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/buffering:** :ref:`cfg=smpi/buffering`
151 - **smpi/bw-factor:** :ref:`cfg=smpi/bw-factor`
152 - **smpi/coll-selector:** :ref:`cfg=smpi/coll-selector`
153 - **smpi/comp-adjustment-file:** :ref:`cfg=smpi/comp-adjustment-file`
154 - **smpi/cpu-threshold:** :ref:`cfg=smpi/cpu-threshold`
155 - **smpi/display-allocs:** :ref:`cfg=smpi/display-allocs`
156 - **smpi/display-timing:** :ref:`cfg=smpi/display-timing`
157 - **smpi/errors-are-fatal:** :ref:`cfg=smpi/errors-are-fatal`
158 - **smpi/finalization-barrier:** :ref:`cfg=smpi/finalization-barrier`
159 - **smpi/grow-injected-times:** :ref:`cfg=smpi/grow-injected-times`
160 - **smpi/host-speed:** :ref:`cfg=smpi/host-speed`
161 - **smpi/IB-penalty-factors:** :ref:`cfg=smpi/IB-penalty-factors`
162 - **smpi/iprobe:** :ref:`cfg=smpi/iprobe`
163 - **smpi/iprobe-cpu-usage:** :ref:`cfg=smpi/iprobe-cpu-usage`
164 - **smpi/init:** :ref:`cfg=smpi/init`
165 - **smpi/keep-temps:** :ref:`cfg=smpi/keep-temps`
166 - **smpi/lat-factor:** :ref:`cfg=smpi/lat-factor`
167 - **smpi/ois:** :ref:`cfg=smpi/ois`
168 - **smpi/or:** :ref:`cfg=smpi/or`
169 - **smpi/os:** :ref:`cfg=smpi/os`
170 - **smpi/papi-events:** :ref:`cfg=smpi/papi-events`
171 - **smpi/pedantic:** :ref:`cfg=smpi/pedantic`
172 - **smpi/privatization:** :ref:`cfg=smpi/privatization`
173 - **smpi/privatize-libs:** :ref:`cfg=smpi/privatize-libs`
174 - **smpi/send-is-detached-thresh:** :ref:`cfg=smpi/send-is-detached-thresh`
175 - **smpi/shared-malloc:** :ref:`cfg=smpi/shared-malloc`
176 - **smpi/shared-malloc-hugepage:** :ref:`cfg=smpi/shared-malloc-hugepage`
177 - **smpi/simulate-computation:** :ref:`cfg=smpi/simulate-computation`
178 - **smpi/test:** :ref:`cfg=smpi/test`
179 - **smpi/wtime:** :ref:`cfg=smpi/wtime`
180 - **smpi/list-leaks** :ref:`cfg=smpi/list-leaks`
182 - **Tracing configuration options** can be found in Section :ref:`tracing_tracing_options`
184 - **storage/model:** :ref:`options_model_select`
186 - **vm/model:** :ref:`options_model_select`
190 Configuring the Platform Models
191 -------------------------------
193 .. _options_model_select:
195 Choosing the Platform Models
196 ............................
198 SimGrid comes with several network, CPU and disk models built in,
199 and you can change the used model at runtime by changing the passed
200 configuration. The three main configuration items are given below.
201 For each of these items, passing the special ``help`` value gives you
202 a short description of all possible values (for example,
203 ``--cfg=network/model:help`` will present all provided network
204 models). Also, ``--help-models`` should provide information about all
205 models for all existing resources.
207 - ``network/model``: specify the used network model. Possible values:
209 - **LV08 (default one):** Realistic network analytic model
210 (slow-start modeled by multiplying latency by 13.01, bandwidth by
211 .97; bottleneck sharing uses a payload of S=20537 for evaluating
212 RTT). Described in `Accuracy Study and Improvement of Network
213 Simulation in the SimGrid Framework
214 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
215 - **Constant:** Simplistic network model where all communication
216 take a constant time (one second). This model provides the lowest
217 realism, but is (marginally) faster.
218 - **SMPI:** Realistic network model specifically tailored for HPC
219 settings (accurate modeling of slow start with correction factors on
220 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). This model can be
221 :ref:`further configured <options_model_network>`.
222 - **IB:** Realistic network model specifically tailored for HPC
223 settings with InfiniBand networks (accurate modeling contention
224 behavior, based on the model explained in `this PhD work
225 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
226 This model can be :ref:`further configured <options_model_network>`.
227 - **CM02:** Legacy network analytic model. Very similar to LV08, but
228 without corrective factors. The timings of small messages are thus
229 poorly modeled. This model is described in `A Network Model for
230 Simulation of Grid Application
231 <https://hal.inria.fr/inria-00071989/document>`_.
232 - **ns-3** (only available if you compiled SimGrid accordingly):
233 Use the packet-level network
234 simulators as network models (see :ref:`model_ns3`).
235 This model can be :ref:`further configured <options_pls>`.
237 - ``cpu/model``: specify the used CPU model. We have only one model
240 - **Cas01:** Simplistic CPU model (time=size/speed)
242 - ``host/model``: The host concept is the aggregation of a CPU with a
243 network card. Three models exists, but actually, only 2 of them are
244 interesting. The "compound" one is simply due to the way our
245 internal code is organized, and can easily be ignored. So at the
246 end, you have two host models: The default one allows aggregation of
247 an existing CPU model with an existing network model, but does not
248 allow parallel tasks because these beasts need some collaboration
249 between the network and CPU model.
251 - **default:** Default host model. Currently, CPU:Cas01 and
252 network:LV08 (with cross traffic enabled)
253 - **compound:** Host model that is automatically chosen if
254 you change the network and CPU models
255 - **ptask_L07:** Host model somehow similar to Cas01+CM02 but
256 allowing "parallel tasks", that are intended to model the moldable
257 tasks of the grid scheduling literature.
259 - ``storage/model``: specify the used storage model. Only one model is
261 - ``vm/model``: specify the model for virtual machines. Only one model
264 .. todo: make 'compound' the default host model.
266 .. _options_model_solver:
271 The different models rely on a linear inequalities solver to share
272 the underlying resources. SimGrid allows you to change the solver, but
273 be cautious, **don't change it unless you are 100% sure**.
275 - items ``cpu/solver``, ``network/solver``, ``disk/solver`` and ``host/solver``
276 allow you to change the solver for each model:
278 - **maxmin:** The default solver for all models except ptask. Provides a
279 max-min fairness allocation.
280 - **fairbottleneck:** The default solver for ptasks. Extends max-min to
281 allow heterogeneous resources.
282 - **bmf:** More realistic solver for heterogeneous resource sharing.
283 Implements BMF (Bottleneck max fairness) fairness. To be used with
284 parallel tasks instead of fair-bottleneck.
286 .. _options_model_optim:
291 The network and CPU models that are based on linear inequalities solver (that
292 is, all our analytical models) accept specific optimization
295 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
297 - **Lazy:** Lazy action management (partial invalidation in lmm +
298 heap in action remaining).
299 - **TI:** Trace integration. Highly optimized mode when using
300 availability traces (only available for the Cas01 CPU model for
302 - **Full:** Full update of remaining and variables. Slow but may be
303 useful when debugging.
305 - items ``network/maxmin-selective-update`` and
306 ``cpu/maxmin-selective-update``: configure whether the underlying
307 should be lazily updated or not. It should have no impact on the
308 computed timings, but should speed up the computation. |br| It is
309 still possible to disable this feature because it can reveal
310 counter-productive in very specific scenarios where the
311 interaction level is high. In particular, if all your
312 communication share a given backbone link, you should disable it:
313 without it, a simple regular loop is used to update each
314 communication. With it, each of them is still updated (because of
315 the dependency induced by the backbone), but through a complicated
316 and slow pattern that follows the actual dependencies.
