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`.
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 - **model-check:** :ref:`options_modelchecking`
112 - **model-check/checkpoint:** :ref:`cfg=model-check/checkpoint`
113 - **model-check/communications-determinism:** :ref:`cfg=model-check/communications-determinism`
114 - **model-check/dot-output:** :ref:`cfg=model-check/dot-output`
115 - **model-check/max-depth:** :ref:`cfg=model-check/max-depth`
116 - **model-check/property:** :ref:`cfg=model-check/property`
117 - **model-check/reduction:** :ref:`cfg=model-check/reduction`
118 - **model-check/replay:** :ref:`cfg=model-check/replay`
119 - **model-check/send-determinism:** :ref:`cfg=model-check/send-determinism`
120 - **model-check/setenv:** :ref:`cfg=model-check/setenv`
121 - **model-check/termination:** :ref:`cfg=model-check/termination`
122 - **model-check/timeout:** :ref:`cfg=model-check/timeout`
123 - **model-check/visited:** :ref:`cfg=model-check/visited`
125 - **network/bandwidth-factor:** :ref:`cfg=network/bandwidth-factor`
126 - **network/crosstraffic:** :ref:`cfg=network/crosstraffic`
127 - **network/latency-factor:** :ref:`cfg=network/latency-factor`
128 - **network/loopback-lat:** :ref:`cfg=network/loopback`
129 - **network/loopback-bw:** :ref:`cfg=network/loopback`
130 - **network/maxmin-selective-update:** :ref:`Network Optimization Level <options_model_optim>`
131 - **network/model:** :ref:`options_model_select`
132 - **network/optim:** :ref:`Network Optimization Level <options_model_optim>`
133 - **network/TCP-gamma:** :ref:`cfg=network/TCP-gamma`
134 - **network/weight-S:** :ref:`cfg=network/weight-S`
136 - **ns3/TcpModel:** :ref:`options_pls`
137 - **ns3/seed:** :ref:`options_pls`
138 - **path:** :ref:`cfg=path`
139 - **plugin:** :ref:`cfg=plugin`
141 - **storage/max_file_descriptors:** :ref:`cfg=storage/max_file_descriptors`
143 - **surf/precision:** :ref:`cfg=surf/precision`
145 - **For collective operations of SMPI,** please refer to Section :ref:`cfg=smpi/coll-selector`
146 - **smpi/auto-shared-malloc-thresh:** :ref:`cfg=smpi/auto-shared-malloc-thresh`
147 - **smpi/async-small-thresh:** :ref:`cfg=smpi/async-small-thresh`
148 - **smpi/barrier-finalization:** :ref:`cfg=smpi/barrier-finalization`
149 - **smpi/barrier-collectives:** :ref:`cfg=smpi/barrier-collectives`
150 - **smpi/buffering:** :ref:`cfg=smpi/buffering`
151 - **smpi/coll-selector:** :ref:`cfg=smpi/coll-selector`
152 - **smpi/comp-adjustment-file:** :ref:`cfg=smpi/comp-adjustment-file`
153 - **smpi/cpu-threshold:** :ref:`cfg=smpi/cpu-threshold`
154 - **smpi/display-allocs:** :ref:`cfg=smpi/display-allocs`
155 - **smpi/display-timing:** :ref:`cfg=smpi/display-timing`
156 - **smpi/errors-are-fatal:** :ref:`cfg=smpi/errors-are-fatal`
157 - **smpi/grow-injected-times:** :ref:`cfg=smpi/grow-injected-times`
158 - **smpi/host-speed:** :ref:`cfg=smpi/host-speed`
159 - **smpi/IB-penalty-factors:** :ref:`cfg=smpi/IB-penalty-factors`
160 - **smpi/iprobe:** :ref:`cfg=smpi/iprobe`
161 - **smpi/iprobe-cpu-usage:** :ref:`cfg=smpi/iprobe-cpu-usage`
162 - **smpi/init:** :ref:`cfg=smpi/init`
163 - **smpi/keep-temps:** :ref:`cfg=smpi/keep-temps`
164 - **smpi/ois:** :ref:`cfg=smpi/ois`
165 - **smpi/or:** :ref:`cfg=smpi/or`
166 - **smpi/os:** :ref:`cfg=smpi/os`
167 - **smpi/papi-events:** :ref:`cfg=smpi/papi-events`
168 - **smpi/pedantic:** :ref:`cfg=smpi/pedantic`
169 - **smpi/privatization:** :ref:`cfg=smpi/privatization`
170 - **smpi/privatize-libs:** :ref:`cfg=smpi/privatize-libs`
171 - **smpi/send-is-detached-thresh:** :ref:`cfg=smpi/send-is-detached-thresh`
172 - **smpi/shared-malloc:** :ref:`cfg=smpi/shared-malloc`
173 - **smpi/shared-malloc-hugepage:** :ref:`cfg=smpi/shared-malloc-hugepage`
174 - **smpi/simulate-computation:** :ref:`cfg=smpi/simulate-computation`
175 - **smpi/test:** :ref:`cfg=smpi/test`
176 - **smpi/wtime:** :ref:`cfg=smpi/wtime`
177 - **smpi/list-leaks** :ref:`cfg=smpi/list-leaks`
179 - **Tracing configuration options** can be found in Section :ref:`tracing_tracing_options`
181 - **storage/model:** :ref:`options_model_select`
183 - **vm/model:** :ref:`options_model_select`
187 Configuring the Platform Models
188 -------------------------------
190 .. _options_model_select:
192 Choosing the Platform Models
193 ............................
195 SimGrid comes with several network, CPU and disk models built in,
196 and you can change the used model at runtime by changing the passed
197 configuration. The three main configuration items are given below.
198 For each of these items, passing the special ``help`` value gives you
199 a short description of all possible values (for example,
200 ``--cfg=network/model:help`` will present all provided network
201 models). Also, ``--help-models`` should provide information about all
202 models for all existing resources.
204 - ``network/model``: specify the used network model. Possible values:
206 - **LV08 (default one):** Realistic network analytic model
207 (slow-start modeled by multiplying latency by 13.01, bandwidth by
208 .97; bottleneck sharing uses a payload of S=20537 for evaluating
209 RTT). Described in `Accuracy Study and Improvement of Network
210 Simulation in the SimGrid Framework
211 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
212 - **Constant:** Simplistic network model where all communication
213 take a constant time (one second). This model provides the lowest
214 realism, but is (marginally) faster.
215 - **SMPI:** Realistic network model specifically tailored for HPC
216 settings (accurate modeling of slow start with correction factors on
217 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). This model can be
218 :ref:`further configured <options_model_network>`.
219 - **IB:** Realistic network model specifically tailored for HPC
220 settings with InfiniBand networks (accurate modeling contention
221 behavior, based on the model explained in `this PhD work
222 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
223 This model can be :ref:`further configured <options_model_network>`.
224 - **CM02:** Legacy network analytic model. Very similar to LV08, but
225 without corrective factors. The timings of small messages are thus
226 poorly modeled. This model is described in `A Network Model for
227 Simulation of Grid Application
228 <https://hal.inria.fr/inria-00071989/document>`_.
229 - **ns-3** (only available if you compiled SimGrid accordingly):
230 Use the packet-level network
231 simulators as network models (see :ref:`models_ns3`).
232 This model can be :ref:`further configured <options_pls>`.
234 - ``cpu/model``: specify the used CPU model. We have only one model
237 - **Cas01:** Simplistic CPU model (time=size/speed)
239 - ``host/model``: The host concept is the aggregation of a CPU with a
240 network card. Three models exists, but actually, only 2 of them are
241 interesting. The "compound" one is simply due to the way our
242 internal code is organized, and can easily be ignored. So at the
243 end, you have two host models: The default one allows aggregation of
244 an existing CPU model with an existing network model, but does not
245 allow parallel tasks because these beasts need some collaboration
246 between the network and CPU model.
248 - **default:** Default host model. Currently, CPU:Cas01 and
249 network:LV08 (with cross traffic enabled)
250 - **compound:** Host model that is automatically chosen if
251 you change the network and CPU models
252 - **ptask_L07:** Host model somehow similar to Cas01+CM02 but
253 allowing "parallel tasks", that are intended to model the moldable
254 tasks of the grid scheduling literature.
