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 - **bmf/max-iterations:** :ref:`cfg=bmf/max-iterations`
88 - **bmf/precision:** :ref:`cfg=bmf/precision`
90 - **contexts/factory:** :ref:`cfg=contexts/factory`
91 - **contexts/guard-size:** :ref:`cfg=contexts/guard-size`
92 - **contexts/nthreads:** :ref:`cfg=contexts/nthreads`
93 - **contexts/stack-size:** :ref:`cfg=contexts/stack-size`
94 - **contexts/synchro:** :ref:`cfg=contexts/synchro`
96 - **cpu/maxmin-selective-update:** :ref:`Cpu Optimization Level <options_model_optim>`
97 - **cpu/model:** :ref:`options_model_select`
98 - **cpu/optim:** :ref:`Cpu Optimization Level <options_model_optim>`
100 - **debug/breakpoint:** :ref:`cfg=debug/breakpoint`
101 - **debug/clean-atexit:** :ref:`cfg=debug/clean-atexit`
102 - **debug/verbose-exit:** :ref:`cfg=debug/verbose-exit`
104 - **exception/cutpath:** :ref:`cfg=exception/cutpath`
106 - **host/model:** :ref:`options_model_select`
108 - **maxmin/precision:** :ref:`cfg=maxmin/precision`
109 - **maxmin/concurrency-limit:** :ref:`cfg=maxmin/concurrency-limit`
111 - **msg/debug-multiple-use:** :ref:`cfg=msg/debug-multiple-use`
113 - **model-check:** :ref:`options_modelchecking`
114 - **model-check/checkpoint:** :ref:`cfg=model-check/checkpoint`
115 - **model-check/communications-determinism:** :ref:`cfg=model-check/communications-determinism`
116 - **model-check/dot-output:** :ref:`cfg=model-check/dot-output`
117 - **model-check/max-depth:** :ref:`cfg=model-check/max-depth`
118 - **model-check/property:** :ref:`cfg=model-check/property`
119 - **model-check/reduction:** :ref:`cfg=model-check/reduction`
120 - **model-check/replay:** :ref:`cfg=model-check/replay`
121 - **model-check/send-determinism:** :ref:`cfg=model-check/send-determinism`
122 - **model-check/setenv:** :ref:`cfg=model-check/setenv`
123 - **model-check/termination:** :ref:`cfg=model-check/termination`
124 - **model-check/timeout:** :ref:`cfg=model-check/timeout`
125 - **model-check/visited:** :ref:`cfg=model-check/visited`
127 - **network/bandwidth-factor:** :ref:`cfg=network/bandwidth-factor`
128 - **network/crosstraffic:** :ref:`cfg=network/crosstraffic`
129 - **network/latency-factor:** :ref:`cfg=network/latency-factor`
130 - **network/loopback-lat:** :ref:`cfg=network/loopback`
131 - **network/loopback-bw:** :ref:`cfg=network/loopback`
132 - **network/maxmin-selective-update:** :ref:`Network Optimization Level <options_model_optim>`
133 - **network/model:** :ref:`options_model_select`
134 - **network/optim:** :ref:`Network Optimization Level <options_model_optim>`
135 - **network/TCP-gamma:** :ref:`cfg=network/TCP-gamma`
136 - **network/weight-S:** :ref:`cfg=network/weight-S`
138 - **ns3/TcpModel:** :ref:`options_pls`
139 - **ns3/seed:** :ref:`options_pls`
140 - **path:** :ref:`cfg=path`
141 - **plugin:** :ref:`cfg=plugin`
143 - **storage/max_file_descriptors:** :ref:`cfg=storage/max_file_descriptors`
145 - **surf/precision:** :ref:`cfg=surf/precision`
147 - **For collective operations of SMPI,** please refer to Section :ref:`cfg=smpi/coll-selector`
148 - **smpi/auto-shared-malloc-thresh:** :ref:`cfg=smpi/auto-shared-malloc-thresh`
149 - **smpi/async-small-thresh:** :ref:`cfg=smpi/async-small-thresh`
150 - **smpi/barrier-finalization:** :ref:`cfg=smpi/barrier-finalization`
151 - **smpi/barrier-collectives:** :ref:`cfg=smpi/barrier-collectives`
152 - **smpi/buffering:** :ref:`cfg=smpi/buffering`
153 - **smpi/bw-factor:** :ref:`cfg=smpi/bw-factor`
154 - **smpi/coll-selector:** :ref:`cfg=smpi/coll-selector`
155 - **smpi/comp-adjustment-file:** :ref:`cfg=smpi/comp-adjustment-file`
156 - **smpi/cpu-threshold:** :ref:`cfg=smpi/cpu-threshold`
157 - **smpi/display-allocs:** :ref:`cfg=smpi/display-allocs`
158 - **smpi/display-timing:** :ref:`cfg=smpi/display-timing`
159 - **smpi/errors-are-fatal:** :ref:`cfg=smpi/errors-are-fatal`
160 - **smpi/grow-injected-times:** :ref:`cfg=smpi/grow-injected-times`
161 - **smpi/host-speed:** :ref:`cfg=smpi/host-speed`
162 - **smpi/IB-penalty-factors:** :ref:`cfg=smpi/IB-penalty-factors`
163 - **smpi/iprobe:** :ref:`cfg=smpi/iprobe`
164 - **smpi/iprobe-cpu-usage:** :ref:`cfg=smpi/iprobe-cpu-usage`
165 - **smpi/init:** :ref:`cfg=smpi/init`
166 - **smpi/keep-temps:** :ref:`cfg=smpi/keep-temps`
167 - **smpi/lat-factor:** :ref:`cfg=smpi/lat-factor`
168 - **smpi/ois:** :ref:`cfg=smpi/ois`
169 - **smpi/or:** :ref:`cfg=smpi/or`
170 - **smpi/os:** :ref:`cfg=smpi/os`
171 - **smpi/papi-events:** :ref:`cfg=smpi/papi-events`
172 - **smpi/pedantic:** :ref:`cfg=smpi/pedantic`
173 - **smpi/privatization:** :ref:`cfg=smpi/privatization`
174 - **smpi/privatize-libs:** :ref:`cfg=smpi/privatize-libs`
175 - **smpi/send-is-detached-thresh:** :ref:`cfg=smpi/send-is-detached-thresh`
176 - **smpi/shared-malloc:** :ref:`cfg=smpi/shared-malloc`
177 - **smpi/shared-malloc-hugepage:** :ref:`cfg=smpi/shared-malloc-hugepage`
178 - **smpi/simulate-computation:** :ref:`cfg=smpi/simulate-computation`
179 - **smpi/test:** :ref:`cfg=smpi/test`
180 - **smpi/wtime:** :ref:`cfg=smpi/wtime`
181 - **smpi/list-leaks** :ref:`cfg=smpi/list-leaks`
183 - **Tracing configuration options** can be found in Section :ref:`tracing_tracing_options`
185 - **storage/model:** :ref:`options_model_select`
187 - **vm/model:** :ref:`options_model_select`
191 Configuring the Platform Models
192 -------------------------------
194 .. _options_model_select:
196 Choosing the Platform Models
197 ............................
199 SimGrid comes with several network, CPU and disk models built in,
200 and you can change the used model at runtime by changing the passed
201 configuration. The three main configuration items are given below.
202 For each of these items, passing the special ``help`` value gives you
203 a short description of all possible values (for example,
204 ``--cfg=network/model:help`` will present all provided network
205 models). Also, ``--help-models`` should provide information about all
206 models for all existing resources.
208 - ``network/model``: specify the used network model. Possible values:
210 - **LV08 (default one):** Realistic network analytic model
211 (slow-start modeled by multiplying latency by 13.01, bandwidth by
212 .97; bottleneck sharing uses a payload of S=20537 for evaluating
213 RTT). Described in `Accuracy Study and Improvement of Network
214 Simulation in the SimGrid Framework
215 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
216 - **Constant:** Simplistic network model where all communication
217 take a constant time (one second). This model provides the lowest
218 realism, but is (marginally) faster.
219 - **SMPI:** Realistic network model specifically tailored for HPC
220 settings (accurate modeling of slow start with correction factors on
221 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). This model can be
222 :ref:`further configured <options_model_network>`.
