248 choose between them.
1250 - **no** (default when not using smpirun): Do not automatically
1251 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1253 - **dlopen** or **yes** (default when using smpirun): Link multiple
1254 times against the binary.
1255 - **mmap** (slower, but maybe somewhat more stable):
1256 Runtime automatic switching of the data segments.
1259 This configuration option cannot be set in your platform file. You can only
1260 pass it as an argument to smpirun.
1262 .. _cfg=smpi/privatize-libs:
1264 Automatic privatization of global variables inside external libraries
1265 .....................................................................
1267 **Option** ``smpi/privatize-libs`` **default:** unset
1269 **Linux/BSD only:** When using dlopen (default) privatization,
1270 privatize specific shared libraries with internal global variables, if
1271 they can't be linked statically. For example libgfortran is usually
1272 used for Fortran I/O and indexes in files can be mixed up.
1274 Multiple libraries can be given, semicolon separated.
1276 This configuration option can only use either full paths to libraries,
1277 or full names. Check with ldd the name of the library you want to
1280 .. code-block:: console
1284 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1287 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1288 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1289 but not ``libgfortran`` nor ``libgfortran.so``.
1291 .. _cfg=smpi/send-is-detached-thresh:
1293 Simulating MPI detached send
1294 ............................
1296 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1298 This threshold specifies the size in bytes under which the send will
1299 return immediately. This is different from the threshold detailed in
1300 :ref:`cfg=smpi/async-small-thresh` because the message is not
1301 really sent when the send is posted. SMPI still waits for the
1302 corresponding receive to be posted, in order to perform the communication
1305 .. _cfg=smpi/coll-selector:
1307 Simulating MPI collective algorithms
1308 ....................................
1310 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1312 SMPI implements more than 100 different algorithms for MPI collective
1313 communication, to accurately simulate the behavior of most of the
1314 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1315 select the decision logic either of the OpenMPI or the MPICH libraries. (By
1316 default SMPI uses naive version of collective operations.)
1318 Each collective operation can be manually selected with a
1319 ``smpi/collective_name:algo_name``. Available algorithms are listed in
1320 :ref:`SMPI_use_colls`.
1322 .. TODO:: All available collective algorithms will be made available
1323 via the ``smpirun --help-coll`` command.
1325 .. _cfg=smpi/finalization-barrier:
1327 Add a barrier in MPI_Finalize
1328 .............................
1330 **Option** ``smpi/finalization-barrier`` **default:** off
1332 By default, SMPI processes are destroyed as soon as soon as their code ends,
1333 so after a successful MPI_Finalize call returns. In some rare cases, some data
1334 might have been attached to MPI objects still active in the remaining processes,
1335 and can be destroyed eagerly by the finished process.
1336 If your code shows issues at finalization, such as segmentation fault, triggering
1337 this option will add an explicit MPI_Barrier(MPI_COMM_WORLD) call inside the
1338 MPI_Finalize, so that all processes will terminate at almost the same point.
1339 It might affect the total timing by the cost of a barrier.
1341 .. _cfg=smpi/errors-are-fatal:
1343 Disable MPI fatal errors
1344 ........................
1346 **Option** ``smpi/errors-are-fatal`` **default:** on
1348 By default, SMPI processes will crash if a MPI error code is returned. MPI allows
1349 to explicitely set MPI_ERRORS_RETURN errhandler to avoid this behaviour. This flag
1350 will turn on this behaviour by default (for all concerned types and errhandlers).
1351 This can ease debugging by going after the first reported error.
1353 .. _cfg=smpi/pedantic:
1355 Disable pedantic MPI errors
1356 ...........................
1358 **Option** ``smpi/pedantic`` **default:** on
1360 By default, SMPI will report all errors it finds in MPI codes. Some of these errors
1361 may not be considered as errors by all developers. This flag can be turned off to
1362 avoid reporting some usually harmless mistakes.
1363 Concerned errors list (will be expanded in the future):
1365 - Calling MPI_Win_fence only once in a program, hence just opening an epoch without
1368 .. _cfg=smpi/iprobe:
1370 Inject constant times for MPI_Iprobe
1371 ....................................
1373 **Option** ``smpi/iprobe`` **default:** 0.0001
1375 The behavior and motivation for this configuration option is identical
1376 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1379 .. _cfg=smpi/iprobe-cpu-usage:
1381 Reduce speed for iprobe calls
1382 .............................