256 - ``storage/model``: specify the used storage model. Only one model is
258 - ``vm/model``: specify the model for virtual machines. Only one model
261 .. todo: make 'compound' the default host model.
263 .. _options_model_solver:
268 The different models rely on a linear inequalities solver to share
269 the underlying resources. SimGrid allows you to change the solver, but
270 be cautious, **don't change it unless you are 100% sure**.
272 - items ``cpu/solver``, ``network/solver``, ``disk/solver`` and ``host/solver``
273 allow you to change the solver for each model:
275 - **maxmin:** The default solver for all models except ptask. Provides a
276 max-min fairness allocation.
277 - **fairbottleneck:** The default solver for ptasks. Extends max-min to
278 allow heterogeneous resources.
279 - **bmf:** More realistic solver for heterogeneous resource sharing.
280 Implements BMF (Bottleneck max fairness) fairness. To be used with
281 parallel tasks instead of fair-bottleneck.
283 .. _options_model_optim:
288 The network and CPU models that are based on linear inequalities solver (that
289 is, all our analytical models) accept specific optimization
292 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
294 - **Lazy:** Lazy action management (partial invalidation in lmm +
295 heap in action remaining).
296 - **TI:** Trace integration. Highly optimized mode when using
297 availability traces (only available for the Cas01 CPU model for
299 - **Full:** Full update of remaining and variables. Slow but may be
300 useful when debugging.
302 - items ``network/maxmin-selective-update`` and
303 ``cpu/maxmin-selective-update``: configure whether the underlying
304 should be lazily updated or not. It should have no impact on the
305 computed timings, but should speed up the computation. |br| It is
306 still possible to disable this feature because it can reveal
307 counter-productive in very specific scenarios where the
308 interaction level is high. In particular, if all your
309 communication share a given backbone link, you should disable it:
310 without it, a simple regular loop is used to update each
311 communication. With it, each of them is still updated (because of
312 the dependency induced by the backbone), but through a complicated
313 and slow pattern that follows the actual dependencies.
315 .. _cfg=bmf/precision:
316 .. _cfg=maxmin/precision:
317 .. _cfg=surf/precision:
322 **Option** ``maxmin/precision`` **Default:** 1e-5 (in flops or bytes) |br|
323 **Option** ``surf/precision`` **Default:** 1e-9 (in seconds) |br|
324 **Option** ``bmf/precision`` **Default:** 1e-12 (no unit)
326 The analytical models handle a lot of floating point values. It is
327 possible to change the epsilon used to update and compare them through
328 this configuration item. Changing it may speedup the simulation by
329 discarding very small actions, at the price of a reduced numerical
330 precision. You can modify separately the precision used to manipulate
331 timings (in seconds) and the one used to manipulate amounts of work
334 .. _cfg=maxmin/concurrency-limit:
339 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
341 The maximum number of variables per resource can be tuned through this
342 option. You can have as many simultaneous actions per resources as you
343 want. If your simulation presents a very high level of concurrency, it
344 may help to use e.g. 100 as a value here. It means that at most 100
345 actions can consume a resource at a given time. The extraneous actions
346 are queued and wait until the amount of concurrency of the considered
347 resource lowers under the given boundary.
349 Such limitations help both to the simulation speed and simulation accuracy
350 on highly constrained scenarios, but the simulation speed suffers of this
351 setting on regular (less constrained) scenarios so it is off by default.
353 .. _cfg=bmf/max-iterations:
358 **Option** ``bmf/max-iterations`` **Default:** 1000
360 It may happen in some settings that the BMF solver fails to converge to
361 a solution, so there is a hard limit on the amount of iteration count to
362 avoid infinite loops.
364 .. _options_model_network:
366 Configuring the Network Model
367 .............................
369 .. _cfg=network/TCP-gamma:
371 Maximal TCP Window Size
372 ^^^^^^^^^^^^^^^^^^^^^^^
374 **Option** ``network/TCP-gamma`` **Default:** 4194304
376 The analytical models need to know the maximal TCP window size to take the TCP congestion mechanism into account (see
377 :ref:`this page <understanding_cm02>` for details). On Linux, this value can be retrieved using the following commands.
378 Both give a set of values, and you should use the last one, which is the maximal size.
380 .. code-block:: console
382 $ cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
383 $ cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
385 If you want to disable the TCP windowing mechanism, set this parameter to 0.
387 .. _cfg=network/bandwidth-factor:
388 .. _cfg=network/latency-factor:
389 .. _cfg=network/weight-S:
391 Manual calibration factors
392 ^^^^^^^^^^^^^^^^^^^^^^^^^^
394 SimGrid can take network irregularities such as a slow startup or changing behavior depending on the message size into account.
395 The values provided by default were computed a long time ago through data fitting one the timings of either packet-level
396 simulators or direct experiments on real platforms. These default values should be OK for most users, but if simulation realism
397 is really important to you, you probably want to recalibrate the models (i.e., devise sensible values for your specific
398 settings). This section only describes how to pass new values to the models while the calibration process involved in the
399 computation of these values is described :ref:`in the relevant chapter <models_calibration>`.
401 We found out that many networking effects can be realistically accounted for with the three following correction factors. They
402 were shown to be enough to capture slow-start effects, the different transmission modes of MPI systems (eager vs. rendez-vous
403 mode), or the non linear effects of wifi sharing.
405 **Option** ``network/latency-factor`` **Default:** 1.0, but overridden by most models
407 This option specifies a multiplier to apply to the *physical* latency (i.e., the one described in the platform) of the set of
408 links involved in a communication. The factor can either be a constant to apply to any communication, or it may depend on the
409 message size. The ``CM02`` model does not use any correction factor, so the latency-factor remains to 1. The ``LV08`` model sets
410 it to 13.01 to model slow-start, while the ``SMPI`` model has several possible values depending on the interval in which the
411 message size falls. The default SMPI setting given below specifies for example that a message smaller than 257 bytes will get a
412 latency multiplier of 2.01467 while a message whose size is in [15424, 65472] will get a latency multiplier of 3.48845. The
413 ``wifi`` model goes further and uses a callback in the program to compute the factor that must be non-linear in this case.
415 This multiplier is applied to the latency computed from the platform, that is the sum of all link *physical* latencies over the
416 :ref:`network path <platform_routing>` used by the considered communication, to derive the *effective* end-to-end latency.
418 Constant factors are easy to express, but the interval-based syntax used in SMPI is somewhat complex. It expects a set of
419 factors separated by semicolons, each of the form ``boundary:factor``. For example if your specification is
420 ``0:1;1000:2;5000:3``, it means that on [0, 1000) the factor is 1. On [1000,5000), the factor is 2 while the factor is 3 for
421 5000 and beyond. If your first interval does include size=0, then the default value of 1 is used before. Changing the factor
422 callback is not possible from the command line and must be done from your code, as shown in `this example
423 <https://framagit.org/simgrid/simgrid/tree/master/examples/cpp/network-factors/s4u-network-factors.cpp>`_. Note that the chosen
424 model only provides some default settings. You may pick a ``LV08`` model to get some of the settings, and override the latency
425 with interval-based values.
427 SMPI default value: 65472:11.6436; 15424:3.48845; 9376:2.59299; 5776:2.18796; 3484:1.88101; 1426:1.61075; 732:1.9503;
428 257:1.95341;0:2.01467 (interval boundaries are sorted automatically). These values were computed by data fitting on the Stampede
429 Supercomputer at TACC, with optimal deployment of processes on nodes. To accurately model your settings, you should redo the
430 :ref:`calibration <models_calibration>`.
432 **Option** ``network/bandwidth-factor`` **Default:** 1.0, but overridden by most models
434 Setting this option automatically adjusts the *effective* bandwidth (i.e., the one perceived by the application) used by any
435 given communication. As with latency-factor above, the value can be a constant (``CM02`` uses 1 -- no correction -- while
436 ``LV08`` uses 0.97 to discount TCP headers while computing the payload bandwidth), interval-based (as the default provided by
437 the ``SMPI``), or using in-program callbacks (as with ``wifi``).