223 - **IB:** Realistic network model specifically tailored for HPC
224 settings with InfiniBand networks (accurate modeling contention
225 behavior, based on the model explained in `this PhD work
226 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
227 This model can be :ref:`further configured <options_model_network>`.
228 - **CM02:** Legacy network analytic model. Very similar to LV08, but
229 without corrective factors. The timings of small messages are thus
230 poorly modeled. This model is described in `A Network Model for
231 Simulation of Grid Application
232 <https://hal.inria.fr/inria-00071989/document>`_.
233 - **ns-3** (only available if you compiled SimGrid accordingly):
234 Use the packet-level network
235 simulators as network models (see :ref:`model_ns3`).
236 This model can be :ref:`further configured <options_pls>`.
238 - ``cpu/model``: specify the used CPU model. We have only one model
241 - **Cas01:** Simplistic CPU model (time=size/speed)
243 - ``host/model``: The host concept is the aggregation of a CPU with a
244 network card. Three models exists, but actually, only 2 of them are
245 interesting. The "compound" one is simply due to the way our
246 internal code is organized, and can easily be ignored. So at the
247 end, you have two host models: The default one allows aggregation of
248 an existing CPU model with an existing network model, but does not
249 allow parallel tasks because these beasts need some collaboration
250 between the network and CPU model.
252 - **default:** Default host model. Currently, CPU:Cas01 and
253 network:LV08 (with cross traffic enabled)
254 - **compound:** Host model that is automatically chosen if
255 you change the network and CPU models
256 - **ptask_L07:** Host model somehow similar to Cas01+CM02 but
257 allowing "parallel tasks", that are intended to model the moldable
258 tasks of the grid scheduling literature.
260 - ``storage/model``: specify the used storage model. Only one model is
262 - ``vm/model``: specify the model for virtual machines. Only one model
265 .. todo: make 'compound' the default host model.
267 .. _options_model_solver:
272 The different models rely on a linear inequalities solver to share
273 the underlying resources. SimGrid allows you to change the solver, but
274 be cautious, **don't change it unless you are 100% sure**.
276 - items ``cpu/solver``, ``network/solver``, ``disk/solver`` and ``host/solver``
277 allow you to change the solver for each model:
279 - **maxmin:** The default solver for all models except ptask. Provides a
280 max-min fairness allocation.
281 - **fairbottleneck:** The default solver for ptasks. Extends max-min to
282 allow heterogeneous resources.
283 - **bmf:** More realistic solver for heterogeneous resource sharing.
284 Implements BMF (Bottleneck max fairness) fairness. To be used with
285 parallel tasks instead of fair-bottleneck.
287 .. _options_model_optim:
292 The network and CPU models that are based on linear inequalities solver (that
293 is, all our analytical models) accept specific optimization
296 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
298 - **Lazy:** Lazy action management (partial invalidation in lmm +
299 heap in action remaining).
300 - **TI:** Trace integration. Highly optimized mode when using
301 availability traces (only available for the Cas01 CPU model for
303 - **Full:** Full update of remaining and variables. Slow but may be
304 useful when debugging.
306 - items ``network/maxmin-selective-update`` and
307 ``cpu/maxmin-selective-update``: configure whether the underlying
308 should be lazily updated or not. It should have no impact on the
309 computed timings, but should speed up the computation. |br| It is
310 still possible to disable this feature because it can reveal
311 counter-productive in very specific scenarios where the
312 interaction level is high. In particular, if all your
313 communication share a given backbone link, you should disable it:
314 without it, a simple regular loop is used to update each
315 communication. With it, each of them is still updated (because of
316 the dependency induced by the backbone), but through a complicated
317 and slow pattern that follows the actual dependencies.
319 .. _cfg=bmf/precision:
320 .. _cfg=maxmin/precision:
321 .. _cfg=surf/precision:
326 **Option** ``maxmin/precision`` **Default:** 1e-5 (in flops or bytes) |br|
327 **Option** ``surf/precision`` **Default:** 1e-9 (in seconds) |br|
328 **Option** ``bmf/precision`` **Default:** 1e-12 (no unit)
330 The analytical models handle a lot of floating point values. It is
331 possible to change the epsilon used to update and compare them through
332 this configuration item. Changing it may speedup the simulation by
333 discarding very small actions, at the price of a reduced numerical
334 precision. You can modify separately the precision used to manipulate
335 timings (in seconds) and the one used to manipulate amounts of work
338 .. _cfg=maxmin/concurrency-limit:
343 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
345 The maximum number of variables per resource can be tuned through this
346 option. You can have as many simultaneous actions per resources as you
347 want. If your simulation presents a very high level of concurrency, it
348 may help to use e.g. 100 as a value here. It means that at most 100
349 actions can consume a resource at a given time. The extraneous actions
350 are queued and wait until the amount of concurrency of the considered
351 resource lowers under the given boundary.
353 Such limitations help both to the simulation speed and simulation accuracy
354 on highly constrained scenarios, but the simulation speed suffers of this
355 setting on regular (less constrained) scenarios so it is off by default.
357 .. _cfg=bmf/max-iterations:
362 **Option** ``bmf/max-iterations`` **Default:** 1000
364 It may happen in some settings that the BMF solver fails to converge to
365 a solution, so there is a hard limit on the amount of iteration count to
366 avoid infinite loops.
368 .. _options_model_network:
370 Configuring the Network Model
371 .............................
373 .. _cfg=network/TCP-gamma:
375 Maximal TCP Window Size
376 ^^^^^^^^^^^^^^^^^^^^^^^
378 **Option** ``network/TCP-gamma`` **Default:** 4194304
380 The analytical models need to know the maximal TCP window size to take
381 the TCP congestion mechanism into account. On Linux, this value can
382 be retrieved using the following commands. Both give a set of values,
383 and you should use the last one, which is the maximal size.
385 .. code-block:: console
387 $ cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
388 $ cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
390 .. _cfg=network/bandwidth-factor:
391 .. _cfg=network/latency-factor:
392 .. _cfg=network/weight-S:
394 Correcting Important Network Parameters
395 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
397 SimGrid can take network irregularities such as a slow startup or
398 changing behavior depending on the message size into account. You
399 should not change these values unless you really know what you're
400 doing. The corresponding values were computed through data fitting
401 one the timings of packet-level simulators, as described in `Accuracy
402 Study and Improvement of Network Simulation in the SimGrid Framework
403 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
405 - **network/latency-factor**: apply a multiplier to latency.
406 Models the TCP slow-start mechanism.
407 - **network/bandwidth-factor**: actual bandwidth perceived by the
409 - **network/weight-S**: bottleneck sharing constant parameter. Used
412 These parameters are the same for all communications in your simulation,
413 independently of message size or source/destination hosts. A more flexible
414 mechanism based on callbacks was introduced in SimGrid. It provides the user
415 a callback that will be called for each communication, allowing the user
416 to set different latency and bandwidth factors, based on the message size, links used
417 or zones traversed. To more details of how to use it, please look at the
418 `examples/cpp/network-factors/s4u-network-factors.cpp <https://framagit.org/simgrid/simgrid/tree/master/examples/cpp/network-factors/s4u-network-factors.cpp>`_.
421 If you are using the SMPI model, these correction coefficients are
422 themselves corrected by constant values depending on the size of the
423 exchange. By default SMPI uses factors computed on the Stampede
424 Supercomputer at TACC, with optimal deployment of processes on
425 nodes. Again, only hardcore experts should bother about this fact.
426 For more details, see SMPI sections about :ref:`cfg=smpi/bw-factor` and :ref:`cfg=smpi/lat-factor`.
429 .. _cfg=smpi/IB-penalty-factors:
434 InfiniBand network behavior can be modeled through 3 parameters
435 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `this PhD
437 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_ (in French)
438 or more concisely in `this paper <https://hal.inria.fr/hal-00953618/document>`_,
439 even if that paper does only describe models for myrinet and ethernet.
440 You can see in Fig 2 some results for Infiniband, for example. This model
441 may be outdated by now for modern infiniband, anyway, so a new
442 validation would be good.