318 .. _cfg=bmf/precision:
319 .. _cfg=maxmin/precision:
320 .. _cfg=surf/precision:
325 **Option** ``maxmin/precision`` **Default:** 1e-5 (in flops or bytes) |br|
326 **Option** ``surf/precision`` **Default:** 1e-9 (in seconds) |br|
327 **Option** ``bmf/precision`` **Default:** 1e-12 (no unit)
329 The analytical models handle a lot of floating point values. It is
330 possible to change the epsilon used to update and compare them through
331 this configuration item. Changing it may speedup the simulation by
332 discarding very small actions, at the price of a reduced numerical
333 precision. You can modify separately the precision used to manipulate
334 timings (in seconds) and the one used to manipulate amounts of work
337 .. _cfg=maxmin/concurrency-limit:
342 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
344 The maximum number of variables per resource can be tuned through this
345 option. You can have as many simultaneous actions per resources as you
346 want. If your simulation presents a very high level of concurrency, it
347 may help to use e.g. 100 as a value here. It means that at most 100
348 actions can consume a resource at a given time. The extraneous actions
349 are queued and wait until the amount of concurrency of the considered
350 resource lowers under the given boundary.
352 Such limitations help both to the simulation speed and simulation accuracy
353 on highly constrained scenarios, but the simulation speed suffers of this
354 setting on regular (less constrained) scenarios so it is off by default.
356 .. _cfg=bmf/max-iterations:
361 **Option** ``bmf/max-iterations`` **Default:** 1000
363 It may happen in some settings that the BMF solver fails to converge to
364 a solution, so there is a hard limit on the amount of iteration count to
365 avoid infinite loops.
367 .. _options_model_network:
369 Configuring the Network Model
370 .............................
372 .. _cfg=network/TCP-gamma:
374 Maximal TCP Window Size
375 ^^^^^^^^^^^^^^^^^^^^^^^
377 **Option** ``network/TCP-gamma`` **Default:** 4194304
379 The analytical models need to know the maximal TCP window size to take
380 the TCP congestion mechanism into account. On Linux, this value can
381 be retrieved using the following commands. Both give a set of values,
382 and you should use the last one, which is the maximal size.
384 .. code-block:: console
386 $ cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
387 $ cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
389 .. _cfg=network/bandwidth-factor:
390 .. _cfg=network/latency-factor:
391 .. _cfg=network/weight-S:
393 Correcting Important Network Parameters
394 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
396 SimGrid can take network irregularities such as a slow startup or
397 changing behavior depending on the message size into account. You
398 should not change these values unless you really know what you're
399 doing. The corresponding values were computed through data fitting
400 one the timings of packet-level simulators, as described in `Accuracy
401 Study and Improvement of Network Simulation in the SimGrid Framework
402 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
404 - **network/latency-factor**: apply a multiplier to latency.
405 Models the TCP slow-start mechanism.
406 - **network/bandwidth-factor**: actual bandwidth perceived by the
408 - **network/weight-S**: bottleneck sharing constant parameter. Used
411 These parameters are the same for all communications in your simulation,
412 independently of message size or source/destination hosts. A more flexible
413 mechanism based on callbacks was introduced in SimGrid. It provides the user
414 a callback that will be called for each communication, allowing the user
415 to set different latency and bandwidth factors, based on the message size, links used
416 or zones traversed. To more details of how to use it, please look at the
417 `examples/cpp/network-factors/s4u-network-factors.cpp <https://framagit.org/simgrid/simgrid/tree/master/examples/cpp/network-factors/s4u-network-factors.cpp>`_.
420 If you are using the SMPI model, these correction coefficients are
421 themselves corrected by constant values depending on the size of the
422 exchange. By default SMPI uses factors computed on the Stampede
423 Supercomputer at TACC, with optimal deployment of processes on
424 nodes. Again, only hardcore experts should bother about this fact.
425 For more details, see SMPI sections about :ref:`cfg=smpi/bw-factor` and :ref:`cfg=smpi/lat-factor`.
428 .. _cfg=smpi/IB-penalty-factors:
433 InfiniBand network behavior can be modeled through 3 parameters
434 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `this PhD
436 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_ (in French)
437 or more concisely in `this paper <https://hal.inria.fr/hal-00953618/document>`_,
438 even if that paper does only describe models for myrinet and ethernet.
439 You can see in Fig 2 some results for Infiniband, for example. This model
440 may be outdated by now for modern infiniband, anyway, so a new
441 validation would be good.
443 The three paramaters are defined as follows:
445 - βs: penalty factor for outgoing messages, computed by running a simple send to
446 two nodes and checking slowdown compared to a single send to one node,
448 - βe: penalty factor for ingoing messages, same computation method but with one
449 node receiving several messages
450 - γr: slowdown factor when communication buffer memory is saturated. It needs a
451 more complicated pattern to run in order to be computed (5.3 in the thesis,
452 page 107), and formula in the end is γr = time(c)/(3×βe×time(ref)), where
453 time(ref) is the time of a single comm with no contention).
455 Once these values are computed, a penalty is assessed for each message (this is
456 the part implemented in the simulator) as shown page 106 of the thesis. Here is
457 a simple translation of this text. First, some notations:
459 - ∆e(e) which corresponds to the incoming degree of node e, that is to say the number of communications having as destination node e.
460 - ∆s (s) which corresponds to the degree outgoing from node s, that is to say the number of communications sent by node s.
461 - Φ (e) which corresponds to the number of communications destined for the node e but coming from a different node.
462 - 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
464 To determine the penalty for a communication, two values need to be calculated. First, the penalty caused by the conflict in transmission, noted ps.
467 - if ∆s (i) = 1 then ps = 1.
468 - if ∆s (i) ≥ 2 and ∆e (i) ≥ 3 then ps = ∆s (i) × βs × γr
469 - else, ps = ∆s (i) × βs
472 Then, the penalty caused by the conflict in reception (noted pe) should be computed as follows:
474 - if ∆e (i) = 1 then pe = 1
475 - else, pe = Φ (e) × βe × Ω (s, e)
477 Finally, the penalty associated with the communication is:
478 p = max (ps ∈ s, pe)
480 .. _cfg=network/crosstraffic:
482 Simulating Cross-Traffic
483 ^^^^^^^^^^^^^^^^^^^^^^^^
485 Since SimGrid v3.7, cross-traffic effects can be taken into account in
486 analytical simulations. It means that ongoing and incoming
487 communication flows are treated independently. In addition, the LV08
488 model adds 0.05 of usage on the opposite direction for each new
489 created flow. This can be useful to simulate some important TCP
490 phenomena such as ack compression.
492 For that to work, your platform must have two links for each
493 pair of interconnected hosts. An example of usable platform is
494 available in ``examples/platforms/crosstraffic.xml``.
496 This is activated through the ``network/crosstraffic`` item, that
497 can be set to 0 (disable this feature) or 1 (enable it).
499 Note that with the default host model this option is activated by default.
501 .. _cfg=network/loopback:
503 Configuring loopback link
504 ^^^^^^^^^^^^^^^^^^^^^^^^^
506 Several network model provide an implicit loopback link to account for local
507 communication on a host. By default it has a 10GBps bandwidth and a null latency.