439 SMPI default value: 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
440 This was also computed on the Stampede Supercomputer.
442 **Option** ``network/weight-S`` **Default:** depends on the model
444 Value used to account for RTT-unfairness when sharing a bottleneck (network connections with a large RTT are generally penalized
445 against those with a small one). Described in `Accuracy Study and Improvement of Network Simulation in the SimGrid Framework
446 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_
448 Default values for ``CM02`` is 0. ``LV08`` sets it to 20537 while both ``SMPI`` and ``IB`` set it to 8775.
450 .. _cfg=network/loopback:
452 Configuring loopback link
453 ^^^^^^^^^^^^^^^^^^^^^^^^^
455 Several network models provide an implicit loopback link to account for local
456 communication on a host. By default it has a 10GBps bandwidth and a null latency.
457 This can be changed with ``network/loopback-lat`` and ``network/loopback-bw``
458 items. Note that this loopback is conveniently modeled with a :ref:`single FATPIPE link <pf_loopback>`
459 for the whole platform. If modeling contention inside nodes is important then you should
460 rather add such loopback links (one for each host) yourself.
462 .. _cfg=smpi/IB-penalty-factors:
467 InfiniBand network behavior can be modeled through 3 parameters
468 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `the PhD
469 thesis of Jean-Marc Vincent
470 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_ (in French)
471 or more concisely in `this paper <https://hal.inria.fr/hal-00953618/document>`_,
472 even if that paper does only describe models for myrinet and ethernet.
473 You can see in Fig 2 some results for Infiniband, for example. This model
474 may be outdated by now for modern infiniband, anyway, so a new
475 validation would be good.
477 The three paramaters are defined as follows:
479 - βs: penalty factor for outgoing messages, computed by running a simple send to
480 two nodes and checking slowdown compared to a single send to one node,
482 - βe: penalty factor for ingoing messages, same computation method but with one
483 node receiving several messages
484 - γr: slowdown factor when communication buffer memory is saturated. It needs a
485 more complicated pattern to run in order to be computed (5.3 in the thesis,
486 page 107), and formula in the end is γr = time(c)/(3×βe×time(ref)), where
487 time(ref) is the time of a single comm with no contention).
489 Once these values are computed, a penalty is assessed for each message (this is
490 the part implemented in the simulator) as shown page 106 of the thesis. Here is
491 a simple translation of this text. First, some notations:
493 - ∆e(e) which corresponds to the incoming degree of node e, that is to say the number of communications having as destination node e.
494 - ∆s (s) which corresponds to the degree outgoing from node s, that is to say the number of communications sent by node s.
495 - Φ (e) which corresponds to the number of communications destined for the node e but coming from a different node.
496 - 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
498 To determine the penalty for a communication, two values need to be calculated. First, the penalty caused by the conflict in transmission, noted ps.
501 - if ∆s (i) = 1 then ps = 1.
502 - if ∆s (i) ≥ 2 and ∆e (i) ≥ 3 then ps = ∆s (i) × βs × γr
503 - else, ps = ∆s (i) × βs
506 Then, the penalty caused by the conflict in reception (noted pe) should be computed as follows:
508 - if ∆e (i) = 1 then pe = 1
509 - else, pe = Φ (e) × βe × Ω (s, e)
511 Finally, the penalty associated with the communication is:
512 p = max (ps ∈ s, pe)
514 .. _cfg=network/crosstraffic:
516 Simulating Cross-Traffic
517 ^^^^^^^^^^^^^^^^^^^^^^^^
519 Since SimGrid v3.7, cross-traffic effects can be taken into account in
520 analytical simulations. It means that ongoing and incoming
521 communication flows are treated independently. In addition, the LV08
522 model adds 0.05 of usage on the opposite direction for each new
523 created flow. This can be useful to simulate some important TCP
524 phenomena such as ack compression.
526 For that to work, your platform must have two links for each
527 pair of interconnected hosts. An example of usable platform is
528 available in ``examples/platforms/crosstraffic.xml``.
530 This is activated through the ``network/crosstraffic`` item, that
531 can be set to 0 (disable this feature) or 1 (enable it).
533 Note that with the default host model this option is activated by default.
535 .. _cfg=smpi/async-small-thresh:
537 Simulating Asynchronous Send
538 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
540 (this configuration item is experimental and may change or disappear)
542 It is possible to specify that messages below a certain size (in bytes) will be
543 sent as soon as the call to MPI_Send is issued, without waiting for
544 the correspondent receive. This threshold can be configured through
545 the ``smpi/async-small-thresh`` item. The default value is 0. This
546 behavior can also be manually set for mailboxes, by setting the
547 receiving mode of the mailbox with a call to
548 :cpp:func:`sg_mailbox_set_receiver`. After this, all messages sent to
549 this mailbox will have this behavior regardless of the message size.
551 This value needs to be smaller than or equals to the threshold set at
552 :ref:`cfg=smpi/send-is-detached-thresh`, because asynchronous messages
553 are meant to be detached as well.
560 **Option** ``ns3/TcpModel`` **Default:** "default" (ns-3 default)
562 When using ns-3, there is an extra item ``ns3/TcpModel``, corresponding
563 to the ``ns3::TcpL4Protocol::SocketType`` configuration item in
564 ns-3. The only valid values (enforced on the SimGrid side) are
565 'default' (no change to the ns-3 configuration), 'NewReno' or 'Reno' or
568 **Option** ``ns3/seed`` **Default:** "" (don't set the seed in ns-3)
570 This option is the random seed to provide to ns-3 with
571 ``ns3::RngSeedManager::SetSeed`` and ``ns3::RngSeedManager::SetRun``.
573 If left blank, no seed is set in ns-3. If the value 'time' is
574 provided, the current amount of seconds since epoch is used as a seed.
575 Otherwise, the provided value must be a number to use as a seed.
577 Configuring the Storage model
578 .............................
580 .. _cfg=storage/max_file_descriptors:
582 File Descriptor Count per Host
583 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
585 **Option** ``storage/max_file_descriptors`` **Default:** 1024
587 Each host maintains a fixed-size array of its file descriptors. You
588 can change its size through this item to either enlarge it if your
589 application requires it or to reduce it to save memory space.
596 SimGrid plugins allow one to extend the framework without changing its
597 source code directly. Read the source code of the existing plugins to
598 learn how to do so (in ``src/plugins``), and ask your questions to the
599 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
600 that plugins usually register callbacks to some signals of interest.
601 If they need to store some information about a given object (Link, CPU
602 or Actor), they do so through the use of a dedicated object extension.
604 Some of the existing plugins can be activated from the command line,
605 meaning that you can activate them from the command line without any
606 modification to your simulation code. For example, you can activate
607 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
610 Here is a partial list of plugins that can be activated this way. You can get
611 the full list by passing ``--cfg=plugin:help`` to your simulator.
613 - :ref:`Host Energy <plugin_host_energy>`: models the energy dissipation of the compute units.
614 - :ref:`Link Energy <plugin_link_energy>`: models the energy dissipation of the network.
615 - :ref:`Host Load <plugin_host_load>`: monitors the load of the compute units.
617 .. _options_modelchecking:
619 Configuring the Model-Checking
620 ------------------------------
622 To enable SimGrid's model-checking support, the program should
623 be executed using the simgrid-mc wrapper:
625 .. code-block:: console
627 $ simgrid-mc ./my_program
629 Safety properties are expressed as assertions using the function
630 :cpp:func:`void MC_assert(int prop)`.
632 .. _cfg=smpi/buffering:
634 Specifying the MPI buffering behavior
635 .....................................
637 **Option** ``smpi/buffering`` **Default:** infty
639 Buffering in MPI has a huge impact on the communication semantic. For example,
640 standard blocking sends are synchronous calls when the system buffers are full
641 while these calls can complete immediately without even requiring a matching
642 receive call for small messages sent when the system buffers are empty.
644 In SMPI, this depends on the message size, that is compared against two thresholds:
646 - if (size < :ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>`) then
647 MPI_Send returns immediately, even if the corresponding receive has not be issued yet.
648 - 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
649 MPI_Send returns as soon as the corresponding receive has been issued. This is known as the eager mode.
650 - if (:ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>` < size) then
651 MPI_Send returns only when the message has actually been sent over the network. This is known as the rendez-vous mode.