444 The three paramaters are defined as follows:
446 - βs: penalty factor for outgoing messages, computed by running a simple send to
447 two nodes and checking slowdown compared to a single send to one node,
449 - βe: penalty factor for ingoing messages, same computation method but with one
450 node receiving several messages
451 - γr: slowdown factor when communication buffer memory is saturated. It needs a
452 more complicated pattern to run in order to be computed (5.3 in the thesis,
453 page 107), and formula in the end is γr = time(c)/(3×βe×time(ref)), where
454 time(ref) is the time of a single comm with no contention).
456 Once these values are computed, a penalty is assessed for each message (this is
457 the part implemented in the simulator) as shown page 106 of the thesis. Here is
458 a simple translation of this text. First, some notations:
460 - ∆e(e) which corresponds to the incoming degree of node e, that is to say the number of communications having as destination node e.
461 - ∆s (s) which corresponds to the degree outgoing from node s, that is to say the number of communications sent by node s.
462 - Φ (e) which corresponds to the number of communications destined for the node e but coming from a different node.
463 - 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
465 To determine the penalty for a communication, two values need to be calculated. First, the penalty caused by the conflict in transmission, noted ps.
468 - if ∆s (i) = 1 then ps = 1.
469 - if ∆s (i) ≥ 2 and ∆e (i) ≥ 3 then ps = ∆s (i) × βs × γr
470 - else, ps = ∆s (i) × βs
473 Then, the penalty caused by the conflict in reception (noted pe) should be computed as follows:
475 - if ∆e (i) = 1 then pe = 1
476 - else, pe = Φ (e) × βe × Ω (s, e)
478 Finally, the penalty associated with the communication is:
479 p = max (ps ∈ s, pe)
481 .. _cfg=network/crosstraffic:
483 Simulating Cross-Traffic
484 ^^^^^^^^^^^^^^^^^^^^^^^^
486 Since SimGrid v3.7, cross-traffic effects can be taken into account in
487 analytical simulations. It means that ongoing and incoming
488 communication flows are treated independently. In addition, the LV08
489 model adds 0.05 of usage on the opposite direction for each new
490 created flow. This can be useful to simulate some important TCP
491 phenomena such as ack compression.
493 For that to work, your platform must have two links for each
494 pair of interconnected hosts. An example of usable platform is
495 available in ``examples/platforms/crosstraffic.xml``.
497 This is activated through the ``network/crosstraffic`` item, that
498 can be set to 0 (disable this feature) or 1 (enable it).
500 Note that with the default host model this option is activated by default.
502 .. _cfg=network/loopback:
504 Configuring loopback link
505 ^^^^^^^^^^^^^^^^^^^^^^^^^
507 Several network model provide an implicit loopback link to account for local
508 communication on a host. By default it has a 10GBps bandwidth and a null latency.
509 This can be changed with ``network/loopback-lat`` and ``network/loopback-bw``
512 .. _cfg=smpi/async-small-thresh:
514 Simulating Asynchronous Send
515 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
517 (this configuration item is experimental and may change or disappear)
519 It is possible to specify that messages below a certain size (in bytes) will be
520 sent as soon as the call to MPI_Send is issued, without waiting for
521 the correspondent receive. This threshold can be configured through
522 the ``smpi/async-small-thresh`` item. The default value is 0. This
523 behavior can also be manually set for mailboxes, by setting the
524 receiving mode of the mailbox with a call to
525 :cpp:func:`MSG_mailbox_set_async`. After this, all messages sent to
526 this mailbox will have this behavior regardless of the message size.
528 This value needs to be smaller than or equals to the threshold set at
529 :ref:`cfg=smpi/send-is-detached-thresh`, because asynchronous messages
530 are meant to be detached as well.
537 **Option** ``ns3/TcpModel`` **Default:** "default" (ns-3 default)
539 When using ns-3, there is an extra item ``ns3/TcpModel``, corresponding
540 to the ``ns3::TcpL4Protocol::SocketType`` configuration item in
541 ns-3. The only valid values (enforced on the SimGrid side) are
542 'default' (no change to the ns-3 configuration), 'NewReno' or 'Reno' or
545 **Option** ``ns3/seed`` **Default:** "" (don't set the seed in ns-3)
547 This option is the random seed to provide to ns-3 with
548 ``ns3::RngSeedManager::SetSeed`` and ``ns3::RngSeedManager::SetRun``.
550 If left blank, no seed is set in ns-3. If the value 'time' is
551 provided, the current amount of seconds since epoch is used as a seed.
552 Otherwise, the provided value must be a number to use as a seed.
554 Configuring the Storage model
555 .............................
557 .. _cfg=storage/max_file_descriptors:
559 File Descriptor Count per Host
560 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
562 **Option** ``storage/max_file_descriptors`` **Default:** 1024
564 Each host maintains a fixed-size array of its file descriptors. You
565 can change its size through this item to either enlarge it if your
566 application requires it or to reduce it to save memory space.
573 SimGrid plugins allow one to extend the framework without changing its
574 source code directly. Read the source code of the existing plugins to
575 learn how to do so (in ``src/plugins``), and ask your questions to the
576 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
577 that plugins usually register callbacks to some signals of interest.
578 If they need to store some information about a given object (Link, CPU
579 or Actor), they do so through the use of a dedicated object extension.
581 Some of the existing plugins can be activated from the command line,
582 meaning that you can activate them from the command line without any
583 modification to your simulation code. For example, you can activate
584 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
587 Here is a partial list of plugins that can be activated this way. You can get
588 the full list by passing ``--cfg=plugin:help`` to your simulator.
590 - :ref:`Host Energy <plugin_host_energy>`: models the energy dissipation of the compute units.
591 - :ref:`Link Energy <plugin_link_energy>`: models the energy dissipation of the network.
592 - :ref:`Host Load <plugin_host_load>`: monitors the load of the compute units.
594 .. _options_modelchecking:
596 Configuring the Model-Checking
597 ------------------------------
599 To enable SimGrid's model-checking support, the program should
600 be executed using the simgrid-mc wrapper:
602 .. code-block:: console
604 $ simgrid-mc ./my_program
606 Safety properties are expressed as assertions using the function
607 :cpp:func:`void MC_assert(int prop)`.
609 .. _cfg=smpi/buffering:
611 Specifying the MPI buffering behavior
612 .....................................
614 **Option** ``smpi/buffering`` **Default:** infty
616 Buffering in MPI has a huge impact on the communication semantic. For example,
617 standard blocking sends are synchronous calls when the system buffers are full
618 while these calls can complete immediately without even requiring a matching
619 receive call for small messages sent when the system buffers are empty.
621 In SMPI, this depends on the message size, that is compared against two thresholds:
623 - if (size < :ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>`) then
624 MPI_Send returns immediately, even if the corresponding receive has not be issued yet.
625 - 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
626 MPI_Send returns as soon as the corresponding receive has been issued. This is known as the eager mode.
627 - if (:ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>` < size) then
628 MPI_Send returns only when the message has actually been sent over the network. This is known as the rendez-vous mode.
630 The ``smpi/buffering`` (only valid with MC) option gives an easier interface to choose between these semantics. It can take two values:
632 - **zero:** means that buffering should be disabled. All communications are actually blocking.
633 - **infty:** means that buffering should be made infinite. All communications are non-blocking.
635 .. _cfg=model-check/property:
637 Specifying a liveness property
638 ..............................
640 **Option** ``model-check/property`` **Default:** unset
642 If you want to specify liveness properties, you have to pass them on
643 the command line, specifying the name of the file containing the
644 property, as formatted by the `ltl2ba <https://github.com/utwente-fmt/ltl2ba>`_ program.
645 Note that ltl2ba is not part of SimGrid and must be installed separately.
647 .. code-block:: console
649 $ simgrid-mc ./my_program --cfg=model-check/property:<filename>
651 .. _cfg=model-check/checkpoint:
653 Going for Stateful Verification
654 ...............................