508 This can be changed with ``network/loopback-lat`` and ``network/loopback-bw``
511 .. _cfg=smpi/async-small-thresh:
513 Simulating Asynchronous Send
514 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
516 (this configuration item is experimental and may change or disappear)
518 It is possible to specify that messages below a certain size (in bytes) will be
519 sent as soon as the call to MPI_Send is issued, without waiting for
520 the correspondent receive. This threshold can be configured through
521 the ``smpi/async-small-thresh`` item. The default value is 0. This
522 behavior can also be manually set for mailboxes, by setting the
523 receiving mode of the mailbox with a call to
524 :cpp:func:`MSG_mailbox_set_async`. After this, all messages sent to
525 this mailbox will have this behavior regardless of the message size.
527 This value needs to be smaller than or equals to the threshold set at
528 :ref:`cfg=smpi/send-is-detached-thresh`, because asynchronous messages
529 are meant to be detached as well.
536 **Option** ``ns3/TcpModel`` **Default:** "default" (ns-3 default)
538 When using ns-3, there is an extra item ``ns3/TcpModel``, corresponding
539 to the ``ns3::TcpL4Protocol::SocketType`` configuration item in
540 ns-3. The only valid values (enforced on the SimGrid side) are
541 'default' (no change to the ns-3 configuration), 'NewReno' or 'Reno' or
544 **Option** ``ns3/seed`` **Default:** "" (don't set the seed in ns-3)
546 This option is the random seed to provide to ns-3 with
547 ``ns3::RngSeedManager::SetSeed`` and ``ns3::RngSeedManager::SetRun``.
549 If left blank, no seed is set in ns-3. If the value 'time' is
550 provided, the current amount of seconds since epoch is used as a seed.
551 Otherwise, the provided value must be a number to use as a seed.
553 Configuring the Storage model
554 .............................
556 .. _cfg=storage/max_file_descriptors:
558 File Descriptor Count per Host
559 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
561 **Option** ``storage/max_file_descriptors`` **Default:** 1024
563 Each host maintains a fixed-size array of its file descriptors. You
564 can change its size through this item to either enlarge it if your
565 application requires it or to reduce it to save memory space.
572 SimGrid plugins allow one to extend the framework without changing its
573 source code directly. Read the source code of the existing plugins to
574 learn how to do so (in ``src/plugins``), and ask your questions to the
575 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
576 that plugins usually register callbacks to some signals of interest.
577 If they need to store some information about a given object (Link, CPU
578 or Actor), they do so through the use of a dedicated object extension.
580 Some of the existing plugins can be activated from the command line,
581 meaning that you can activate them from the command line without any
582 modification to your simulation code. For example, you can activate
583 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
586 Here is a partial list of plugins that can be activated this way. You can get
587 the full list by passing ``--cfg=plugin:help`` to your simulator.
589 - :ref:`Host Energy <plugin_host_energy>`: models the energy dissipation of the compute units.
590 - :ref:`Link Energy <plugin_link_energy>`: models the energy dissipation of the network.
591 - :ref:`Host Load <plugin_host_load>`: monitors the load of the compute units.
593 .. _options_modelchecking:
595 Configuring the Model-Checking
596 ------------------------------
598 To enable SimGrid's model-checking support, the program should
599 be executed using the simgrid-mc wrapper:
601 .. code-block:: console
603 $ simgrid-mc ./my_program
605 Safety properties are expressed as assertions using the function
606 :cpp:func:`void MC_assert(int prop)`.
608 .. _cfg=smpi/buffering:
610 Specifying the MPI buffering behavior
611 .....................................
613 **Option** ``smpi/buffering`` **Default:** infty
615 Buffering in MPI has a huge impact on the communication semantic. For example,
616 standard blocking sends are synchronous calls when the system buffers are full
617 while these calls can complete immediately without even requiring a matching
618 receive call for small messages sent when the system buffers are empty.
620 In SMPI, this depends on the message size, that is compared against two thresholds:
622 - if (size < :ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>`) then
623 MPI_Send returns immediately, even if the corresponding receive has not be issued yet.
624 - 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
625 MPI_Send returns as soon as the corresponding receive has been issued. This is known as the eager mode.
626 - if (:ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>` < size) then
627 MPI_Send returns only when the message has actually been sent over the network. This is known as the rendez-vous mode.
629 The ``smpi/buffering`` (only valid with MC) option gives an easier interface to choose between these semantics. It can take two values:
631 - **zero:** means that buffering should be disabled. All communications are actually blocking.
632 - **infty:** means that buffering should be made infinite. All communications are non-blocking.
634 .. _cfg=model-check/property:
636 Specifying a liveness property
637 ..............................
639 **Option** ``model-check/property`` **Default:** unset
641 If you want to specify liveness properties, you have to pass them on
642 the command line, specifying the name of the file containing the
643 property, as formatted by the `ltl2ba <https://github.com/utwente-fmt/ltl2ba>`_ program.
644 Note that ltl2ba is not part of SimGrid and must be installed separately.
646 .. code-block:: console
648 $ simgrid-mc ./my_program --cfg=model-check/property:<filename>
650 .. _cfg=model-check/checkpoint:
652 Going for Stateful Verification
653 ...............................
655 By default, the system is backtracked to its initial state to explore
656 another path, instead of backtracking to the exact step before the fork
657 that we want to explore (this is called stateless verification). This
658 is done this way because saving intermediate states can rapidly
659 exhaust the available memory. If you want, you can change the value of
660 the ``model-check/checkpoint`` item. For example,
661 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
662 step. Beware, this will certainly explode your memory. Larger values
663 are probably better, make sure to experiment a bit to find the right
664 setting for your specific system.
666 .. _cfg=model-check/reduction:
668 Specifying the kind of reduction
669 ................................
671 The main issue when using the model-checking is the state space
672 explosion. You can activate some reduction technique with
673 ``--cfg=model-check/reduction:<technique>``. For now, this
674 configuration variable can take 2 values:
676 - **none:** Do not apply any kind of reduction (mandatory for
677 liveness properties, as our current DPOR algorithm breaks cycles)
678 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if
679 you verify local safety properties (default value for safety
682 Another way to mitigate the state space explosion is to search for
683 cycles in the exploration with the :ref:`cfg=model-check/visited`
684 configuration. Note that DPOR and state-equality reduction may not
685 play well together. You should choose between them.
687 Our current DPOR implementation could be improved in may ways. We are
688 currently improving its efficiency (both in term of reduction ability
689 and computational speed), and future work could make it compatible
690 with liveness properties.
692 .. _cfg=model-check/visited:
694 Size of Cycle Detection Set (state equality reduction)
695 ......................................................
697 Mc SimGrid can be asked to search for cycles during the exploration,
698 i.e. situations where a new explored state is in fact the same state
699 than a previous one.. This can prove useful to mitigate the state
700 space explosion with safety properties, and this is the crux when
701 searching for counter-examples to the liveness properties.
703 Note that this feature may break the current implementation of the
704 DPOR reduction technique.
706 The ``model-check/visited`` item is the maximum number of states, which
707 are stored in memory. If the maximum number of snapshotted state is
708 reached, some states will be removed from the memory and some cycles
709 might be missed. Small values can lead to incorrect verifications, but
710 large values can exhaust your memory and be CPU intensive as each new
711 state must be compared to that amount of older saved states.
713 The default settings depend on the kind of exploration. With safety
714 checking, no state is snapshotted and cycles cannot be detected. With
715 liveness checking, all states are snapshotted because missing a cycle
716 could hinder the exploration soundness.