1384 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1386 MPI_Iprobe calls can be heavily used in applications. To account
1387 correctly for the energy that cores spend probing, it is necessary to
1388 reduce the load that these calls cause inside SimGrid.
1390 For instance, we measured a maximum power consumption of 220 W for a
1391 particular application but only 180 W while this application was
1392 probing. Hence, the correct factor that should be passed to this
1393 option would be 180/220 = 0.81.
1397 Inject constant times for MPI_Init
1398 ..................................
1400 **Option** ``smpi/init`` **default:** 0
1402 The behavior and motivation for this configuration option is identical
1403 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1407 Inject constant times for MPI_Isend()
1408 .....................................
1410 **Option** ``smpi/ois``
1412 The behavior and motivation for this configuration option is identical
1413 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1417 Inject constant times for MPI_send()
1418 ....................................
1420 **Option** ``smpi/os``
1422 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1423 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1424 time). SMPI can factor these costs in as well, but the user has to
1425 configure SMPI accordingly as these values may vary by machine. This
1426 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1427 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1428 exactly as ``smpi/ois``.
1430 This item can consist of multiple sections; each section takes three
1431 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1432 so this example contains two sections. Furthermore, each section
1433 consists of three values.
1435 1. The first value denotes the minimum size in bytes for this section to take effect;
1436 read it as "if message size is greater than this value (and other section has a larger
1437 first value that is also smaller than the message size), use this".
1438 In the first section above, this value is "1".
1440 2. The second value is the startup time; this is a constant value that will always
1441 be charged, no matter what the size of the message. In the first section above,
1444 3. The third value is the `per-byte` cost. That is, it is charged for every
1445 byte of the message (incurring cost messageSize*cost_per_byte)
1446 and hence accounts also for larger messages. In the first
1447 section of the example above, this value is "2".
1449 Now, SMPI always checks which section it should use for a given
1450 message; that is, if a message of size 11 is sent with the
1451 configuration of the example above, only the second section will be
1452 used, not the first, as the first value of the second section is
1453 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1454 message of size 11 incurs the following cost inside MPI_Send:
1455 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1457 Note that the order of sections can be arbitrary; they will be ordered internally.
1461 Inject constant times for MPI_Recv()
1462 ....................................
1464 **Option** ``smpi/or``
1466 The behavior and motivation for this configuration option is identical
1467 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1470 .. _cfg=smpi/grow-injected-times:
1472 Inject constant times for MPI_Test
1473 ..................................
1475 **Option** ``smpi/test`` **default:** 0.0001
1477 By setting this option, you can control the amount of time a process
1478 sleeps when MPI_Test() is called; this is important, because SimGrid
1479 normally only advances the time while communication is happening and
1480 thus, MPI_Test will not add to the time, resulting in deadlock if it is
1481 used as a break-condition as in the following example:
1486 MPI_Test(request, flag, status);
1490 To speed up execution, we use a counter to keep track of how often we
1491 checked if the handle is now valid or not. Hence, we actually
1492 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1493 process to sleep increases linearly with the number of previously
1494 failed tests. This behavior can be disabled by setting
1495 ``smpi/grow-injected-times`` to **no**. This will also disable this
1496 behavior for MPI_Iprobe.
1498 .. _cfg=smpi/shared-malloc:
1499 .. _cfg=smpi/shared-malloc-hugepage:
1504 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1506 If your simulation consumes too much memory, you may want to modify
1507 your code so that the working areas are shared by all MPI ranks. For
1508 example, in a block-cyclic matrix multiplication, you will only
1509 allocate one set of blocks, and all processes will share them.
1510 Naturally, this will lead to very wrong results, but this will save a
1511 lot of memory. So this is still desirable for some studies. For more on
1512 the motivation for that feature, please refer to the `relevant section
1513 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1514 of the SMPI CourseWare (see Activity #2.2 of the pointed
1515 assignment). In practice, change the calls for malloc() and free() into
1516 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1518 SMPI provides two algorithms for this feature. The first one, called
1519 ``local``, allocates one block per call to SMPI_SHARED_MALLOC()
1520 (each call site gets its own block) ,and this block is shared
1521 among all MPI ranks. This is implemented with the shm_* functions
1522 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1523 for each shared block.