653 The ``smpi/buffering`` (only valid with MC) option gives an easier interface to choose between these semantics. It can take two values:
655 - **zero:** means that buffering should be disabled. All communications are actually blocking.
656 - **infty:** means that buffering should be made infinite. All communications are non-blocking.
658 .. _cfg=model-check/property:
660 Specifying a liveness property
661 ..............................
663 **Option** ``model-check/property`` **Default:** unset
665 If you want to specify liveness properties, you have to pass them on
666 the command line, specifying the name of the file containing the
667 property, as formatted by the `ltl2ba <https://github.com/utwente-fmt/ltl2ba>`_ program.
668 Note that ltl2ba is not part of SimGrid and must be installed separately.
670 .. code-block:: console
672 $ simgrid-mc ./my_program --cfg=model-check/property:<filename>
674 .. _cfg=model-check/checkpoint:
676 Going for Stateful Verification
677 ...............................
679 By default, the system is backtracked to its initial state to explore
680 another path, instead of backtracking to the exact step before the fork
681 that we want to explore (this is called stateless verification). This
682 is done this way because saving intermediate states can rapidly
683 exhaust the available memory. If you want, you can change the value of
684 the ``model-check/checkpoint`` item. For example,
685 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
686 step. Beware, this will certainly explode your memory. Larger values
687 are probably better, make sure to experiment a bit to find the right
688 setting for your specific system.
690 .. _cfg=model-check/reduction:
692 Specifying the kind of reduction
693 ................................
695 The main issue when using the model-checking is the state space
696 explosion. You can activate some reduction technique with
697 ``--cfg=model-check/reduction:<technique>``. For now, this
698 configuration variable can take 2 values:
700 - **none:** Do not apply any kind of reduction (mandatory for
701 liveness properties, as our current DPOR algorithm breaks cycles)
702 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if
703 you verify local safety properties (default value for safety
706 Another way to mitigate the state space explosion is to search for
707 cycles in the exploration with the :ref:`cfg=model-check/visited`
708 configuration. Note that DPOR and state-equality reduction may not
709 play well together. You should choose between them.
711 Our current DPOR implementation could be improved in may ways. We are
712 currently improving its efficiency (both in term of reduction ability
713 and computational speed), and future work could make it compatible
714 with liveness properties.
716 .. _cfg=model-check/visited:
718 Size of Cycle Detection Set (state equality reduction)
719 ......................................................
721 Mc SimGrid can be asked to search for cycles during the exploration,
722 i.e. situations where a new explored state is in fact the same state
723 than a previous one.. This can prove useful to mitigate the state
724 space explosion with safety properties, and this is the crux when
725 searching for counter-examples to the liveness properties.
727 Note that this feature may break the current implementation of the
728 DPOR reduction technique.
730 The ``model-check/visited`` item is the maximum number of states, which
731 are stored in memory. If the maximum number of snapshotted state is
732 reached, some states will be removed from the memory and some cycles
733 might be missed. Small values can lead to incorrect verifications, but
734 large values can exhaust your memory and be CPU intensive as each new
735 state must be compared to that amount of older saved states.
737 The default settings depend on the kind of exploration. With safety
738 checking, no state is snapshotted and cycles cannot be detected. With
739 liveness checking, all states are snapshotted because missing a cycle
740 could hinder the exploration soundness.
742 .. _cfg=model-check/termination:
744 Non-Termination Detection
745 .........................
747 The ``model-check/termination`` configuration item can be used to
748 report if a non-termination execution path has been found. This is a
749 path with a cycle, which means that the program might never terminate.
751 This only works in safety mode, not in liveness mode.
753 This options is disabled by default.
755 .. _cfg=model-check/dot-output:
760 If set, the ``model-check/dot-output`` configuration item is the name
761 of a file in which to write a dot file of the path leading to the
762 property violation discovered (safety or liveness violation), as well
763 as the cycle for liveness properties. This dot file can then be fed to the
764 graphviz dot tool to generate a corresponding graphical representation.
766 .. _cfg=model-check/max-depth:
768 Exploration Depth Limit
769 .......................
771 The ``model-check/max-depth`` can set the maximum depth of the
772 exploration graph of the model checker. If this limit is reached, a
773 logging message is sent and the results might not be exact.
775 By default, the exploration is limited to the depth of 1000.
777 .. _cfg=model-check/timeout:
782 By default, the model checker does not handle timeout conditions: the `wait`
783 operations never time out. With the ``model-check/timeout`` configuration item
784 set to **yes**, the model checker will explore timeouts of `wait` operations.
786 .. _cfg=model-check/communications-determinism:
787 .. _cfg=model-check/send-determinism:
789 Communication Determinism
790 .........................
792 The ``model-check/communications-determinism`` and
793 ``model-check/send-determinism`` items can be used to select the
794 communication determinism mode of the model checker, which checks
795 determinism properties of the communications of an application.
797 .. _cfg=model-check/setenv:
799 Passing environment variables
800 .............................
802 You can specify extra environment variables to be set in the verified application
803 with ``model-check/setenv``. For example, you can preload a library as follows:
804 ``-cfg=model-check/setenv:LD_PRELOAD=toto;LD_LIBRARY_PATH=/tmp``.
808 Verification Performance Considerations
809 .......................................
811 The size of the stacks can have a huge impact on the memory
812 consumption when using model-checking. By default, each snapshot will
813 save a copy of the whole stacks and not only of the part that is
814 really meaningful: you should expect the contribution of the memory
815 consumption of the snapshots to be:
816 :math:`\text{number of processes} \times \text{stack size} \times \text{number of states}`.
818 When compiled against the model checker, the stacks are not
819 protected with guards: if the stack size is too small for your
820 application, the stack will silently overflow into other parts of the
821 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
823 .. _cfg=model-check/replay:
825 Replaying buggy execution paths from the model checker
826 ......................................................
828 Debugging the problems reported by the model checker is challenging:
829 First, the application under verification cannot be debugged with gdb
830 because the model checker already traces it. Then, the model checker may
831 explore several execution paths before encountering the issue, making it
832 very difficult to understand the output. Fortunately, SimGrid provides
833 the execution path leading to any reported issue so that you can replay
834 this path reported by the model checker, enabling the usage of classical
837 When the model checker finds an interesting path in the application
838 execution graph (where a safety or liveness property is violated), it
839 generates an identifier for this path. Here is an example of the output:
841 .. code-block:: console
843 [ 0.000000] (0:@) Check a safety property
844 [ 0.000000] (0:@) **************************
845 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
846 [ 0.000000] (0:@) **************************
847 [ 0.000000] (0:@) Counter-example execution trace:
848 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
849 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
850 [ 0.000000] (0:@) Path = 1/3;1/4
851 [ 0.000000] (0:@) Expanded states = 27
852 [ 0.000000] (0:@) Visited states = 68
853 [ 0.000000] (0:@) Executed transitions = 46
855 The interesting line is ``Path = 1/3;1/4``, which means that you should use
856 ``--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
857 execution path. All options (but the model checker related ones) must
858 remain the same. In particular, if you ran your application with
859 ``smpirun -wrapper simgrid-mc``, then do it again. Remove all
860 MC-related options, keep non-MC-related ones and add
861 ``--cfg=model-check/replay:???``.
863 Currently, if the path is of the form ``X;Y;Z``, each number denotes
864 the actor's pid that is selected at each indecision point. If it's of
865 the form ``X/a;Y/b``, the X and Y are the selected pids while the a
866 and b are the return values of their simcalls. In the previous
867 example, ``1/3;1/4``, you can see from the full output that the actor
868 1 is doing MC_RANDOM simcalls, so the 3 and 4 simply denote the values
869 that these simcall return on the execution branch leading to the
872 Configuring the User Code Virtualization
873 ----------------------------------------
875 .. _cfg=contexts/factory:
877 Selecting the Virtualization Factory
878 ....................................