656 By default, the system is backtracked to its initial state to explore
657 another path, instead of backtracking to the exact step before the fork
658 that we want to explore (this is called stateless verification). This
659 is done this way because saving intermediate states can rapidly
660 exhaust the available memory. If you want, you can change the value of
661 the ``model-check/checkpoint`` item. For example,
662 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
663 step. Beware, this will certainly explode your memory. Larger values
664 are probably better, make sure to experiment a bit to find the right
665 setting for your specific system.
667 .. _cfg=model-check/reduction:
669 Specifying the kind of reduction
670 ................................
672 The main issue when using the model-checking is the state space
673 explosion. You can activate some reduction technique with
674 ``--cfg=model-check/reduction:<technique>``. For now, this
675 configuration variable can take 2 values:
677 - **none:** Do not apply any kind of reduction (mandatory for
678 liveness properties, as our current DPOR algorithm breaks cycles)
679 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if
680 you verify local safety properties (default value for safety
683 Another way to mitigate the state space explosion is to search for
684 cycles in the exploration with the :ref:`cfg=model-check/visited`
685 configuration. Note that DPOR and state-equality reduction may not
686 play well together. You should choose between them.
688 Our current DPOR implementation could be improved in may ways. We are
689 currently improving its efficiency (both in term of reduction ability
690 and computational speed), and future work could make it compatible
691 with liveness properties.
693 .. _cfg=model-check/visited:
695 Size of Cycle Detection Set (state equality reduction)
696 ......................................................
698 Mc SimGrid can be asked to search for cycles during the exploration,
699 i.e. situations where a new explored state is in fact the same state
700 than a previous one.. This can prove useful to mitigate the state
701 space explosion with safety properties, and this is the crux when
702 searching for counter-examples to the liveness properties.
704 Note that this feature may break the current implementation of the
705 DPOR reduction technique.
707 The ``model-check/visited`` item is the maximum number of states, which
708 are stored in memory. If the maximum number of snapshotted state is
709 reached, some states will be removed from the memory and some cycles
710 might be missed. Small values can lead to incorrect verifications, but
711 large values can exhaust your memory and be CPU intensive as each new
712 state must be compared to that amount of older saved states.
714 The default settings depend on the kind of exploration. With safety
715 checking, no state is snapshotted and cycles cannot be detected. With
716 liveness checking, all states are snapshotted because missing a cycle
717 could hinder the exploration soundness.
719 .. _cfg=model-check/termination:
721 Non-Termination Detection
722 .........................
724 The ``model-check/termination`` configuration item can be used to
725 report if a non-termination execution path has been found. This is a
726 path with a cycle, which means that the program might never terminate.
728 This only works in safety mode, not in liveness mode.
730 This options is disabled by default.
732 .. _cfg=model-check/dot-output:
737 If set, the ``model-check/dot-output`` configuration item is the name
738 of a file in which to write a dot file of the path leading to the
739 property violation discovered (safety or liveness violation), as well
740 as the cycle for liveness properties. This dot file can then be fed to the
741 graphviz dot tool to generate a corresponding graphical representation.
743 .. _cfg=model-check/max-depth:
745 Exploration Depth Limit
746 .......................
748 The ``model-check/max-depth`` can set the maximum depth of the
749 exploration graph of the model checker. If this limit is reached, a
750 logging message is sent and the results might not be exact.
752 By default, the exploration is limited to the depth of 1000.
754 .. _cfg=model-check/timeout:
759 By default, the model checker does not handle timeout conditions: the `wait`
760 operations never time out. With the ``model-check/timeout`` configuration item
761 set to **yes**, the model checker will explore timeouts of `wait` operations.
763 .. _cfg=model-check/communications-determinism:
764 .. _cfg=model-check/send-determinism:
766 Communication Determinism
767 .........................
769 The ``model-check/communications-determinism`` and
770 ``model-check/send-determinism`` items can be used to select the
771 communication determinism mode of the model checker, which checks
772 determinism properties of the communications of an application.
774 .. _cfg=model-check/setenv:
776 Passing environment variables
777 .............................
779 You can specify extra environment variables to be set in the verified application
780 with ``model-check/setenv``. For example, you can preload a library as follows:
781 ``-cfg=model-check/setenv:LD_PRELOAD=toto;LD_LIBRARY_PATH=/tmp``.
785 Verification Performance Considerations
786 .......................................
788 The size of the stacks can have a huge impact on the memory
789 consumption when using model-checking. By default, each snapshot will
790 save a copy of the whole stacks and not only of the part that is
791 really meaningful: you should expect the contribution of the memory
792 consumption of the snapshots to be:
793 :math:`\text{number of processes} \times \text{stack size} \times \text{number of states}`.
795 When compiled against the model checker, the stacks are not
796 protected with guards: if the stack size is too small for your
797 application, the stack will silently overflow into other parts of the
798 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
800 .. _cfg=model-check/replay:
802 Replaying buggy execution paths from the model checker
803 ......................................................
805 Debugging the problems reported by the model checker is challenging:
806 First, the application under verification cannot be debugged with gdb
807 because the model checker already traces it. Then, the model checker may
808 explore several execution paths before encountering the issue, making it
809 very difficult to understand the output. Fortunately, SimGrid provides
810 the execution path leading to any reported issue so that you can replay
811 this path reported by the model checker, enabling the usage of classical
814 When the model checker finds an interesting path in the application
815 execution graph (where a safety or liveness property is violated), it
816 generates an identifier for this path. Here is an example of the output:
818 .. code-block:: console
820 [ 0.000000] (0:@) Check a safety property
821 [ 0.000000] (0:@) **************************
822 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
823 [ 0.000000] (0:@) **************************
824 [ 0.000000] (0:@) Counter-example execution trace:
825 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
826 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
827 [ 0.000000] (0:@) Path = 1/3;1/4
828 [ 0.000000] (0:@) Expanded states = 27
829 [ 0.000000] (0:@) Visited states = 68
830 [ 0.000000] (0:@) Executed transitions = 46
832 The interesting line is ``Path = 1/3;1/4``, which means that you should use
833 ``--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
834 execution path. All options (but the model checker related ones) must
835 remain the same. In particular, if you ran your application with
836 ``smpirun -wrapper simgrid-mc``, then do it again. Remove all
837 MC-related options, keep non-MC-related ones and add
838 ``--cfg=model-check/replay:???``.
840 Currently, if the path is of the form ``X;Y;Z``, each number denotes
841 the actor's pid that is selected at each indecision point. If it's of
842 the form ``X/a;Y/b``, the X and Y are the selected pids while the a
843 and b are the return values of their simcalls. In the previous
844 example, ``1/3;1/4``, you can see from the full output that the actor
845 1 is doing MC_RANDOM simcalls, so the 3 and 4 simply denote the values
846 that these simcall return on the execution branch leading to the
849 Configuring the User Code Virtualization
850 ----------------------------------------
852 .. _cfg=contexts/factory:
854 Selecting the Virtualization Factory
855 ....................................
857 **Option** contexts/factory **Default:** "raw"
859 In SimGrid, the user code is virtualized in a specific mechanism that
860 allows the simulation kernel to control its execution: when a user
861 process requires a blocking action (such as sending a message), it is
862 interrupted, and only gets released when the simulated clock reaches
863 the point where the blocking operation is done. This is explained
864 graphically in the `relevant tutorial, available online
865 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
867 In SimGrid, the containers in which user processes are virtualized are
868 called contexts. Several context factory are provided, and you can
869 select the one you want to use with the ``contexts/factory``
870 configuration item. Some of the following may not exist on your
871 machine because of portability issues. In any case, the default one
872 should be the most effcient one (please report bugs if the
873 auto-detection fails for you). They are approximately sorted here from
874 the slowest to the most efficient:
876 - **thread:** very slow factory using full featured threads (either
877 pthreads or windows native threads). They are slow but very
878 standard. Some debuggers or profilers only work with this factory.
879 - **java:** Java applications are virtualized onto java threads (that
880 are regular pthreads registered to the JVM)
881 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
882 - **boost:** This uses the `context
883 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
884 of the boost library for a performance that is comparable to our
886 |br| Install the relevant library (e.g. with the
887 libboost-contexts-dev package on Debian/Ubuntu) and recompile
889 - **raw:** amazingly fast factory using a context switching mechanism
890 of our own, directly implemented in assembly (only available for x86
891 and amd64 platforms for now) and without any unneeded system call.