718 .. _cfg=model-check/termination:
720 Non-Termination Detection
721 .........................
723 The ``model-check/termination`` configuration item can be used to
724 report if a non-termination execution path has been found. This is a
725 path with a cycle, which means that the program might never terminate.
727 This only works in safety mode, not in liveness mode.
729 This options is disabled by default.
731 .. _cfg=model-check/dot-output:
736 If set, the ``model-check/dot-output`` configuration item is the name
737 of a file in which to write a dot file of the path leading to the
738 property violation discovered (safety or liveness violation), as well
739 as the cycle for liveness properties. This dot file can then be fed to the
740 graphviz dot tool to generate a corresponding graphical representation.
742 .. _cfg=model-check/max-depth:
744 Exploration Depth Limit
745 .......................
747 The ``model-check/max-depth`` can set the maximum depth of the
748 exploration graph of the model checker. If this limit is reached, a
749 logging message is sent and the results might not be exact.
751 By default, the exploration is limited to the depth of 1000.
753 .. _cfg=model-check/timeout:
758 By default, the model checker does not handle timeout conditions: the `wait`
759 operations never time out. With the ``model-check/timeout`` configuration item
760 set to **yes**, the model checker will explore timeouts of `wait` operations.
762 .. _cfg=model-check/communications-determinism:
763 .. _cfg=model-check/send-determinism:
765 Communication Determinism
766 .........................
768 The ``model-check/communications-determinism`` and
769 ``model-check/send-determinism`` items can be used to select the
770 communication determinism mode of the model checker, which checks
771 determinism properties of the communications of an application.
773 .. _cfg=model-check/setenv:
775 Passing environment variables
776 .............................
778 You can specify extra environment variables to be set in the verified application
779 with ``model-check/setenv``. For example, you can preload a library as follows:
780 ``-cfg=model-check/setenv:LD_PRELOAD=toto;LD_LIBRARY_PATH=/tmp``.
784 Verification Performance Considerations
785 .......................................
787 The size of the stacks can have a huge impact on the memory
788 consumption when using model-checking. By default, each snapshot will
789 save a copy of the whole stacks and not only of the part that is
790 really meaningful: you should expect the contribution of the memory
791 consumption of the snapshots to be:
792 :math:`\text{number of processes} \times \text{stack size} \times \text{number of states}`.
794 When compiled against the model checker, the stacks are not
795 protected with guards: if the stack size is too small for your
796 application, the stack will silently overflow into other parts of the
797 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
799 .. _cfg=model-check/replay:
801 Replaying buggy execution paths from the model checker
802 ......................................................
804 Debugging the problems reported by the model checker is challenging:
805 First, the application under verification cannot be debugged with gdb
806 because the model checker already traces it. Then, the model checker may
807 explore several execution paths before encountering the issue, making it
808 very difficult to understand the output. Fortunately, SimGrid provides
809 the execution path leading to any reported issue so that you can replay
810 this path reported by the model checker, enabling the usage of classical
813 When the model checker finds an interesting path in the application
814 execution graph (where a safety or liveness property is violated), it
815 generates an identifier for this path. Here is an example of the output:
817 .. code-block:: console
819 [ 0.000000] (0:@) Check a safety property
820 [ 0.000000] (0:@) **************************
821 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
822 [ 0.000000] (0:@) **************************
823 [ 0.000000] (0:@) Counter-example execution trace:
824 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
825 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
826 [ 0.000000] (0:@) Path = 1/3;1/4
827 [ 0.000000] (0:@) Expanded states = 27
828 [ 0.000000] (0:@) Visited states = 68
829 [ 0.000000] (0:@) Executed transitions = 46
831 The interesting line is ``Path = 1/3;1/4``, which means that you should use
832 ``--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
833 execution path. All options (but the model checker related ones) must
834 remain the same. In particular, if you ran your application with
835 ``smpirun -wrapper simgrid-mc``, then do it again. Remove all
836 MC-related options, keep non-MC-related ones and add
837 ``--cfg=model-check/replay:???``.
839 Currently, if the path is of the form ``X;Y;Z``, each number denotes
840 the actor's pid that is selected at each indecision point. If it's of
841 the form ``X/a;Y/b``, the X and Y are the selected pids while the a
842 and b are the return values of their simcalls. In the previous
843 example, ``1/3;1/4``, you can see from the full output that the actor
844 1 is doing MC_RANDOM simcalls, so the 3 and 4 simply denote the values
845 that these simcall return on the execution branch leading to the
848 Configuring the User Code Virtualization
849 ----------------------------------------
851 .. _cfg=contexts/factory:
853 Selecting the Virtualization Factory
854 ....................................
856 **Option** contexts/factory **Default:** "raw"
858 In SimGrid, the user code is virtualized in a specific mechanism that
859 allows the simulation kernel to control its execution: when a user
860 process requires a blocking action (such as sending a message), it is
861 interrupted, and only gets released when the simulated clock reaches
862 the point where the blocking operation is done. This is explained
863 graphically in the `relevant tutorial, available online
864 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
866 In SimGrid, the containers in which user processes are virtualized are
867 called contexts. Several context factory are provided, and you can
868 select the one you want to use with the ``contexts/factory``
869 configuration item. Some of the following may not exist on your
870 machine because of portability issues. In any case, the default one
871 should be the most effcient one (please report bugs if the
872 auto-detection fails for you). They are approximately sorted here from
873 the slowest to the most efficient:
875 - **thread:** very slow factory using full featured threads (either
876 pthreads or windows native threads). They are slow but very
877 standard. Some debuggers or profilers only work with this factory.
878 - **java:** Java applications are virtualized onto java threads (that
879 are regular pthreads registered to the JVM)
880 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
881 - **boost:** This uses the `context
882 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
883 of the boost library for a performance that is comparable to our
885 |br| Install the relevant library (e.g. with the
886 libboost-contexts-dev package on Debian/Ubuntu) and recompile
888 - **raw:** amazingly fast factory using a context switching mechanism
889 of our own, directly implemented in assembly (only available for x86
890 and amd64 platforms for now) and without any unneeded system call.
892 The main reason to change this setting is when the debugging tools become
893 fooled by the optimized context factories. Threads are the most
894 debugging-friendly contexts, as they allow one to set breakpoints
895 anywhere with gdb and visualize backtraces for all processes, in order
896 to debug concurrency issues. Valgrind is also more comfortable with
897 threads, but it should be usable with all factories (Exception: the
898 callgrind tool really dislikes raw and ucontext factories).
900 .. _cfg=contexts/stack-size:
902 Adapting the Stack Size
903 .......................
905 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
907 Each virtualized used process is executed using a specific system
908 stack. The size of this stack has a huge impact on the simulation
909 scalability, but its default value is rather large. This is because
910 the error messages that you get when the stack size is too small are
911 rather disturbing: this leads to stack overflow (overwriting other
912 stacks), leading to segfaults with corrupted stack traces.
914 If you want to push the scalability limits of your code, you might
915 want to reduce the ``contexts/stack-size`` item. Its default value is
916 8192 (in KiB), while our Chord simulation works with stacks as small
917 as 16 KiB, for example. You can ensure that some actors have a specific
918 size by simply changing the value of this configuration item before
919 creating these actors. The :cpp:func:`simgrid::s4u::Engine::set_config`
920 functions are handy for that.
922 This *setting is ignored* when using the thread factory (because there
923 is no way to modify the stack size with C++ system threads). Instead,
924 you should compile SimGrid and your application with
925 ``-fsplit-stack``. Note that this compilation flag is not compatible
926 with the model checker right now.