1525 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1526 returns a new address, but it only points to a shadow block: its memory
1527 area is mapped on a 1 MiB file on disk. If the returned block is of size
1528 N MiB, then the same file is mapped N times to cover the whole block.
1529 At the end, no matter how many times you call SMPI_SHARED_MALLOC, this will
1530 only consume 1 MiB in memory.
1532 You can disable this behavior and come back to regular mallocs (for
1533 example for debugging purposes) using ``no`` as a value.
1535 If you want to keep private some parts of the buffer, for instance if these
1536 parts are used by the application logic and should not be corrupted, you
1537 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). For example:
1541 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1543 This will allocate 500 bytes to mem, such that mem[27..41] and
1544 mem[100..199] are shared while other area remain private.
1546 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1548 When smpi/shared-malloc:global is used, the memory consumption problem
1549 is solved, but it may induce too much load on the kernel's pages table.
1550 In this case, you should use huge pages so that the kernel creates only one
1551 entry per MB of malloced data instead of one entry per 4 kB.
1552 To activate this, you must mount a hugetlbfs on your system and allocate
1553 at least one huge page:
1555 .. code-block:: console
1558 $ sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1559 $ sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1561 Then, you can pass the option
1562 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1563 actually activate the huge page support in shared mallocs.
1565 .. _cfg=smpi/auto-shared-malloc-thresh:
1567 Automatically share allocations
1568 ...............................
1570 **Option** ``smpi/auto-shared-malloc-thresh:`` **Default:** 0 (false)
1571 This value in bytes represents the size above which all allocations
1572 will be "shared" by default (as if they were performed through
1573 SMPI_SHARED_MALLOC macros). Default = 0 = disabled feature.
1574 The value must be carefully chosen to only select data buffers which
1575 will not modify execution path or cause crash if their content is false.
1576 Option :ref:`cfg=smpi/display-allocs` can be used to locate the largest
1577 allocation detected in a run, and provide a good starting threshold.
1578 Note : malloc, calloc and free are overridden by smpicc/cxx by default.
1579 This can cause some troubles if codes are already overriding these. If this
1580 is the case, defining SMPI_NO_OVERRIDE_MALLOC in the compilation flags can
1581 help, but will make this feature unusable.
1585 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1586 ...................................................................
1588 **Option** ``smpi/wtime`` **default:** 10 ns
1590 This option controls the amount of (simulated) time spent in calls to
1591 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1592 to 0, the simulated clock is not advanced in these calls, which leads
1593 to issues if your application contains such a loop:
1597 while(MPI_Wtime() < some_time_bound) {
1598 /* some tests, with no communication nor computation */
1601 When the option smpi/wtime is set to 0, the time advances only on
1602 communications and computations. So the previous code results in an
1603 infinite loop: the current [simulated] time will never reach
1604 ``some_time_bound``. This infinite loop is avoided when that option
1605 is set to a small value, as it is by default since SimGrid v3.21.
1607 Note that if your application does not contain any loop depending on
1608 the current time only, then setting this option to a non-zero value
1609 will slow down your simulations by a tiny bit: the simulation loop has
1610 to be broken out of and reset each time your code asks for the current time.
1611 If the simulation speed really matters to you, you can avoid this
1612 extra delay by setting smpi/wtime to 0.
1614 .. _cfg=smpi/list-leaks:
1616 Report leaked MPI objects
1617 .........................
1619 **Option** ``smpi/list-leaks`` **default:** 0
1621 This option controls whether to report leaked MPI objects.
1622 The parameter is the number of leaks to report.
1624 Other Configurations
1625 --------------------
1627 .. _cfg=debug/clean-atexit:
1629 Cleanup at Termination
1630 ......................
1632 **Option** ``debug/clean-atexit`` **default:** on
1634 If your code is segfaulting during its finalization, it may help to
1635 disable this option to request that SimGrid not attempt any cleanups at
1636 the end of the simulation. Since the Unix process is ending anyway,
1637 the operating system will wipe it all.
1644 **Option** ``path`` **default:** . (current dir)
1646 It is possible to specify a list of directories to search in for the
1647 trace files (see :ref:`pf_trace`) by using this configuration
1648 item. To add several directory to the path, set the configuration
1649 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1651 .. _cfg=debug/breakpoint:
1656 **Option** ``debug/breakpoint`` **default:** unset
1658 This configuration option sets a breakpoint: when the simulated clock
1659 reaches the given time, a SIGTRAP is raised. This can be used to stop
1660 the execution and get a backtrace with a debugger.