880 **Option** contexts/factory **Default:** "raw"
882 In SimGrid, the user code is virtualized in a specific mechanism that
883 allows the simulation kernel to control its execution: when a user
884 process requires a blocking action (such as sending a message), it is
885 interrupted, and only gets released when the simulated clock reaches
886 the point where the blocking operation is done. This is explained
887 graphically in the `relevant tutorial, available online
888 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
890 In SimGrid, the containers in which user processes are virtualized are
891 called contexts. Several context factory are provided, and you can
892 select the one you want to use with the ``contexts/factory``
893 configuration item. Some of the following may not exist on your
894 machine because of portability issues. In any case, the default one
895 should be the most effcient one (please report bugs if the
896 auto-detection fails for you). They are approximately sorted here from
897 the slowest to the most efficient:
899 - **thread:** very slow factory using full featured, standard threads.
900 They are slow but very standard. Some debuggers or profilers only work with this factory.
901 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
902 - **boost:** This uses the `context
903 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
904 of the boost library for a performance that is comparable to our
906 |br| Install the relevant library (e.g. with the
907 libboost-contexts-dev package on Debian/Ubuntu) and recompile
909 - **raw:** amazingly fast factory using a context switching mechanism
910 of our own, directly implemented in assembly (only available for x86
911 and amd64 platforms for now) and without any unneeded system call.
913 The main reason to change this setting is when the debugging tools become
914 fooled by the optimized context factories. Threads are the most
915 debugging-friendly contexts, as they allow one to set breakpoints
916 anywhere with gdb and visualize backtraces for all processes, in order
917 to debug concurrency issues. Valgrind is also more comfortable with
918 threads, but it should be usable with all factories (Exception: the
919 callgrind tool really dislikes raw and ucontext factories).
921 .. _cfg=contexts/stack-size:
923 Adapting the Stack Size
924 .......................
926 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
928 Each virtualized used process is executed using a specific system
929 stack. The size of this stack has a huge impact on the simulation
930 scalability, but its default value is rather large. This is because
931 the error messages that you get when the stack size is too small are
932 rather disturbing: this leads to stack overflow (overwriting other
933 stacks), leading to segfaults with corrupted stack traces.
935 If you want to push the scalability limits of your code, you might
936 want to reduce the ``contexts/stack-size`` item. Its default value is
937 8192 (in KiB), while our Chord simulation works with stacks as small
938 as 16 KiB, for example. You can ensure that some actors have a specific
939 size by simply changing the value of this configuration item before
940 creating these actors. The :cpp:func:`simgrid::s4u::Engine::set_config`
941 functions are handy for that.
943 This *setting is ignored* when using the thread factory (because there
944 is no way to modify the stack size with C++ system threads). Instead,
945 you should compile SimGrid and your application with
946 ``-fsplit-stack``. Note that this compilation flag is not compatible
947 with the model checker right now.
949 The operating system should only allocate memory for the pages of the
950 stack which are actually used and you might not need to use this in
951 most cases. However, this setting is very important when using the
952 model checker (see :ref:`options_mc_perf`).
954 .. _cfg=contexts/guard-size:
956 Disabling Stack Guard Pages
957 ...........................
959 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages with MC)
961 Unless you use the threads context factory (see
962 :ref:`cfg=contexts/factory`), a stack guard page is usually used
963 which prevents the stack of a given actor from overflowing on another
964 stack. But the performance impact may become prohibitive when the
965 amount of actors increases. The option ``contexts/guard-size`` is the
966 number of stack guard pages used. By setting it to 0, no guard pages
967 will be used: in this case, you should avoid using small stacks (with
968 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
969 will silently overflow on other parts of the memory.
971 When no stack guard page is created, stacks may then silently overflow
972 on other parts of the memory if their size is too small for the
975 .. _cfg=contexts/nthreads:
976 .. _cfg=contexts/synchro:
978 Running User Code in Parallel
979 .............................
981 Parallel execution of the user code is only considered stable in
982 SimGrid v3.7 and higher, and mostly for S4U simulations. SMPI
983 simulations may well fail in parallel mode. It is described in
984 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
986 If you are using the **ucontext** or **raw** context factories, you can
987 request to execute the user code in parallel. Several threads are
988 launched, each of them handling the same number of user contexts at each
989 run. To activate this, set the ``contexts/nthreads`` item to the amount
990 of cores that you have in your computer (or lower than 1 to have the
991 amount of cores auto-detected).
993 When parallel execution is activated, you can choose the
994 synchronization schema used with the ``contexts/synchro`` item,
995 which value is either:
997 - **futex:** ultra optimized synchronisation schema, based on futexes
998 (fast user-mode mutexes), and thus only available on Linux systems.
999 This is the default mode when available.
1000 - **posix:** slow but portable synchronisation using only POSIX
1002 - **busy_wait:** not really a synchronisation: the worker threads
1003 constantly request new contexts to execute. It should be the most
1004 efficient synchronisation schema, but it loads all the cores of
1005 your machine for no good reason. You probably prefer the other less
1008 Configuring the Tracing
1009 -----------------------
1011 The :ref:`tracing subsystem <outcome_vizu>` can be configured in
1012 several different ways depending on the used interface (S4U, SMPI)
1013 and the kind of traces that needs to be obtained. See the
1014 :ref:`Tracing Configuration Options subsection
1015 <tracing_tracing_options>` for a full description of each
1016 configuration option.
1018 We detail here a simple way to get the traces working for you, even if
1019 you never used the tracing API.
1022 - Any SimGrid-based simulator (S4U, SMPI, ...) and raw traces:
1024 .. code-block:: none
1026 --cfg=tracing:yes --cfg=tracing/uncategorized:yes
1028 The first parameter activates the tracing subsystem, and the second
1029 tells it to trace host and link utilization (without any
1032 - S4U-based simulator and categorized traces (you need to
1033 declare categories and classify your tasks according to them)
1035 .. code-block:: none
1037 --cfg=tracing:yes --cfg=tracing/categorized:yes
1039 The first parameter activates the tracing subsystem, and the second
1040 tells it to trace host and link categorized utilization.
1042 - SMPI simulator and traces for a space/time view:
1044 .. code-block:: console
1046 $ smpirun -trace ...
1048 The `-trace` parameter for the smpirun script runs the simulation
1049 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
1050 smpirun's `-help` parameter for additional tracing options.
1052 Sometimes you might want to put additional information on the trace to
1053 correctly identify them later, or to provide data that can be used to
1054 reproduce an experiment. You have two ways to do that:
1056 - Add a string on top of the trace file as comment:
1058 .. code-block:: none
1060 --cfg=tracing/comment:my_simulation_identifier
1062 - Add the contents of a textual file on top of the trace file as comment:
1064 .. code-block:: none
1066 --cfg=tracing/comment-file:my_file_with_additional_information.txt
1068 Please, use these two parameters (for comments) to make reproducible
1069 simulations. For additional details about this and all tracing
1070 options, check See the :ref:`tracing_tracing_options`.
1075 The SMPI interface provides several specific configuration items.
1076 These are not easy to see with ``--help-cfg``, since SMPI binaries are usually launched through the ``smiprun`` script.
1078 .. _cfg=smpi/host-speed:
1079 .. _cfg=smpi/cpu-threshold:
1080 .. _cfg=smpi/simulate-computation:
1082 Automatic Benchmarking of SMPI Code
1083 ...................................
1085 In SMPI, the sequential code is automatically benchmarked, and these
1086 computations are automatically reported to the simulator. That is to
1087 say that if you have a large computation between a ``MPI_Recv()`` and
1088 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
1089 this code, and create an execution task within the simulator to take
1090 this into account. For that, the actual duration is measured on the
1091 host machine and then scaled to the power of the corresponding
1092 simulated machine. The variable ``smpi/host-speed`` allows one to
1093 specify the computational speed of the host machine (in flop/s by
1094 default) to use when scaling the execution times.
1096 The default value is ``smpi/host-speed=20kf`` (= 20,000 flop/s). This
1097 is probably underestimated for most machines, leading SimGrid to
1098 overestimate the amount of flops in the execution blocks that are
1099 automatically injected in the simulator. As a result, the execution
1100 time of the whole application will probably be overestimated until you
1101 use a realistic value.
1103 When the code consists of numerous consecutive MPI calls, the
1104 previous mechanism feeds the simulation kernel with numerous tiny
1105 computations. The ``smpi/cpu-threshold`` item becomes handy when this
1106 impacts badly on the simulation performance. It specifies a threshold (in
1107 seconds) below which the execution chunks are not reported to the
1108 simulation kernel (default value: 1e-6).