928 The operating system should only allocate memory for the pages of the
929 stack which are actually used and you might not need to use this in
930 most cases. However, this setting is very important when using the
931 model checker (see :ref:`options_mc_perf`).
933 .. _cfg=contexts/guard-size:
935 Disabling Stack Guard Pages
936 ...........................
938 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages on Windows or with MC)
940 Unless you use the threads context factory (see
941 :ref:`cfg=contexts/factory`), a stack guard page is usually used
942 which prevents the stack of a given actor from overflowing on another
943 stack. But the performance impact may become prohibitive when the
944 amount of actors increases. The option ``contexts/guard-size`` is the
945 number of stack guard pages used. By setting it to 0, no guard pages
946 will be used: in this case, you should avoid using small stacks (with
947 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
948 will silently overflow on other parts of the memory.
950 When no stack guard page is created, stacks may then silently overflow
951 on other parts of the memory if their size is too small for the
954 .. _cfg=contexts/nthreads:
955 .. _cfg=contexts/synchro:
957 Running User Code in Parallel
958 .............................
960 Parallel execution of the user code is only considered stable in
961 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
962 simulations may well fail in parallel mode. It is described in
963 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
965 If you are using the **ucontext** or **raw** context factories, you can
966 request to execute the user code in parallel. Several threads are
967 launched, each of them handling the same number of user contexts at each
968 run. To activate this, set the ``contexts/nthreads`` item to the amount
969 of cores that you have in your computer (or lower than 1 to have the
970 amount of cores auto-detected).
972 When parallel execution is activated, you can choose the
973 synchronization schema used with the ``contexts/synchro`` item,
974 which value is either:
976 - **futex:** ultra optimized synchronisation schema, based on futexes
977 (fast user-mode mutexes), and thus only available on Linux systems.
978 This is the default mode when available.
979 - **posix:** slow but portable synchronisation using only POSIX
981 - **busy_wait:** not really a synchronisation: the worker threads
982 constantly request new contexts to execute. It should be the most
983 efficient synchronisation schema, but it loads all the cores of
984 your machine for no good reason. You probably prefer the other less
987 Configuring the Tracing
988 -----------------------
990 The :ref:`tracing subsystem <outcome_vizu>` can be configured in
991 several different ways depending on the used interface (S4U, SMPI)
992 and the kind of traces that needs to be obtained. See the
993 :ref:`Tracing Configuration Options subsection
994 <tracing_tracing_options>` for a full description of each
995 configuration option.
997 We detail here a simple way to get the traces working for you, even if
998 you never used the tracing API.
1001 - Any SimGrid-based simulator (MSG, SMPI, ...) and raw traces:
1003 .. code-block:: none
1005 --cfg=tracing:yes --cfg=tracing/uncategorized:yes
1007 The first parameter activates the tracing subsystem, and the second
1008 tells it to trace host and link utilization (without any
1011 - MSG-based simulator and categorized traces (you need to
1012 declare categories and classify your tasks according to them)
1014 .. code-block:: none
1016 --cfg=tracing:yes --cfg=tracing/categorized:yes
1018 The first parameter activates the tracing subsystem, and the second
1019 tells it to trace host and link categorized utilization.
1021 - SMPI simulator and traces for a space/time view:
1023 .. code-block:: console
1025 $ smpirun -trace ...
1027 The `-trace` parameter for the smpirun script runs the simulation
1028 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
1029 smpirun's `-help` parameter for additional tracing options.
1031 Sometimes you might want to put additional information on the trace to
1032 correctly identify them later, or to provide data that can be used to
1033 reproduce an experiment. You have two ways to do that:
1035 - Add a string on top of the trace file as comment:
1037 .. code-block:: none
1039 --cfg=tracing/comment:my_simulation_identifier
1041 - Add the contents of a textual file on top of the trace file as comment:
1043 .. code-block:: none
1045 --cfg=tracing/comment-file:my_file_with_additional_information.txt
1047 Please, use these two parameters (for comments) to make reproducible
1048 simulations. For additional details about this and all tracing
1049 options, check See the :ref:`tracing_tracing_options`.
1054 .. _cfg=msg/debug-multiple-use:
1059 **Option** ``msg/debug-multiple-use`` **Default:** off
1061 Sometimes your application may try to send a task that is still being
1062 executed somewhere else, making it impossible to send this task. However,
1063 for debugging purposes, one may want to know what the other host is/was
1064 doing. This option shows a backtrace of the other process.
1069 The SMPI interface provides several specific configuration items.
1070 These are not easy to see, since the code is usually launched through the
1071 ``smiprun`` script directly.
1073 .. _cfg=smpi/host-speed:
1074 .. _cfg=smpi/cpu-threshold:
1075 .. _cfg=smpi/simulate-computation:
1077 Automatic Benchmarking of SMPI Code
1078 ...................................
1080 In SMPI, the sequential code is automatically benchmarked, and these
1081 computations are automatically reported to the simulator. That is to
1082 say that if you have a large computation between a ``MPI_Recv()`` and
1083 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
1084 this code, and create an execution task within the simulator to take
1085 this into account. For that, the actual duration is measured on the
1086 host machine and then scaled to the power of the corresponding
1087 simulated machine. The variable ``smpi/host-speed`` allows one to
1088 specify the computational speed of the host machine (in flop/s by
1089 default) to use when scaling the execution times.
1091 The default value is ``smpi/host-speed=20kf`` (= 20,000 flop/s). This
1092 is probably underestimated for most machines, leading SimGrid to
1093 overestimate the amount of flops in the execution blocks that are
1094 automatically injected in the simulator. As a result, the execution
1095 time of the whole application will probably be overestimated until you
1096 use a realistic value.
1098 When the code consists of numerous consecutive MPI calls, the
1099 previous mechanism feeds the simulation kernel with numerous tiny
1100 computations. The ``smpi/cpu-threshold`` item becomes handy when this
1101 impacts badly on the simulation performance. It specifies a threshold (in
1102 seconds) below which the execution chunks are not reported to the
1103 simulation kernel (default value: 1e-6).
1105 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
1106 time spent below this threshold. SMPI does not consider the
1107 `amount of time` of these computations; there is no offset for
1108 this. Hence, a value that is too small, may lead to unreliable
1111 In some cases, however, one may wish to disable simulation of
1112 the computation of an application. This is the case when SMPI is used not to
1113 simulate an MPI application, but instead an MPI code that performs
1114 "live replay" of another MPI app (e.g., ScalaTrace's replay tool, or
1115 various on-line simulators that run an app at scale). In this case the
1116 computation of the replay/simulation logic should not be simulated by
1117 SMPI. Instead, the replay tool or on-line simulator will issue
1118 "computation events", which correspond to the actual MPI simulation
1119 being replayed/simulated. At the moment, these computation events can
1120 be simulated using SMPI by calling internal smpi_execute*() functions.
1122 To disable the benchmarking/simulation of a computation in the simulated
1123 application, the variable ``smpi/simulate-computation`` should be set
1124 to **no**. This option just ignores the timings in your simulation; it
1125 still executes the computations itself. If you want to stop SMPI from
1126 doing that, you should check the SMPI_SAMPLE macros, documented in
1127 Section :ref:`SMPI_use_faster`.
1129 +------------------------------------+-------------------------+-----------------------------+
1130 | Solution | Computations executed? | Computations simulated? |
1131 +====================================+=========================+=============================+
1132 | --cfg=smpi/simulate-computation:no | Yes | Never |
1133 +------------------------------------+-------------------------+-----------------------------+
1134 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
1135 +------------------------------------+-------------------------+-----------------------------+
1136 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
1137 +------------------------------------+-------------------------+-----------------------------+
1139 .. _cfg=smpi/comp-adjustment-file:
1141 Slow-down or speed-up parts of your code
1142 ........................................