893 The main reason to change this setting is when the debugging tools become
894 fooled by the optimized context factories. Threads are the most
895 debugging-friendly contexts, as they allow one to set breakpoints
896 anywhere with gdb and visualize backtraces for all processes, in order
897 to debug concurrency issues. Valgrind is also more comfortable with
898 threads, but it should be usable with all factories (Exception: the
899 callgrind tool really dislikes raw and ucontext factories).
901 .. _cfg=contexts/stack-size:
903 Adapting the Stack Size
904 .......................
906 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
908 Each virtualized used process is executed using a specific system
909 stack. The size of this stack has a huge impact on the simulation
910 scalability, but its default value is rather large. This is because
911 the error messages that you get when the stack size is too small are
912 rather disturbing: this leads to stack overflow (overwriting other
913 stacks), leading to segfaults with corrupted stack traces.
915 If you want to push the scalability limits of your code, you might
916 want to reduce the ``contexts/stack-size`` item. Its default value is
917 8192 (in KiB), while our Chord simulation works with stacks as small
918 as 16 KiB, for example. You can ensure that some actors have a specific
919 size by simply changing the value of this configuration item before
920 creating these actors. The :cpp:func:`simgrid::s4u::Engine::set_config`
921 functions are handy for that.
923 This *setting is ignored* when using the thread factory (because there
924 is no way to modify the stack size with C++ system threads). Instead,
925 you should compile SimGrid and your application with
926 ``-fsplit-stack``. Note that this compilation flag is not compatible
927 with the model checker right now.
929 The operating system should only allocate memory for the pages of the
930 stack which are actually used and you might not need to use this in
931 most cases. However, this setting is very important when using the
932 model checker (see :ref:`options_mc_perf`).
934 .. _cfg=contexts/guard-size:
936 Disabling Stack Guard Pages
937 ...........................
939 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages on Windows or with MC)
941 Unless you use the threads context factory (see
942 :ref:`cfg=contexts/factory`), a stack guard page is usually used
943 which prevents the stack of a given actor from overflowing on another
944 stack. But the performance impact may become prohibitive when the
945 amount of actors increases. The option ``contexts/guard-size`` is the
946 number of stack guard pages used. By setting it to 0, no guard pages
947 will be used: in this case, you should avoid using small stacks (with
948 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
949 will silently overflow on other parts of the memory.
951 When no stack guard page is created, stacks may then silently overflow
952 on other parts of the memory if their size is too small for the
955 .. _cfg=contexts/nthreads:
956 .. _cfg=contexts/synchro:
958 Running User Code in Parallel
959 .............................
961 Parallel execution of the user code is only considered stable in
962 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
963 simulations may well fail in parallel mode. It is described in
964 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
966 If you are using the **ucontext** or **raw** context factories, you can
967 request to execute the user code in parallel. Several threads are
968 launched, each of them handling the same number of user contexts at each
969 run. To activate this, set the ``contexts/nthreads`` item to the amount
970 of cores that you have in your computer (or lower than 1 to have the
971 amount of cores auto-detected).
973 When parallel execution is activated, you can choose the
974 synchronization schema used with the ``contexts/synchro`` item,
975 which value is either:
977 - **futex:** ultra optimized synchronisation schema, based on futexes
978 (fast user-mode mutexes), and thus only available on Linux systems.
979 This is the default mode when available.
980 - **posix:** slow but portable synchronisation using only POSIX
982 - **busy_wait:** not really a synchronisation: the worker threads
983 constantly request new contexts to execute. It should be the most
984 efficient synchronisation schema, but it loads all the cores of
985 your machine for no good reason. You probably prefer the other less
988 Configuring the Tracing
989 -----------------------
991 The :ref:`tracing subsystem <outcome_vizu>` can be configured in
992 several different ways depending on the used interface (S4U, SMPI)
993 and the kind of traces that needs to be obtained. See the
994 :ref:`Tracing Configuration Options subsection
995 <tracing_tracing_options>` for a full description of each
996 configuration option.
998 We detail here a simple way to get the traces working for you, even if
999 you never used the tracing API.
1002 - Any SimGrid-based simulator (MSG, SMPI, ...) and raw traces:
1004 .. code-block:: none
1006 --cfg=tracing:yes --cfg=tracing/uncategorized:yes
1008 The first parameter activates the tracing subsystem, and the second
1009 tells it to trace host and link utilization (without any
1012 - MSG-based simulator and categorized traces (you need to
1013 declare categories and classify your tasks according to them)
1015 .. code-block:: none
1017 --cfg=tracing:yes --cfg=tracing/categorized:yes
1019 The first parameter activates the tracing subsystem, and the second
1020 tells it to trace host and link categorized utilization.
1022 - SMPI simulator and traces for a space/time view:
1024 .. code-block:: console
1026 $ smpirun -trace ...
1028 The `-trace` parameter for the smpirun script runs the simulation
1029 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
1030 smpirun's `-help` parameter for additional tracing options.
1032 Sometimes you might want to put additional information on the trace to
1033 correctly identify them later, or to provide data that can be used to
1034 reproduce an experiment. You have two ways to do that:
1036 - Add a string on top of the trace file as comment:
1038 .. code-block:: none
1040 --cfg=tracing/comment:my_simulation_identifier
1042 - Add the contents of a textual file on top of the trace file as comment:
1044 .. code-block:: none
1046 --cfg=tracing/comment-file:my_file_with_additional_information.txt
1048 Please, use these two parameters (for comments) to make reproducible
1049 simulations. For additional details about this and all tracing
1050 options, check See the :ref:`tracing_tracing_options`.
1055 .. _cfg=msg/debug-multiple-use:
1060 **Option** ``msg/debug-multiple-use`` **Default:** off
1062 Sometimes your application may try to send a task that is still being
1063 executed somewhere else, making it impossible to send this task. However,
1064 for debugging purposes, one may want to know what the other host is/was
1065 doing. This option shows a backtrace of the other process.
1070 The SMPI interface provides several specific configuration items.
1071 These are not easy to see, since the code is usually launched through the
1072 ``smiprun`` script directly.
1074 .. _cfg=smpi/host-speed:
1075 .. _cfg=smpi/cpu-threshold:
1076 .. _cfg=smpi/simulate-computation:
1078 Automatic Benchmarking of SMPI Code
1079 ...................................
1081 In SMPI, the sequential code is automatically benchmarked, and these
1082 computations are automatically reported to the simulator. That is to
1083 say that if you have a large computation between a ``MPI_Recv()`` and
1084 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
1085 this code, and create an execution task within the simulator to take
1086 this into account. For that, the actual duration is measured on the
1087 host machine and then scaled to the power of the corresponding
1088 simulated machine. The variable ``smpi/host-speed`` allows one to
1089 specify the computational speed of the host machine (in flop/s by
1090 default) to use when scaling the execution times.
1092 The default value is ``smpi/host-speed=20kf`` (= 20,000 flop/s). This
1093 is probably underestimated for most machines, leading SimGrid to
1094 overestimate the amount of flops in the execution blocks that are
1095 automatically injected in the simulator. As a result, the execution
1096 time of the whole application will probably be overestimated until you
1097 use a realistic value.
1099 When the code consists of numerous consecutive MPI calls, the
1100 previous mechanism feeds the simulation kernel with numerous tiny
1101 computations. The ``smpi/cpu-threshold`` item becomes handy when this
1102 impacts badly on the simulation performance. It specifies a threshold (in
1103 seconds) below which the execution chunks are not reported to the
1104 simulation kernel (default value: 1e-6).
1106 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
1107 time spent below this threshold. SMPI does not consider the
1108 `amount of time` of these computations; there is no offset for
1109 this. Hence, a value that is too small, may lead to unreliable
1112 In some cases, however, one may wish to disable simulation of
1113 the computation of an application. This is the case when SMPI is used not to
1114 simulate an MPI application, but instead an MPI code that performs
1115 "live replay" of another MPI app (e.g., ScalaTrace's replay tool, or
1116 various on-line simulators that run an app at scale). In this case the
1117 computation of the replay/simulation logic should not be simulated by
1118 SMPI. Instead, the replay tool or on-line simulator will issue
1119 "computation events", which correspond to the actual MPI simulation
1120 being replayed/simulated. At the moment, these computation events can
1121 be simulated using SMPI by calling internal smpi_execute*() functions.