1662 It is also possible to set the breakpoint from inside the debugger, by
1663 writing in global variable simgrid::simix::breakpoint. For example,
1666 .. code-block:: none
1668 set variable simgrid::simix::breakpoint = 3.1416
1670 .. _cfg=debug/verbose-exit:
1675 **Option** ``debug/verbose-exit`` **default:** on
1677 By default, when Ctrl-C is pressed, the status of all existing actors
1678 is displayed before exiting the simulation. This is very useful to
1679 debug your code, but it can become troublesome if you have many
1680 actors. Set this configuration item to **off** to disable this
1683 .. _cfg=exception/cutpath:
1685 Truncate local path from exception backtrace
1686 ............................................
1688 **Option** ``exception/cutpath`` **default:** off
1690 This configuration option is used to remove the path from the
1691 backtrace shown when an exception is thrown. This is mainly useful for
1692 the tests: the full file path would makes the tests non-reproducible because
1693 the paths of source files depend of the build settings. That would
1694 break most of the tests since their output is continually compared.
1698 Logging configuration
1699 ---------------------
1701 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
1702 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
1703 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
1706 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
1707 messages from your code.
1709 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
1710 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
1711 practice, the following is equivalent to the above settings: ``--log=root.thresh:error --log=s4u_host.thresh:debug``.
1713 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
1714 your settings, as in ``--log="root.thresh:error s4u_host.thresh:debug"``. The parameters are interpreted in order, from left to right.
1717 Threshold configuration
1718 .......................
1720 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
1721 ``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
1722 see, ``threshold`` can be abbreviated here.
1724 Existing thresholds:
1726 - ``trace`` some functions display a message at this level when entering or returning
1727 - ``debug`` output that is mostly useful when debugging the corresponding module.
1728 - ``verbose`` verbose output that is only mildly interesting and can easily be ignored
1729 - ``info`` usual output (this is the default threshold of all categories)
1730 - ``warning`` minor issue encountered
1731 - ``error`` issue encountered
1732 - ``critical`` major issue encountered, such as assertions failures
1736 Format configuration
1737 ....................
1739 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
1740 as the date, or the actor ID, everything. Existing format directives:
1743 - %n: line separator (LOG4J compatible)
1744 - %e: plain old space (SimGrid extension)
1746 - %m: user-provided message
1748 - %c: Category name (LOG4J compatible)
1749 - %p: Priority name (LOG4J compatible)
1751 - %h: Hostname (SimGrid extension)
1752 - %a: Actor name (SimGrid extension -- note that with SMPI this is the integer value of the process rank)
1753 - %i: Actor PID (SimGrid extension -- this is a 'i' as in 'i'dea)
1754 - %t: Thread "name" (LOG4J compatible -- actually the address of the thread in memory)
1756 - %F: file name where the log event was raised (LOG4J compatible)
1757 - %l: location where the log event was raised (LOG4J compatible, like '%%F:%%L' -- this is a l as in 'l'etter)
1758 - %L: line number where the log event was raised (LOG4J compatible)
1759 - %M: function name (LOG4J compatible -- called method name here of course).
1761 - %d: date (UNIX-like epoch)
1762 - %r: application age (time elapsed since the beginning of the application)
1765 ``--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
1766 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
1767 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
1768 provided layout is used for every messages.
1770 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
1774 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
1775 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'"``.
1776 Another option is to use the ``%e`` directive for spaces, as in ``--log=root.fmt:%l:%e[%p/%c]:%e%m%n``.
1781 The keyword ``app`` controls the appended of a logging category. For example ``--log=root.app:file:mylogfile`` redirects every output to the file ``mylogfile``.
1783 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
1784 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.
1786 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``
1787 ensures that the log file ``mylog`` will never overpass 500 bytes in size.
1789 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
1790 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.
1795 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
1796 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
1797 ``on`` (or ``yes`` or ``1``), the produced messages will also be passed to the upper appender.
1799 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
1800 ``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
1801 will only be sent to ``all.log``.
1806 ``--help-logs`` displays a complete help message about logging in SimGrid.
1808 ``--help-log-categories`` displays the actual hierarchy of log categories for this binary.
1810 ``--log=no_loc`` hides the source locations (file names and line numbers) from the messages. This is useful to make tests reproducible.