1144 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
1146 This option allows you to pass a file that contains two columns: The
1147 first column defines the section that will be subject to a speedup;
1148 the second column is the speedup. For instance:
1150 .. code-block:: none
1152 "start:stop","ratio"
1153 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1155 The first line is the header - you must include it. The following
1156 line means that the code between two consecutive MPI calls on line 30
1157 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1158 of 1.18244559422142. The value for the second column is therefore a
1159 speedup, if it is larger than 1 and a slowdown if it is smaller
1160 than 1. Nothing will be changed if it is equal to 1.
1162 Of course, you can set any arbitrary filenames you want (so the start
1163 and end don't have to be in the same file), but be aware that this
1164 mechanism only supports `consecutive calls!`
1166 Please note that you must pass the ``-trace-call-location`` flag to
1167 smpicc or smpiff, respectively. This flag activates some internal
1168 macro definitions that help with obtaining the call location.
1170 .. _cfg=smpi/bw-factor:
1175 **Option** ``smpi/bw-factor``
1176 |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
1178 The possible throughput of network links is often dependent on the
1179 message sizes, as protocols may adapt to different message sizes. With
1180 this option, a series of message sizes and factors are given, helping
1181 the simulation to be more realistic. For instance, the current default
1182 value means that messages with size 65472 bytes and more will get a total of
1183 MAX_BANDWIDTH*0.940694, messages of size 15424 to 65471 will get
1184 MAX_BANDWIDTH*0.697866, and so on (where MAX_BANDWIDTH denotes the
1185 bandwidth of the link).
1187 An experimental script to compute these factors is available online. See
1188 https://framagit.org/simgrid/platform-calibration/
1189 https://simgrid.org/contrib/smpi-saturation-doc.html
1191 .. _cfg=smpi/display-timing:
1193 Reporting Simulation Time
1194 .........................
1196 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1198 Most of the time, you run MPI code with SMPI to compute the time it
1199 would take to run it on a platform. But since the code is run through
1200 the ``smpirun`` script, you don't have any control on the launcher
1201 code, making it difficult to report the simulated time when the
1202 simulation ends. If you enable the ``smpi/display-timing`` item,
1203 ``smpirun`` will display this information when the simulation
1205 SMPI will also display information about the amout of real time spent
1206 in application code and in SMPI internals, to provide hints about the
1207 need to use sampling to reduce simulation time.
1209 .. _cfg=smpi/display-allocs:
1211 Reporting memory allocations
1212 ............................
1214 **Option** ``smpi/display-allocs`` **Default:** 0 (false)
1216 SMPI intercepts malloc and calloc calls performed inside the running
1217 application, if it wasn't compiled with SMPI_NO_OVERRIDE_MALLOC.
1218 With this option, SMPI will show at the end of execution the amount of
1219 memory allocated through these calls, and locate the most expensive one.
1220 This helps finding the targets for manual memory sharing, or the threshold
1221 to use for smpi/auto-shared-malloc-thresh option (see :ref:`cfg=smpi/auto-shared-malloc-thresh`).
1223 .. _cfg=smpi/keep-temps:
1225 Keeping temporary files after simulation
1226 ........................................
1228 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1230 SMPI usually generates a lot of temporary files that are cleaned after
1231 use. This option requests to preserve them, for example to debug or
1232 profile your code. Indeed, the binary files are removed very early
1233 under the dlopen privatization schema, which tends to fool the
1236 .. _cfg=smpi/lat-factor:
1241 **Option** ``smpi/lat-factor`` |br|
1242 **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
1244 The motivation and syntax for this option is identical to the motivation/syntax
1245 of :ref:`cfg=smpi/bw-factor`.
1247 There is an important difference, though: While smpi/bw-factor `reduces` the
1248 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
1249 increase the latency, i.e., values larger than or equal to 1 are valid here.
1251 .. _cfg=smpi/papi-events:
1253 Trace hardware counters with PAPI
1254 .................................
1256 **Option** ``smpi/papi-events`` **default:** unset
1258 When the PAPI support is compiled into SimGrid, this option takes the
1259 names of PAPI counters and adds their respective values to the trace
1260 files (See Section :ref:`tracing_tracing_options`).
1264 This feature currently requires superuser privileges, as registers
1265 are queried. Only use this feature with code you trust! Call
1266 smpirun for instance via ``smpirun -wrapper "sudo "
1267 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1268 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1269 will not be required.
1271 It is planned to make this feature available on a per-process (or per-thread?) basis.
1272 The first draft, however, just implements a "global" (i.e., for all processes) set
1273 of counters, the "default" set.
1275 .. code-block:: none
1277 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1279 .. _cfg=smpi/privatization:
1281 Automatic Privatization of Global Variables
1282 ...........................................
1284 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1286 MPI executables are usually meant to be executed in separate
1287 processes, but SMPI is executed in only one process. Global variables
1288 from executables will be placed in the same memory region and shared
1289 between processes, causing intricate bugs. Several options are
1290 possible to avoid this, as described in the main `SMPI publication
1291 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1292 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1293 automatically privatizing the globals, and this option allows one to
1294 choose between them.
1296 - **no** (default when not using smpirun): Do not automatically
1297 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1299 - **dlopen** or **yes** (default when using smpirun): Link multiple
1300 times against the binary.
1301 - **mmap** (slower, but maybe somewhat more stable):
1302 Runtime automatic switching of the data segments.
1305 This configuration option cannot be set in your platform file. You can only
1306 pass it as an argument to smpirun.
1308 .. _cfg=smpi/privatize-libs:
1310 Automatic privatization of global variables inside external libraries
1311 .....................................................................
1313 **Option** ``smpi/privatize-libs`` **default:** unset
1315 **Linux/BSD only:** When using dlopen (default) privatization,
1316 privatize specific shared libraries with internal global variables, if
1317 they can't be linked statically. For example libgfortran is usually
1318 used for Fortran I/O and indexes in files can be mixed up.
1320 Multiple libraries can be given, semicolon separated.
1322 This configuration option can only use either full paths to libraries,
1323 or full names. Check with ldd the name of the library you want to
1326 .. code-block:: console
1330 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1333 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1334 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1335 but not ``libgfortran`` nor ``libgfortran.so``.
1337 .. _cfg=smpi/send-is-detached-thresh:
1339 Simulating MPI detached send
1340 ............................
1342 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1344 This threshold specifies the size in bytes under which the send will
1345 return immediately. This is different from the threshold detailed in
1346 :ref:`cfg=smpi/async-small-thresh` because the message is not
1347 really sent when the send is posted. SMPI still waits for the
1348 corresponding receive to be posted, in order to perform the communication
1351 .. _cfg=smpi/coll-selector:
1353 Simulating MPI collective algorithms
1354 ....................................
1356 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1358 SMPI implements more than 100 different algorithms for MPI collective
1359 communication, to accurately simulate the behavior of most of the
1360 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1361 select the decision logic either of the OpenMPI or the MPICH libraries. (By
1362 default SMPI uses naive version of collective operations.)
1364 Each collective operation can be manually selected with a
1365 ``smpi/collective_name:algo_name``. Available algorithms are listed in
1366 :ref:`SMPI_use_colls`.
1368 .. TODO:: All available collective algorithms will be made available
1369 via the ``smpirun --help-coll`` command.