1110 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
1111 time spent below this threshold. SMPI does not consider the
1112 `amount of time` of these computations; there is no offset for
1113 this. Hence, a value that is too small, may lead to unreliable
1116 In some cases, however, one may wish to disable simulation of
1117 the computation of an application. This is the case when SMPI is used not to
1118 simulate an MPI application, but instead an MPI code that performs
1119 "live replay" of another MPI app (e.g., ScalaTrace's replay tool, or
1120 various on-line simulators that run an app at scale). In this case the
1121 computation of the replay/simulation logic should not be simulated by
1122 SMPI. Instead, the replay tool or on-line simulator will issue
1123 "computation events", which correspond to the actual MPI simulation
1124 being replayed/simulated. At the moment, these computation events can
1125 be simulated using SMPI by calling internal smpi_execute*() functions.
1127 To disable the benchmarking/simulation of a computation in the simulated
1128 application, the variable ``smpi/simulate-computation`` should be set
1129 to **no**. This option just ignores the timings in your simulation; it
1130 still executes the computations itself. If you want to stop SMPI from
1131 doing that, you should check the SMPI_SAMPLE macros, documented in
1132 Section :ref:`SMPI_use_faster`.
1134 +------------------------------------+-------------------------+-----------------------------+
1135 | Solution | Computations executed? | Computations simulated? |
1136 +====================================+=========================+=============================+
1137 | --cfg=smpi/simulate-computation:no | Yes | Never |
1138 +------------------------------------+-------------------------+-----------------------------+
1139 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
1140 +------------------------------------+-------------------------+-----------------------------+
1141 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
1142 +------------------------------------+-------------------------+-----------------------------+
1144 .. _cfg=smpi/comp-adjustment-file:
1146 Slow-down or speed-up parts of your code
1147 ........................................
1149 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
1151 This option allows you to pass a file that contains two columns: The
1152 first column defines the section that will be subject to a speedup;
1153 the second column is the speedup. For instance:
1155 .. code-block:: none
1157 "start:stop","ratio"
1158 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1160 The first line is the header - you must include it. The following
1161 line means that the code between two consecutive MPI calls on line 30
1162 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1163 of 1.18244559422142. The value for the second column is therefore a
1164 speedup, if it is larger than 1 and a slowdown if it is smaller
1165 than 1. Nothing will be changed if it is equal to 1.
1167 Of course, you can set any arbitrary filenames you want (so the start
1168 and end don't have to be in the same file), but be aware that this
1169 mechanism only supports `consecutive calls!`
1171 Please note that you must pass the ``-trace-call-location`` flag to
1172 smpicc or smpiff, respectively. This flag activates some internal
1173 macro definitions that help with obtaining the call location.
1175 Bandwidth and latency factors
1176 .............................
1178 Adapting the bandwidth and latency acurately to the network conditions is of a paramount importance to get realistic results.
1179 This is done through the :ref:`network/bandwidth-factor <cfg=network/bandwidth-factor>` and :ref:`network/latency-factor
1180 <cfg=network/latency-factor>` items. You probably also want to read the following section: :ref:`models_calibration`.
1182 .. _cfg=smpi/display-timing:
1184 Reporting Simulation Time
1185 .........................
1187 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1189 Most of the time, you run MPI code with SMPI to compute the time it
1190 would take to run it on a platform. But since the code is run through
1191 the ``smpirun`` script, you don't have any control on the launcher
1192 code, making it difficult to report the simulated time when the
1193 simulation ends. If you enable the ``smpi/display-timing`` item,
1194 ``smpirun`` will display this information when the simulation
1196 SMPI will also display information about the amout of real time spent
1197 in application code and in SMPI internals, to provide hints about the
1198 need to use sampling to reduce simulation time.
1200 .. _cfg=smpi/display-allocs:
1202 Reporting memory allocations
1203 ............................
1205 **Option** ``smpi/display-allocs`` **Default:** 0 (false)
1207 SMPI intercepts malloc and calloc calls performed inside the running
1208 application, if it wasn't compiled with SMPI_NO_OVERRIDE_MALLOC.
1209 With this option, SMPI will show at the end of execution the amount of
1210 memory allocated through these calls, and locate the most expensive one.
1211 This helps finding the targets for manual memory sharing, or the threshold
1212 to use for smpi/auto-shared-malloc-thresh option (see :ref:`cfg=smpi/auto-shared-malloc-thresh`).
1214 .. _cfg=smpi/keep-temps:
1216 Keeping temporary files after simulation
1217 ........................................
1219 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1221 SMPI usually generates a lot of temporary files that are cleaned after
1222 use. This option requests to preserve them, for example to debug or
1223 profile your code. Indeed, the binary files are removed very early
1224 under the dlopen privatization schema, which tends to fool the
1227 .. _cfg=smpi/papi-events:
1229 Trace hardware counters with PAPI
1230 .................................
1232 **Option** ``smpi/papi-events`` **default:** unset
1234 When the PAPI support is compiled into SimGrid, this option takes the
1235 names of PAPI counters and adds their respective values to the trace
1236 files (See Section :ref:`tracing_tracing_options`).
1240 This feature currently requires superuser privileges, as registers
1241 are queried. Only use this feature with code you trust! Call
1242 smpirun for instance via ``smpirun -wrapper "sudo "
1243 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1244 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1245 will not be required.
1247 It is planned to make this feature available on a per-process (or per-thread?) basis.
1248 The first draft, however, just implements a "global" (i.e., for all processes) set
1249 of counters, the "default" set.
1251 .. code-block:: none
1253 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1255 .. _cfg=smpi/privatization:
1257 Automatic Privatization of Global Variables
1258 ...........................................
1260 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1262 MPI executables are usually meant to be executed in separate
1263 processes, but SMPI is executed in only one process. Global variables
1264 from executables will be placed in the same memory region and shared
1265 between processes, causing intricate bugs. Several options are
1266 possible to avoid this, as described in the main `SMPI publication
1267 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1268 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1269 automatically privatizing the globals, and this option allows one to
1270 choose between them.
1272 - **no** (default when not using smpirun): Do not automatically
1273 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1275 - **dlopen** or **yes** (default when using smpirun): Link multiple
1276 times against the binary.
1277 - **mmap** (slower, but maybe somewhat more stable):
1278 Runtime automatic switching of the data segments.
1281 This configuration option cannot be set in your platform file. You can only
1282 pass it as an argument to smpirun.
1284 .. _cfg=smpi/privatize-libs:
1286 Automatic privatization of global variables inside external libraries
1287 .....................................................................
1289 **Option** ``smpi/privatize-libs`` **default:** unset
1291 **Linux/BSD only:** When using dlopen (default) privatization,
1292 privatize specific shared libraries with internal global variables, if
1293 they can't be linked statically. For example libgfortran is usually
1294 used for Fortran I/O and indexes in files can be mixed up.
1296 Multiple libraries can be given, semicolon separated.
1298 This configuration option can only use either full paths to libraries,
1299 or full names. Check with ldd the name of the library you want to
1302 .. code-block:: console
1306 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1309 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1310 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1311 but not ``libgfortran`` nor ``libgfortran.so``.
1313 .. _cfg=smpi/send-is-detached-thresh:
1315 Simulating MPI detached send
1316 ............................
1318 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1320 This threshold specifies the size in bytes under which the send will
1321 return immediately. This is different from the threshold detailed in
1322 :ref:`cfg=smpi/async-small-thresh` because the message is not
1323 really sent when the send is posted. SMPI still waits for the
1324 corresponding receive to be posted, in order to perform the communication
1327 .. _cfg=smpi/coll-selector:
1329 Simulating MPI collective algorithms
1330 ....................................
1332 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1334 SMPI implements more than 100 different algorithms for MPI collective
1335 communication, to accurately simulate the behavior of most of the
1336 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1337 select the decision logic either of the OpenMPI or the MPICH libraries. (By
1338 default SMPI uses naive version of collective operations.)
1340 Each collective operation can be manually selected with a
1341 ``smpi/collective_name:algo_name``. Available algorithms are listed in
1342 :ref:`SMPI_use_colls`.