1123 To disable the benchmarking/simulation of a computation in the simulated
1124 application, the variable ``smpi/simulate-computation`` should be set
1125 to **no**. This option just ignores the timings in your simulation; it
1126 still executes the computations itself. If you want to stop SMPI from
1127 doing that, you should check the SMPI_SAMPLE macros, documented in
1128 Section :ref:`SMPI_use_faster`.
1130 +------------------------------------+-------------------------+-----------------------------+
1131 | Solution | Computations executed? | Computations simulated? |
1132 +====================================+=========================+=============================+
1133 | --cfg=smpi/simulate-computation:no | Yes | Never |
1134 +------------------------------------+-------------------------+-----------------------------+
1135 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
1136 +------------------------------------+-------------------------+-----------------------------+
1137 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
1138 +------------------------------------+-------------------------+-----------------------------+
1140 .. _cfg=smpi/comp-adjustment-file:
1142 Slow-down or speed-up parts of your code
1143 ........................................
1145 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
1147 This option allows you to pass a file that contains two columns: The
1148 first column defines the section that will be subject to a speedup;
1149 the second column is the speedup. For instance:
1151 .. code-block:: none
1153 "start:stop","ratio"
1154 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1156 The first line is the header - you must include it. The following
1157 line means that the code between two consecutive MPI calls on line 30
1158 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1159 of 1.18244559422142. The value for the second column is therefore a
1160 speedup, if it is larger than 1 and a slowdown if it is smaller
1161 than 1. Nothing will be changed if it is equal to 1.
1163 Of course, you can set any arbitrary filenames you want (so the start
1164 and end don't have to be in the same file), but be aware that this
1165 mechanism only supports `consecutive calls!`
1167 Please note that you must pass the ``-trace-call-location`` flag to
1168 smpicc or smpiff, respectively. This flag activates some internal
1169 macro definitions that help with obtaining the call location.
1171 .. _cfg=smpi/bw-factor:
1176 **Option** ``smpi/bw-factor``
1177 |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
1179 The possible throughput of network links is often dependent on the
1180 message sizes, as protocols may adapt to different message sizes. With
1181 this option, a series of message sizes and factors are given, helping
1182 the simulation to be more realistic. For instance, the current default
1183 value means that messages with size 65472 bytes and more will get a total of
1184 MAX_BANDWIDTH*0.940694, messages of size 15424 to 65471 will get
1185 MAX_BANDWIDTH*0.697866, and so on (where MAX_BANDWIDTH denotes the
1186 bandwidth of the link).
1188 An experimental script to compute these factors is available online. See
1189 https://framagit.org/simgrid/platform-calibration/
1190 https://simgrid.org/contrib/smpi-saturation-doc.html
1192 .. _cfg=smpi/display-timing:
1194 Reporting Simulation Time
1195 .........................
1197 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1199 Most of the time, you run MPI code with SMPI to compute the time it
1200 would take to run it on a platform. But since the code is run through
1201 the ``smpirun`` script, you don't have any control on the launcher
1202 code, making it difficult to report the simulated time when the
1203 simulation ends. If you enable the ``smpi/display-timing`` item,
1204 ``smpirun`` will display this information when the simulation
1206 SMPI will also display information about the amout of real time spent
1207 in application code and in SMPI internals, to provide hints about the
1208 need to use sampling to reduce simulation time.
1210 .. _cfg=smpi/display-allocs:
1212 Reporting memory allocations
1213 ............................
1215 **Option** ``smpi/display-allocs`` **Default:** 0 (false)
1217 SMPI intercepts malloc and calloc calls performed inside the running
1218 application, if it wasn't compiled with SMPI_NO_OVERRIDE_MALLOC.
1219 With this option, SMPI will show at the end of execution the amount of
1220 memory allocated through these calls, and locate the most expensive one.
1221 This helps finding the targets for manual memory sharing, or the threshold
1222 to use for smpi/auto-shared-malloc-thresh option (see :ref:`cfg=smpi/auto-shared-malloc-thresh`).
1224 .. _cfg=smpi/keep-temps:
1226 Keeping temporary files after simulation
1227 ........................................
1229 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1231 SMPI usually generates a lot of temporary files that are cleaned after
1232 use. This option requests to preserve them, for example to debug or
1233 profile your code. Indeed, the binary files are removed very early
1234 under the dlopen privatization schema, which tends to fool the
1237 .. _cfg=smpi/lat-factor:
1242 **Option** ``smpi/lat-factor`` |br|
1243 **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
1245 The motivation and syntax for this option is identical to the motivation/syntax
1246 of :ref:`cfg=smpi/bw-factor`.
1248 There is an important difference, though: While smpi/bw-factor `reduces` the
1249 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
1250 increase the latency, i.e., values larger than or equal to 1 are valid here.
1252 .. _cfg=smpi/papi-events:
1254 Trace hardware counters with PAPI
1255 .................................
1257 **Option** ``smpi/papi-events`` **default:** unset
1259 When the PAPI support is compiled into SimGrid, this option takes the
1260 names of PAPI counters and adds their respective values to the trace
1261 files (See Section :ref:`tracing_tracing_options`).
1265 This feature currently requires superuser privileges, as registers
1266 are queried. Only use this feature with code you trust! Call
1267 smpirun for instance via ``smpirun -wrapper "sudo "
1268 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1269 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1270 will not be required.
1272 It is planned to make this feature available on a per-process (or per-thread?) basis.
1273 The first draft, however, just implements a "global" (i.e., for all processes) set
1274 of counters, the "default" set.
1276 .. code-block:: none
1278 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1280 .. _cfg=smpi/privatization:
1282 Automatic Privatization of Global Variables
1283 ...........................................
1285 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1287 MPI executables are usually meant to be executed in separate
1288 processes, but SMPI is executed in only one process. Global variables
1289 from executables will be placed in the same memory region and shared
1290 between processes, causing intricate bugs. Several options are
1291 possible to avoid this, as described in the main `SMPI publication
1292 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1293 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1294 automatically privatizing the globals, and this option allows one to
1295 choose between them.
1297 - **no** (default when not using smpirun): Do not automatically
1298 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1300 - **dlopen** or **yes** (default when using smpirun): Link multiple
1301 times against the binary.
1302 - **mmap** (slower, but maybe somewhat more stable):
1303 Runtime automatic switching of the data segments.
1306 This configuration option cannot be set in your platform file. You can only
1307 pass it as an argument to smpirun.
1309 .. _cfg=smpi/privatize-libs:
1311 Automatic privatization of global variables inside external libraries
1312 .....................................................................
1314 **Option** ``smpi/privatize-libs`` **default:** unset
1316 **Linux/BSD only:** When using dlopen (default) privatization,
1317 privatize specific shared libraries with internal global variables, if
1318 they can't be linked statically. For example libgfortran is usually
1319 used for Fortran I/O and indexes in files can be mixed up.
1321 Multiple libraries can be given, semicolon separated.
1323 This configuration option can only use either full paths to libraries,
1324 or full names. Check with ldd the name of the library you want to
1327 .. code-block:: console
1331 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1334 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1335 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1336 but not ``libgfortran`` nor ``libgfortran.so``.
1338 .. _cfg=smpi/send-is-detached-thresh:
1340 Simulating MPI detached send
1341 ............................
1343 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1345 This threshold specifies the size in bytes under which the send will
1346 return immediately. This is different from the threshold detailed in
1347 :ref:`cfg=smpi/async-small-thresh` because the message is not
1348 really sent when the send is posted. SMPI still waits for the
1349 corresponding receive to be posted, in order to perform the communication
1352 .. _cfg=smpi/coll-selector:
1354 Simulating MPI collective algorithms
1355 ....................................
1357 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1359 SMPI implements more than 100 different algorithms for MPI collective
1360 communication, to accurately simulate the behavior of most of the
1361 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1362 select the decision logic either of the OpenMPI or the MPICH libraries. (By
1363 default SMPI uses naive version of collective operations.)