1371 .. _cfg=smpi/finalization-barrier:
1373 Add a barrier in MPI_Finalize
1374 .............................
1376 **Option** ``smpi/finalization-barrier`` **default:** off
1378 By default, SMPI processes are destroyed as soon as soon as their code ends,
1379 so after a successful MPI_Finalize call returns. In some rare cases, some data
1380 might have been attached to MPI objects still active in the remaining processes,
1381 and can be destroyed eagerly by the finished process.
1382 If your code shows issues at finalization, such as segmentation fault, triggering
1383 this option will add an explicit MPI_Barrier(MPI_COMM_WORLD) call inside the
1384 MPI_Finalize, so that all processes will terminate at almost the same point.
1385 It might affect the total timing by the cost of a barrier.
1387 .. _cfg=smpi/errors-are-fatal:
1389 Disable MPI fatal errors
1390 ........................
1392 **Option** ``smpi/errors-are-fatal`` **default:** on
1394 By default, SMPI processes will crash if a MPI error code is returned. MPI allows
1395 to explicitely set MPI_ERRORS_RETURN errhandler to avoid this behaviour. This flag
1396 will turn on this behaviour by default (for all concerned types and errhandlers).
1397 This can ease debugging by going after the first reported error.
1399 .. _cfg=smpi/pedantic:
1401 Disable pedantic MPI errors
1402 ...........................
1404 **Option** ``smpi/pedantic`` **default:** on
1406 By default, SMPI will report all errors it finds in MPI codes. Some of these errors
1407 may not be considered as errors by all developers. This flag can be turned off to
1408 avoid reporting some usually harmless mistakes.
1409 Concerned errors list (will be expanded in the future):
1411 - Calling MPI_Win_fence only once in a program, hence just opening an epoch without
1414 .. _cfg=smpi/iprobe:
1416 Inject constant times for MPI_Iprobe
1417 ....................................
1419 **Option** ``smpi/iprobe`` **default:** 0.0001
1421 The behavior and motivation for this configuration option is identical
1422 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1425 .. _cfg=smpi/iprobe-cpu-usage:
1427 Reduce speed for iprobe calls
1428 .............................
1430 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1432 MPI_Iprobe calls can be heavily used in applications. To account
1433 correctly for the energy that cores spend probing, it is necessary to
1434 reduce the load that these calls cause inside SimGrid.
1436 For instance, we measured a maximum power consumption of 220 W for a
1437 particular application but only 180 W while this application was
1438 probing. Hence, the correct factor that should be passed to this
1439 option would be 180/220 = 0.81.
1443 Inject constant times for MPI_Init
1444 ..................................
1446 **Option** ``smpi/init`` **default:** 0
1448 The behavior and motivation for this configuration option is identical
1449 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1453 Inject constant times for MPI_Isend()
1454 .....................................
1456 **Option** ``smpi/ois``
1458 The behavior and motivation for this configuration option is identical
1459 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1463 Inject constant times for MPI_send()
1464 ....................................
1466 **Option** ``smpi/os``
1468 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1469 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1470 time). SMPI can factor these costs in as well, but the user has to
1471 configure SMPI accordingly as these values may vary by machine. This
1472 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1473 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1474 exactly as ``smpi/ois``.
1476 This item can consist of multiple sections; each section takes three
1477 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1478 so this example contains two sections. Furthermore, each section
1479 consists of three values.
1481 1. The first value denotes the minimum size in bytes for this section to take effect;
1482 read it as "if message size is greater than this value (and other section has a larger
1483 first value that is also smaller than the message size), use this".
1484 In the first section above, this value is "1".
1486 2. The second value is the startup time; this is a constant value that will always
1487 be charged, no matter what the size of the message. In the first section above,
1490 3. The third value is the `per-byte` cost. That is, it is charged for every
1491 byte of the message (incurring cost messageSize*cost_per_byte)
1492 and hence accounts also for larger messages. In the first
1493 section of the example above, this value is "2".
1495 Now, SMPI always checks which section it should use for a given
1496 message; that is, if a message of size 11 is sent with the
1497 configuration of the example above, only the second section will be
1498 used, not the first, as the first value of the second section is
1499 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1500 message of size 11 incurs the following cost inside MPI_Send:
1501 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1503 Note that the order of sections can be arbitrary; they will be ordered internally.
1507 Inject constant times for MPI_Recv()
1508 ....................................
1510 **Option** ``smpi/or``
1512 The behavior and motivation for this configuration option is identical
1513 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1516 .. _cfg=smpi/grow-injected-times:
1518 Inject constant times for MPI_Test
1519 ..................................
1521 **Option** ``smpi/test`` **default:** 0.0001
1523 By setting this option, you can control the amount of time a process
1524 sleeps when MPI_Test() is called; this is important, because SimGrid
1525 normally only advances the time while communication is happening and
1526 thus, MPI_Test will not add to the time, resulting in deadlock if it is
1527 used as a break-condition as in the following example:
1532 MPI_Test(request, flag, status);
1536 To speed up execution, we use a counter to keep track of how often we
1537 checked if the handle is now valid or not. Hence, we actually
1538 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1539 process to sleep increases linearly with the number of previously
1540 failed tests. This behavior can be disabled by setting
1541 ``smpi/grow-injected-times`` to **no**. This will also disable this
1542 behavior for MPI_Iprobe.
1544 .. _cfg=smpi/shared-malloc:
1545 .. _cfg=smpi/shared-malloc-hugepage:
1550 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1552 If your simulation consumes too much memory, you may want to modify
1553 your code so that the working areas are shared by all MPI ranks. For
1554 example, in a block-cyclic matrix multiplication, you will only
1555 allocate one set of blocks, and all processes will share them.
1556 Naturally, this will lead to very wrong results, but this will save a
1557 lot of memory. So this is still desirable for some studies. For more on
1558 the motivation for that feature, please refer to the `relevant section
1559 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1560 of the SMPI CourseWare (see Activity #2.2 of the pointed
1561 assignment). In practice, change the calls for malloc() and free() into
1562 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1564 SMPI provides two algorithms for this feature. The first one, called
1565 ``local``, allocates one block per call to SMPI_SHARED_MALLOC()
1566 (each call site gets its own block) ,and this block is shared
1567 among all MPI ranks. This is implemented with the shm_* functions
1568 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1569 for each shared block.
1571 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1572 returns a new address, but it only points to a shadow block: its memory
1573 area is mapped on a 1 MiB file on disk. If the returned block is of size
1574 N MiB, then the same file is mapped N times to cover the whole block.
1575 At the end, no matter how many times you call SMPI_SHARED_MALLOC, this will
1576 only consume 1 MiB in memory.
1578 You can disable this behavior and come back to regular mallocs (for
1579 example for debugging purposes) using ``no`` as a value.
1581 If you want to keep private some parts of the buffer, for instance if these
1582 parts are used by the application logic and should not be corrupted, you
1583 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). For example:
1587 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1589 This will allocate 500 bytes to mem, such that mem[27..41] and
1590 mem[100..199] are shared while other area remain private.
1592 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1594 When smpi/shared-malloc:global is used, the memory consumption problem
1595 is solved, but it may induce too much load on the kernel's pages table.
1596 In this case, you should use huge pages so that the kernel creates only one
1597 entry per MB of malloced data instead of one entry per 4 kB.
1598 To activate this, you must mount a hugetlbfs on your system and allocate
1599 at least one huge page:
1601 .. code-block:: console
1604 $ sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1605 $ sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1607 Then, you can pass the option
1608 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1609 actually activate the huge page support in shared mallocs.