1344 .. TODO:: All available collective algorithms will be made available
1345 via the ``smpirun --help-coll`` command.
1347 .. _cfg=smpi/barrier-collectives:
1349 Add a barrier in all collectives
1350 ................................
1352 **Option** ``smpi/barrier-collectives`` **default:** off
1354 This option adds a simple barrier in some collective operations to catch dangerous
1355 code that may or may not work depending on the MPI implementation: Bcast, Exscan,
1356 Gather, Gatherv, Scan, Scatter, Scatterv and Reduce.
1358 For example, the following code works with OpenMPI while it deadlocks in MPICH and
1359 Intel MPI. Broadcast seem to be "fire and forget" in OpenMPI while other
1360 implementations expect to receive a message.
1365 MPI_Bcast(buf1, buff_size, MPI_CHAR, 0, newcom);
1366 MPI_Send(&buf2, buff_size, MPI_CHAR, 1, tag, newcom);
1367 } else if (rank==1) {
1368 MPI_Recv(&buf2, buff_size, MPI_CHAR, 0, tag, newcom, MPI_STATUS_IGNORE);
1369 MPI_Bcast(buf1, buff_size, MPI_CHAR, 0, newcom);
1372 The barrier is only simulated and does not involve any additional message (it is a S4U barrier).
1373 This option is disabled by default, and activated by the `-analyze` flag of smpirun.
1375 .. _cfg=smpi/barrier-finalization:
1377 Add a barrier in MPI_Finalize
1378 .............................
1380 **Option** ``smpi/finalization-barrier`` **default:** off
1382 By default, SMPI processes are destroyed as soon as soon as their code ends,
1383 so after a successful MPI_Finalize call returns. In some rare cases, some data
1384 might have been attached to MPI objects still active in the remaining processes,
1385 and can be destroyed eagerly by the finished process.
1386 If your code shows issues at finalization, such as segmentation fault, triggering
1387 this option will add an explicit MPI_Barrier(MPI_COMM_WORLD) call inside the
1388 MPI_Finalize, so that all processes will terminate at almost the same point.
1389 It might affect the total timing by the cost of a barrier.
1391 .. _cfg=smpi/errors-are-fatal:
1393 Disable MPI fatal errors
1394 ........................
1396 **Option** ``smpi/errors-are-fatal`` **default:** on
1398 By default, SMPI processes will crash if a MPI error code is returned. MPI allows
1399 to explicitely set MPI_ERRORS_RETURN errhandler to avoid this behaviour. This flag
1400 will turn on this behaviour by default (for all concerned types and errhandlers).
1401 This can ease debugging by going after the first reported error.
1403 .. _cfg=smpi/pedantic:
1405 Disable pedantic MPI errors
1406 ...........................
1408 **Option** ``smpi/pedantic`` **default:** on
1410 By default, SMPI will report all errors it finds in MPI codes. Some of these errors
1411 may not be considered as errors by all developers. This flag can be turned off to
1412 avoid reporting some usually harmless mistakes.
1413 Concerned errors list (will be expanded in the future):
1415 - Calling MPI_Win_fence only once in a program, hence just opening an epoch without
1418 .. _cfg=smpi/iprobe:
1420 Inject constant times for MPI_Iprobe
1421 ....................................
1423 **Option** ``smpi/iprobe`` **default:** 0.0001
1425 The behavior and motivation for this configuration option is identical
1426 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1429 .. _cfg=smpi/iprobe-cpu-usage:
1431 Reduce speed for iprobe calls
1432 .............................
1434 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1436 MPI_Iprobe calls can be heavily used in applications. To account
1437 correctly for the energy that cores spend probing, it is necessary to
1438 reduce the load that these calls cause inside SimGrid.
1440 For instance, we measured a maximum power consumption of 220 W for a
1441 particular application but only 180 W while this application was
1442 probing. Hence, the correct factor that should be passed to this
1443 option would be 180/220 = 0.81.
1447 Inject constant times for MPI_Init
1448 ..................................
1450 **Option** ``smpi/init`` **default:** 0
1452 The behavior and motivation for this configuration option is identical
1453 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1457 Inject constant times for MPI_Isend()
1458 .....................................
1460 **Option** ``smpi/ois``
1462 The behavior and motivation for this configuration option is identical
1463 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1467 Inject constant times for MPI_send()
1468 ....................................
1470 **Option** ``smpi/os``
1472 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1473 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1474 time). SMPI can factor these costs in as well, but the user has to
1475 configure SMPI accordingly as these values may vary by machine. This
1476 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1477 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1478 exactly as ``smpi/ois``.
1480 This item can consist of multiple sections; each section takes three
1481 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1482 so this example contains two sections. Furthermore, each section
1483 consists of three values.
1485 1. The first value denotes the minimum size in bytes for this section to take effect;
1486 read it as "if message size is greater than this value (and other section has a larger
1487 first value that is also smaller than the message size), use this".
1488 In the first section above, this value is "1".
1490 2. The second value is the startup time; this is a constant value that will always
1491 be charged, no matter what the size of the message. In the first section above,
1494 3. The third value is the `per-byte` cost. That is, it is charged for every
1495 byte of the message (incurring cost messageSize*cost_per_byte)
1496 and hence accounts also for larger messages. In the first
1497 section of the example above, this value is "2".
1499 Now, SMPI always checks which section it should use for a given
1500 message; that is, if a message of size 11 is sent with the
1501 configuration of the example above, only the second section will be
1502 used, not the first, as the first value of the second section is
1503 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1504 message of size 11 incurs the following cost inside MPI_Send:
1505 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1507 Note that the order of sections can be arbitrary; they will be ordered internally.
1511 Inject constant times for MPI_Recv()
1512 ....................................
1514 **Option** ``smpi/or``
1516 The behavior and motivation for this configuration option is identical
1517 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1520 .. _cfg=smpi/grow-injected-times:
1522 Inject constant times for MPI_Test
1523 ..................................
1525 **Option** ``smpi/test`` **default:** 0.0001
1527 By setting this option, you can control the amount of time a process
1528 sleeps when MPI_Test() is called; this is important, because SimGrid
1529 normally only advances the time while communication is happening and
1530 thus, MPI_Test will not add to the time, resulting in deadlock if it is
1531 used as a break-condition as in the following example:
1536 MPI_Test(request, flag, status);
1540 To speed up execution, we use a counter to keep track of how often we
1541 checked if the handle is now valid or not. Hence, we actually
1542 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1543 process to sleep increases linearly with the number of previously
1544 failed tests. This behavior can be disabled by setting
1545 ``smpi/grow-injected-times`` to **no**. This will also disable this
1546 behavior for MPI_Iprobe.
1548 .. _cfg=smpi/shared-malloc:
1549 .. _cfg=smpi/shared-malloc-hugepage:
1554 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1556 If your simulation consumes too much memory, you may want to modify
1557 your code so that the working areas are shared by all MPI ranks. For
1558 example, in a block-cyclic matrix multiplication, you will only
1559 allocate one set of blocks, and all processes will share them.
1560 Naturally, this will lead to very wrong results, but this will save a
1561 lot of memory. So this is still desirable for some studies. For more on
1562 the motivation for that feature, please refer to the `relevant section
1563 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1564 of the SMPI CourseWare (see Activity #2.2 of the pointed
1565 assignment). In practice, change the calls for malloc() and free() into
1566 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1568 SMPI provides two algorithms for this feature. The first one, called
1569 ``local``, allocates one block per call to SMPI_SHARED_MALLOC()
1570 (each call site gets its own block) ,and this block is shared
1571 among all MPI ranks. This is implemented with the shm_* functions
1572 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1573 for each shared block.
1575 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1576 returns a new address, but it only points to a shadow block: its memory
1577 area is mapped on a 1 MiB file on disk. If the returned block is of size
1578 N MiB, then the same file is mapped N times to cover the whole block.
1579 At the end, no matter how many times you call SMPI_SHARED_MALLOC, this will
1580 only consume 1 MiB in memory.
1582 You can disable this behavior and come back to regular mallocs (for
1583 example for debugging purposes) using ``no`` as a value.