1365 Each collective operation can be manually selected with a
1366 ``smpi/collective_name:algo_name``. Available algorithms are listed in
1367 :ref:`SMPI_use_colls`.
1369 .. TODO:: All available collective algorithms will be made available
1370 via the ``smpirun --help-coll`` command.
1372 .. _cfg=smpi/barrier-collectives:
1374 Add a barrier in all collectives
1375 ................................
1377 **Option** ``smpi/barrier-collectives`` **default:** off
1379 This option adds a simple barrier in all collectives operation to catch dangerous
1380 code that may or may not work depending on the MPI implementation. It is disabled
1381 by default, and activated by the `-analyze` flag of smpirun.
1383 For example, the following code works with OpenMPI while it deadlocks in MPICH and
1384 Intel MPI. It seems to mean that OpenMPI has a "fire and forget" implementation for
1390 MPI_Bcast(buf1, buff_size, MPI_CHAR, 0, newcom);
1391 MPI_Send(&buf2, buff_size, MPI_CHAR, 1, tag, newcom);
1392 } else if (rank==1) {
1393 MPI_Recv(&buf2, buff_size, MPI_CHAR, 0, tag, newcom, MPI_STATUS_IGNORE);
1394 MPI_Bcast(buf1, buff_size, MPI_CHAR, 0, newcom);
1397 .. _cfg=smpi/barrier-finalization:
1399 Add a barrier in MPI_Finalize
1400 .............................
1402 **Option** ``smpi/finalization-barrier`` **default:** off
1404 By default, SMPI processes are destroyed as soon as soon as their code ends,
1405 so after a successful MPI_Finalize call returns. In some rare cases, some data
1406 might have been attached to MPI objects still active in the remaining processes,
1407 and can be destroyed eagerly by the finished process.
1408 If your code shows issues at finalization, such as segmentation fault, triggering
1409 this option will add an explicit MPI_Barrier(MPI_COMM_WORLD) call inside the
1410 MPI_Finalize, so that all processes will terminate at almost the same point.
1411 It might affect the total timing by the cost of a barrier.
1413 .. _cfg=smpi/errors-are-fatal:
1415 Disable MPI fatal errors
1416 ........................
1418 **Option** ``smpi/errors-are-fatal`` **default:** on
1420 By default, SMPI processes will crash if a MPI error code is returned. MPI allows
1421 to explicitely set MPI_ERRORS_RETURN errhandler to avoid this behaviour. This flag
1422 will turn on this behaviour by default (for all concerned types and errhandlers).
1423 This can ease debugging by going after the first reported error.
1425 .. _cfg=smpi/pedantic:
1427 Disable pedantic MPI errors
1428 ...........................
1430 **Option** ``smpi/pedantic`` **default:** on
1432 By default, SMPI will report all errors it finds in MPI codes. Some of these errors
1433 may not be considered as errors by all developers. This flag can be turned off to
1434 avoid reporting some usually harmless mistakes.
1435 Concerned errors list (will be expanded in the future):
1437 - Calling MPI_Win_fence only once in a program, hence just opening an epoch without
1440 .. _cfg=smpi/iprobe:
1442 Inject constant times for MPI_Iprobe
1443 ....................................
1445 **Option** ``smpi/iprobe`` **default:** 0.0001
1447 The behavior and motivation for this configuration option is identical
1448 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1451 .. _cfg=smpi/iprobe-cpu-usage:
1453 Reduce speed for iprobe calls
1454 .............................
1456 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1458 MPI_Iprobe calls can be heavily used in applications. To account
1459 correctly for the energy that cores spend probing, it is necessary to
1460 reduce the load that these calls cause inside SimGrid.
1462 For instance, we measured a maximum power consumption of 220 W for a
1463 particular application but only 180 W while this application was
1464 probing. Hence, the correct factor that should be passed to this
1465 option would be 180/220 = 0.81.
1469 Inject constant times for MPI_Init
1470 ..................................
1472 **Option** ``smpi/init`` **default:** 0
1474 The behavior and motivation for this configuration option is identical
1475 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1479 Inject constant times for MPI_Isend()
1480 .....................................
1482 **Option** ``smpi/ois``
1484 The behavior and motivation for this configuration option is identical
1485 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1489 Inject constant times for MPI_send()
1490 ....................................
1492 **Option** ``smpi/os``
1494 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1495 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1496 time). SMPI can factor these costs in as well, but the user has to
1497 configure SMPI accordingly as these values may vary by machine. This
1498 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1499 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1500 exactly as ``smpi/ois``.
1502 This item can consist of multiple sections; each section takes three
1503 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1504 so this example contains two sections. Furthermore, each section
1505 consists of three values.
1507 1. The first value denotes the minimum size in bytes for this section to take effect;
1508 read it as "if message size is greater than this value (and other section has a larger
1509 first value that is also smaller than the message size), use this".
1510 In the first section above, this value is "1".
1512 2. The second value is the startup time; this is a constant value that will always
1513 be charged, no matter what the size of the message. In the first section above,
1516 3. The third value is the `per-byte` cost. That is, it is charged for every
1517 byte of the message (incurring cost messageSize*cost_per_byte)
1518 and hence accounts also for larger messages. In the first
1519 section of the example above, this value is "2".
1521 Now, SMPI always checks which section it should use for a given
1522 message; that is, if a message of size 11 is sent with the
1523 configuration of the example above, only the second section will be
1524 used, not the first, as the first value of the second section is
1525 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1526 message of size 11 incurs the following cost inside MPI_Send:
1527 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1529 Note that the order of sections can be arbitrary; they will be ordered internally.
1533 Inject constant times for MPI_Recv()
1534 ....................................
1536 **Option** ``smpi/or``
1538 The behavior and motivation for this configuration option is identical
1539 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1542 .. _cfg=smpi/grow-injected-times:
1544 Inject constant times for MPI_Test
1545 ..................................
1547 **Option** ``smpi/test`` **default:** 0.0001
1549 By setting this option, you can control the amount of time a process
1550 sleeps when MPI_Test() is called; this is important, because SimGrid
1551 normally only advances the time while communication is happening and
1552 thus, MPI_Test will not add to the time, resulting in deadlock if it is
1553 used as a break-condition as in the following example:
1558 MPI_Test(request, flag, status);
1562 To speed up execution, we use a counter to keep track of how often we
1563 checked if the handle is now valid or not. Hence, we actually
1564 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1565 process to sleep increases linearly with the number of previously
1566 failed tests. This behavior can be disabled by setting
1567 ``smpi/grow-injected-times`` to **no**. This will also disable this
1568 behavior for MPI_Iprobe.
1570 .. _cfg=smpi/shared-malloc:
1571 .. _cfg=smpi/shared-malloc-hugepage:
1576 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1578 If your simulation consumes too much memory, you may want to modify
1579 your code so that the working areas are shared by all MPI ranks. For
1580 example, in a block-cyclic matrix multiplication, you will only
1581 allocate one set of blocks, and all processes will share them.
1582 Naturally, this will lead to very wrong results, but this will save a
1583 lot of memory. So this is still desirable for some studies. For more on
1584 the motivation for that feature, please refer to the `relevant section
1585 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1586 of the SMPI CourseWare (see Activity #2.2 of the pointed
1587 assignment). In practice, change the calls for malloc() and free() into
1588 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1590 SMPI provides two algorithms for this feature. The first one, called
1591 ``local``, allocates one block per call to SMPI_SHARED_MALLOC()
1592 (each call site gets its own block) ,and this block is shared
1593 among all MPI ranks. This is implemented with the shm_* functions
1594 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1595 for each shared block.
1597 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1598 returns a new address, but it only points to a shadow block: its memory
1599 area is mapped on a 1 MiB file on disk. If the returned block is of size
1600 N MiB, then the same file is mapped N times to cover the whole block.
1601 At the end, no matter how many times you call SMPI_SHARED_MALLOC, this will
1602 only consume 1 MiB in memory.
1604 You can disable this behavior and come back to regular mallocs (for
1605 example for debugging purposes) using ``no`` as a value.