1611 .. _cfg=smpi/auto-shared-malloc-thresh:
1613 Automatically share allocations
1614 ...............................
1616 **Option** ``smpi/auto-shared-malloc-thresh:`` **Default:** 0 (false)
1617 This value in bytes represents the size above which all allocations
1618 will be "shared" by default (as if they were performed through
1619 SMPI_SHARED_MALLOC macros). Default = 0 = disabled feature.
1620 The value must be carefully chosen to only select data buffers which
1621 will not modify execution path or cause crash if their content is false.
1622 Option :ref:`cfg=smpi/display-allocs` can be used to locate the largest
1623 allocation detected in a run, and provide a good starting threshold.
1624 Note : malloc, calloc and free are overridden by smpicc/cxx by default.
1625 This can cause some troubles if codes are already overriding these. If this
1626 is the case, defining SMPI_NO_OVERRIDE_MALLOC in the compilation flags can
1627 help, but will make this feature unusable.
1631 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1632 ...................................................................
1634 **Option** ``smpi/wtime`` **default:** 10 ns
1636 This option controls the amount of (simulated) time spent in calls to
1637 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1638 to 0, the simulated clock is not advanced in these calls, which leads
1639 to issues if your application contains such a loop:
1643 while(MPI_Wtime() < some_time_bound) {
1644 /* some tests, with no communication nor computation */
1647 When the option smpi/wtime is set to 0, the time advances only on
1648 communications and computations. So the previous code results in an
1649 infinite loop: the current [simulated] time will never reach
1650 ``some_time_bound``. This infinite loop is avoided when that option
1651 is set to a small value, as it is by default since SimGrid v3.21.
1653 Note that if your application does not contain any loop depending on
1654 the current time only, then setting this option to a non-zero value
1655 will slow down your simulations by a tiny bit: the simulation loop has
1656 to be broken out of and reset each time your code asks for the current time.
1657 If the simulation speed really matters to you, you can avoid this
1658 extra delay by setting smpi/wtime to 0.
1660 .. _cfg=smpi/list-leaks:
1662 Report leaked MPI objects
1663 .........................
1665 **Option** ``smpi/list-leaks`` **default:** 0
1667 This option controls whether to report leaked MPI objects.
1668 The parameter is the number of leaks to report.
1670 Other Configurations
1671 --------------------
1673 .. _cfg=debug/clean-atexit:
1675 Cleanup at Termination
1676 ......................
1678 **Option** ``debug/clean-atexit`` **default:** on
1680 If your code is segfaulting during its finalization, it may help to
1681 disable this option to request that SimGrid not attempt any cleanups at
1682 the end of the simulation. Since the Unix process is ending anyway,
1683 the operating system will wipe it all.
1690 **Option** ``path`` **default:** . (current dir)
1692 It is possible to specify a list of directories to search in for the
1693 trace files (see :ref:`pf_trace`) by using this configuration
1694 item. To add several directory to the path, set the configuration
1695 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1697 .. _cfg=debug/breakpoint:
1702 **Option** ``debug/breakpoint`` **default:** unset
1704 This configuration option sets a breakpoint: when the simulated clock
1705 reaches the given time, a SIGTRAP is raised. This can be used to stop
1706 the execution and get a backtrace with a debugger.
1708 It is also possible to set the breakpoint from inside the debugger, by
1709 writing in global variable simgrid::kernel::cfg_breakpoint. For example,
1712 .. code-block:: none
1714 set variable simgrid::kernel::cfg_breakpoint = 3.1416
1716 .. _cfg=debug/verbose-exit:
1721 **Option** ``debug/verbose-exit`` **default:** on
1723 By default, when Ctrl-C is pressed, the status of all existing actors
1724 is displayed before exiting the simulation. This is very useful to
1725 debug your code, but it can become troublesome if you have many
1726 actors. Set this configuration item to **off** to disable this
1729 .. _cfg=exception/cutpath:
1731 Truncate local path from exception backtrace
1732 ............................................
1734 **Option** ``exception/cutpath`` **default:** off
1736 This configuration option is used to remove the path from the
1737 backtrace shown when an exception is thrown. This is mainly useful for
1738 the tests: the full file path would makes the tests non-reproducible because
1739 the paths of source files depend of the build settings. That would
1740 break most of the tests since their output is continually compared.
1744 Logging configuration
1745 ---------------------
1747 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
1748 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
1749 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
1752 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
1753 messages from your code.
1755 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
1756 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
1757 practice, the following is equivalent to the above settings: ``--log=root.thresh:error --log=s4u_host.thresh:debug``.
1759 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
1760 your settings, as in ``--log="root.thresh:error s4u_host.thresh:debug"``. The parameters are interpreted in order, from left to right.
1763 Threshold configuration
1764 .......................
1766 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
1767 ``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
1768 see, ``threshold`` can be abbreviated here.
1770 Existing thresholds:
1772 - ``trace`` some functions display a message at this level when entering or returning
1773 - ``debug`` output that is mostly useful when debugging the corresponding module.
1774 - ``verbose`` verbose output that is only mildly interesting and can easily be ignored
1775 - ``info`` usual output (this is the default threshold of all categories)
1776 - ``warning`` minor issue encountered
1777 - ``error`` issue encountered
1778 - ``critical`` major issue encountered, such as assertions failures
1782 Format configuration
1783 ....................
1785 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
1786 as the date, or the actor ID, everything. Existing format directives:
1789 - %n: line separator (LOG4J compatible)
1790 - %e: plain old space (SimGrid extension)
1792 - %m: user-provided message
1794 - %c: Category name (LOG4J compatible)
1795 - %p: Priority name (LOG4J compatible)
1797 - %h: Hostname (SimGrid extension)
1798 - %a: Actor name (SimGrid extension -- note that with SMPI this is the integer value of the process rank)
1799 - %i: Actor PID (SimGrid extension -- this is a 'i' as in 'i'dea)
1800 - %t: Thread "name" (LOG4J compatible -- actually the address of the thread in memory)
1802 - %F: file name where the log event was raised (LOG4J compatible)
1803 - %l: location where the log event was raised (LOG4J compatible, like '%%F:%%L' -- this is a l as in 'l'etter)
1804 - %L: line number where the log event was raised (LOG4J compatible)
1805 - %M: function name (LOG4J compatible -- called method name here of course).
1807 - %d: date (UNIX-like epoch)
1808 - %r: application age (time elapsed since the beginning of the application)
1811 ``--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
1812 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
1813 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
1814 provided layout is used for every messages.
1816 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
1820 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
1821 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'"``.
1822 Another option is to use the ``%e`` directive for spaces, as in ``--log=root.fmt:%l:%e[%p/%c]:%e%m%n``.
1827 The keyword ``app`` controls the appended of a logging category. For example ``--log=root.app:file:mylogfile`` redirects every output to the file ``mylogfile``.
1829 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
1830 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.
1832 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``
1833 ensures that the log file ``mylog`` will never overpass 500 bytes in size.
1835 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
1836 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.
1841 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
1842 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
1843 ``on`` (or ``yes`` or ``1``), the produced messages will also be passed to the upper appender.
1845 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
1846 ``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
1847 will only be sent to ``all.log``.
1852 ``--help-logs`` displays a complete help message about logging in SimGrid.
1854 ``--help-log-categories`` displays the actual hierarchy of log categories for this binary.
1856 ``--log=no_loc`` hides the source locations (file names and line numbers) from the messages. This is useful to make tests reproducible.