1585 If you want to keep private some parts of the buffer, for instance if these
1586 parts are used by the application logic and should not be corrupted, you
1587 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). For example:
1591 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1593 This will allocate 500 bytes to mem, such that mem[27..41] and
1594 mem[100..199] are shared while other area remain private.
1596 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1598 When smpi/shared-malloc:global is used, the memory consumption problem
1599 is solved, but it may induce too much load on the kernel's pages table.
1600 In this case, you should use huge pages so that the kernel creates only one
1601 entry per MB of malloced data instead of one entry per 4 kB.
1602 To activate this, you must mount a hugetlbfs on your system and allocate
1603 at least one huge page:
1605 .. code-block:: console
1608 $ sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1609 $ sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1611 Then, you can pass the option
1612 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1613 actually activate the huge page support in shared mallocs.
1615 .. _cfg=smpi/auto-shared-malloc-thresh:
1617 Automatically share allocations
1618 ...............................
1620 **Option** ``smpi/auto-shared-malloc-thresh:`` **Default:** 0 (false)
1621 This value in bytes represents the size above which all allocations
1622 will be "shared" by default (as if they were performed through
1623 SMPI_SHARED_MALLOC macros). Default = 0 = disabled feature.
1624 The value must be carefully chosen to only select data buffers which
1625 will not modify execution path or cause crash if their content is false.
1626 Option :ref:`cfg=smpi/display-allocs` can be used to locate the largest
1627 allocation detected in a run, and provide a good starting threshold.
1628 Note : malloc, calloc and free are overridden by smpicc/cxx by default.
1629 This can cause some troubles if codes are already overriding these. If this
1630 is the case, defining SMPI_NO_OVERRIDE_MALLOC in the compilation flags can
1631 help, but will make this feature unusable.
1635 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1636 ...................................................................
1638 **Option** ``smpi/wtime`` **default:** 10 ns
1640 This option controls the amount of (simulated) time spent in calls to
1641 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1642 to 0, the simulated clock is not advanced in these calls, which leads
1643 to issues if your application contains such a loop:
1647 while(MPI_Wtime() < some_time_bound) {
1648 /* some tests, with no communication nor computation */
1651 When the option smpi/wtime is set to 0, the time advances only on
1652 communications and computations. So the previous code results in an
1653 infinite loop: the current [simulated] time will never reach
1654 ``some_time_bound``. This infinite loop is avoided when that option
1655 is set to a small value, as it is by default since SimGrid v3.21.
1657 Note that if your application does not contain any loop depending on
1658 the current time only, then setting this option to a non-zero value
1659 will slow down your simulations by a tiny bit: the simulation loop has
1660 to be broken out of and reset each time your code asks for the current time.
1661 If the simulation speed really matters to you, you can avoid this
1662 extra delay by setting smpi/wtime to 0.
1664 .. _cfg=smpi/list-leaks:
1666 Report leaked MPI objects
1667 .........................
1669 **Option** ``smpi/list-leaks`` **default:** 0
1671 This option controls whether to report leaked MPI objects.
1672 The parameter is the number of leaks to report.
1674 Other Configurations
1675 --------------------
1677 .. _cfg=debug/clean-atexit:
1679 Cleanup at Termination
1680 ......................
1682 **Option** ``debug/clean-atexit`` **default:** on
1684 If your code is segfaulting during its finalization, it may help to
1685 disable this option to request that SimGrid not attempt any cleanups at
1686 the end of the simulation. Since the Unix process is ending anyway,
1687 the operating system will wipe it all.
1694 **Option** ``path`` **default:** . (current dir)
1696 It is possible to specify a list of directories to search in for the
1697 trace files (see :ref:`pf_trace`) by using this configuration
1698 item. To add several directory to the path, set the configuration
1699 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1701 .. _cfg=debug/breakpoint:
1706 **Option** ``debug/breakpoint`` **default:** unset
1708 This configuration option sets a breakpoint: when the simulated clock
1709 reaches the given time, a SIGTRAP is raised. This can be used to stop
1710 the execution and get a backtrace with a debugger.
1712 It is also possible to set the breakpoint from inside the debugger, by
1713 writing in global variable simgrid::kernel::cfg_breakpoint. For example,
1716 .. code-block:: none
1718 set variable simgrid::kernel::cfg_breakpoint = 3.1416
1720 .. _cfg=debug/verbose-exit:
1725 **Option** ``debug/verbose-exit`` **default:** on
1727 By default, when Ctrl-C is pressed, the status of all existing actors
1728 is displayed before exiting the simulation. This is very useful to
1729 debug your code, but it can become troublesome if you have many
1730 actors. Set this configuration item to **off** to disable this
1733 .. _cfg=exception/cutpath:
1735 Truncate local path from exception backtrace
1736 ............................................
1738 **Option** ``exception/cutpath`` **default:** off
1740 This configuration option is used to remove the path from the
1741 backtrace shown when an exception is thrown. This is mainly useful for
1742 the tests: the full file path would makes the tests non-reproducible because
1743 the paths of source files depend of the build settings. That would
1744 break most of the tests since their output is continually compared.
1748 Logging configuration
1749 ---------------------
1751 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
1752 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
1753 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
1756 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
1757 messages from your code.
1759 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
1760 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
1761 practice, the following is equivalent to the above settings: ``--log=root.thresh:error --log=s4u_host.thresh:debug``.
1763 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
1764 your settings, as in ``--log="root.thresh:error s4u_host.thresh:debug"``. The parameters are interpreted in order, from left to right.
1767 Threshold configuration
1768 .......................
1770 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
1771 ``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
1772 see, ``threshold`` can be abbreviated here.
1774 Existing thresholds:
1776 - ``trace`` some functions display a message at this level when entering or returning
1777 - ``debug`` output that is mostly useful when debugging the corresponding module.
1778 - ``verbose`` verbose output that is only mildly interesting and can easily be ignored
1779 - ``info`` usual output (this is the default threshold of all categories)
1780 - ``warning`` minor issue encountered
1781 - ``error`` issue encountered
1782 - ``critical`` major issue encountered, such as assertions failures
1786 Format configuration
1787 ....................
1789 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
1790 as the date, or the actor ID, everything. Existing format directives:
1793 - %n: line separator (LOG4J compatible)
1794 - %e: plain old space (SimGrid extension)
1796 - %m: user-provided message
1798 - %c: Category name (LOG4J compatible)
1799 - %p: Priority name (LOG4J compatible)
1801 - %h: Hostname (SimGrid extension)
1802 - %a: Actor name (SimGrid extension -- note that with SMPI this is the integer value of the process rank)
1803 - %i: Actor PID (SimGrid extension -- this is a 'i' as in 'i'dea)
1804 - %t: Thread "name" (LOG4J compatible -- actually the address of the thread in memory)
1806 - %F: file name where the log event was raised (LOG4J compatible)
1807 - %l: location where the log event was raised (LOG4J compatible, like '%%F:%%L' -- this is a l as in 'l'etter)
1808 - %L: line number where the log event was raised (LOG4J compatible)
1809 - %M: function name (LOG4J compatible -- called method name here of course).
1811 - %d: date (UNIX-like epoch)
1812 - %r: application age (time elapsed since the beginning of the application)
1815 ``--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
1816 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
1817 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
1818 provided layout is used for every messages.
1820 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
1824 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
1825 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'"``.
1826 Another option is to use the ``%e`` directive for spaces, as in ``--log=root.fmt:%l:%e[%p/%c]:%e%m%n``.
1831 The keyword ``app`` controls the appended of a logging category. For example ``--log=root.app:file:mylogfile`` redirects every output to the file ``mylogfile``.
1833 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
1834 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.
1836 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``
1837 ensures that the log file ``mylog`` will never overpass 500 bytes in size.
1839 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
1840 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.
1845 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
1846 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
1847 ``on`` (or ``yes`` or ``1``), the produced messages will also be passed to the upper appender.
1849 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
1850 ``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
1851 will only be sent to ``all.log``.
1856 ``--help-logs`` displays a complete help message about logging in SimGrid.
1858 ``--help-log-categories`` displays the actual hierarchy of log categories for this binary.
1860 ``--log=no_loc`` hides the source locations (file names and line numbers) from the messages. This is useful to make tests reproducible.