1607 If you want to keep private some parts of the buffer, for instance if these
1608 parts are used by the application logic and should not be corrupted, you
1609 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). For example:
1613 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1615 This will allocate 500 bytes to mem, such that mem[27..41] and
1616 mem[100..199] are shared while other area remain private.
1618 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1620 When smpi/shared-malloc:global is used, the memory consumption problem
1621 is solved, but it may induce too much load on the kernel's pages table.
1622 In this case, you should use huge pages so that the kernel creates only one
1623 entry per MB of malloced data instead of one entry per 4 kB.
1624 To activate this, you must mount a hugetlbfs on your system and allocate
1625 at least one huge page:
1627 .. code-block:: console
1630 $ sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1631 $ sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1633 Then, you can pass the option
1634 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1635 actually activate the huge page support in shared mallocs.
1637 .. _cfg=smpi/auto-shared-malloc-thresh:
1639 Automatically share allocations
1640 ...............................
1642 **Option** ``smpi/auto-shared-malloc-thresh:`` **Default:** 0 (false)
1643 This value in bytes represents the size above which all allocations
1644 will be "shared" by default (as if they were performed through
1645 SMPI_SHARED_MALLOC macros). Default = 0 = disabled feature.
1646 The value must be carefully chosen to only select data buffers which
1647 will not modify execution path or cause crash if their content is false.
1648 Option :ref:`cfg=smpi/display-allocs` can be used to locate the largest
1649 allocation detected in a run, and provide a good starting threshold.
1650 Note : malloc, calloc and free are overridden by smpicc/cxx by default.
1651 This can cause some troubles if codes are already overriding these. If this
1652 is the case, defining SMPI_NO_OVERRIDE_MALLOC in the compilation flags can
1653 help, but will make this feature unusable.
1657 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1658 ...................................................................
1660 **Option** ``smpi/wtime`` **default:** 10 ns
1662 This option controls the amount of (simulated) time spent in calls to
1663 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1664 to 0, the simulated clock is not advanced in these calls, which leads
1665 to issues if your application contains such a loop:
1669 while(MPI_Wtime() < some_time_bound) {
1670 /* some tests, with no communication nor computation */
1673 When the option smpi/wtime is set to 0, the time advances only on
1674 communications and computations. So the previous code results in an
1675 infinite loop: the current [simulated] time will never reach
1676 ``some_time_bound``. This infinite loop is avoided when that option
1677 is set to a small value, as it is by default since SimGrid v3.21.
1679 Note that if your application does not contain any loop depending on
1680 the current time only, then setting this option to a non-zero value
1681 will slow down your simulations by a tiny bit: the simulation loop has
1682 to be broken out of and reset each time your code asks for the current time.
1683 If the simulation speed really matters to you, you can avoid this
1684 extra delay by setting smpi/wtime to 0.
1686 .. _cfg=smpi/list-leaks:
1688 Report leaked MPI objects
1689 .........................
1691 **Option** ``smpi/list-leaks`` **default:** 0
1693 This option controls whether to report leaked MPI objects.
1694 The parameter is the number of leaks to report.
1696 Other Configurations
1697 --------------------
1699 .. _cfg=debug/clean-atexit:
1701 Cleanup at Termination
1702 ......................
1704 **Option** ``debug/clean-atexit`` **default:** on
1706 If your code is segfaulting during its finalization, it may help to
1707 disable this option to request that SimGrid not attempt any cleanups at
1708 the end of the simulation. Since the Unix process is ending anyway,
1709 the operating system will wipe it all.
1716 **Option** ``path`` **default:** . (current dir)
1718 It is possible to specify a list of directories to search in for the
1719 trace files (see :ref:`pf_trace`) by using this configuration
1720 item. To add several directory to the path, set the configuration
1721 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1723 .. _cfg=debug/breakpoint:
1728 **Option** ``debug/breakpoint`` **default:** unset
1730 This configuration option sets a breakpoint: when the simulated clock
1731 reaches the given time, a SIGTRAP is raised. This can be used to stop
1732 the execution and get a backtrace with a debugger.
1734 It is also possible to set the breakpoint from inside the debugger, by
1735 writing in global variable simgrid::kernel::cfg_breakpoint. For example,
1738 .. code-block:: none
1740 set variable simgrid::kernel::cfg_breakpoint = 3.1416
1742 .. _cfg=debug/verbose-exit:
1747 **Option** ``debug/verbose-exit`` **default:** on
1749 By default, when Ctrl-C is pressed, the status of all existing actors
1750 is displayed before exiting the simulation. This is very useful to
1751 debug your code, but it can become troublesome if you have many
1752 actors. Set this configuration item to **off** to disable this
1755 .. _cfg=exception/cutpath:
1757 Truncate local path from exception backtrace
1758 ............................................
1760 **Option** ``exception/cutpath`` **default:** off
1762 This configuration option is used to remove the path from the
1763 backtrace shown when an exception is thrown. This is mainly useful for
1764 the tests: the full file path would makes the tests non-reproducible because
1765 the paths of source files depend of the build settings. That would
1766 break most of the tests since their output is continually compared.
1770 Logging configuration
1771 ---------------------
1773 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
1774 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
1775 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
1778 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
1779 messages from your code.
1781 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
1782 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
1783 practice, the following is equivalent to the above settings: ``--log=root.thresh:error --log=s4u_host.thresh:debug``.
1785 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
1786 your settings, as in ``--log="root.thresh:error s4u_host.thresh:debug"``. The parameters are interpreted in order, from left to right.
1789 Threshold configuration
1790 .......................
1792 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
1793 ``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
1794 see, ``threshold`` can be abbreviated here.
1796 Existing thresholds:
1798 - ``trace`` some functions display a message at this level when entering or returning
1799 - ``debug`` output that is mostly useful when debugging the corresponding module.
1800 - ``verbose`` verbose output that is only mildly interesting and can easily be ignored
1801 - ``info`` usual output (this is the default threshold of all categories)
1802 - ``warning`` minor issue encountered
1803 - ``error`` issue encountered
1804 - ``critical`` major issue encountered, such as assertions failures
1808 Format configuration
1809 ....................
1811 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
1812 as the date, or the actor ID, everything. Existing format directives:
1815 - %n: line separator (LOG4J compatible)
1816 - %e: plain old space (SimGrid extension)
1818 - %m: user-provided message
1820 - %c: Category name (LOG4J compatible)
1821 - %p: Priority name (LOG4J compatible)
1823 - %h: Hostname (SimGrid extension)
1824 - %a: Actor name (SimGrid extension -- note that with SMPI this is the integer value of the process rank)
1825 - %i: Actor PID (SimGrid extension -- this is a 'i' as in 'i'dea)
1826 - %t: Thread "name" (LOG4J compatible -- actually the address of the thread in memory)
1828 - %F: file name where the log event was raised (LOG4J compatible)
1829 - %l: location where the log event was raised (LOG4J compatible, like '%%F:%%L' -- this is a l as in 'l'etter)
1830 - %L: line number where the log event was raised (LOG4J compatible)
1831 - %M: function name (LOG4J compatible -- called method name here of course).
1833 - %d: date (UNIX-like epoch)
1834 - %r: application age (time elapsed since the beginning of the application)
1837 ``--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
1838 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
1839 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
1840 provided layout is used for every messages.
1842 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
1846 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
1847 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'"``.
1848 Another option is to use the ``%e`` directive for spaces, as in ``--log=root.fmt:%l:%e[%p/%c]:%e%m%n``.
1853 The keyword ``app`` controls the appended of a logging category. For example ``--log=root.app:file:mylogfile`` redirects every output to the file ``mylogfile``.
1855 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
1856 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.
1858 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``
1859 ensures that the log file ``mylog`` will never overpass 500 bytes in size.
1861 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
1862 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.
1867 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
1868 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
1869 ``on`` (or ``yes`` or ``1``), the produced messages will also be passed to the upper appender.
1871 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
1872 ``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
1873 will only be sent to ``all.log``.
1878 ``--help-logs`` displays a complete help message about logging in SimGrid.
1880 ``--help-log-categories`` displays the actual hierarchy of log categories for this binary.
1882 ``--log=no_loc`` hides the source locations (file names and line numbers) from the messages. This is useful to make tests reproducible.