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/concurrency-limit:** :ref:`cfg=maxmin/concurrency-limit`
110 - **model-check:** :ref:`options_modelchecking`
111 - **model-check/checkpoint:** :ref:`cfg=model-check/checkpoint`
112 - **model-check/communications-determinism:** :ref:`cfg=model-check/communications-determinism`
113 - **model-check/dot-output:** :ref:`cfg=model-check/dot-output`
114 - **model-check/max-depth:** :ref:`cfg=model-check/max-depth`
115 - **model-check/property:** :ref:`cfg=model-check/property`
116 - **model-check/reduction:** :ref:`cfg=model-check/reduction`
117 - **model-check/replay:** :ref:`cfg=model-check/replay`
118 - **model-check/send-determinism:** :ref:`cfg=model-check/send-determinism`
119 - **model-check/setenv:** :ref:`cfg=model-check/setenv`
120 - **model-check/sleep-set:** :ref:`cfg=model-check/sleep-set`
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 - **precision/timing:** :ref:`cfg=precision/timing`
144 - **precision/work-amount:** :ref:`cfg=precision/work-amount`
146 - **For collective operations of SMPI,** please refer to Section :ref:`cfg=smpi/coll-selector`
147 - **smpi/auto-shared-malloc-thresh:** :ref:`cfg=smpi/auto-shared-malloc-thresh`
148 - **smpi/async-small-thresh:** :ref:`cfg=smpi/async-small-thresh`
149 - **smpi/barrier-finalization:** :ref:`cfg=smpi/barrier-finalization`
150 - **smpi/barrier-collectives:** :ref:`cfg=smpi/barrier-collectives`
151 - **smpi/buffering:** :ref:`cfg=smpi/buffering`
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/grow-injected-times:** :ref:`cfg=smpi/grow-injected-times`
159 - **smpi/host-speed:** :ref:`cfg=smpi/host-speed`
160 - **smpi/IB-penalty-factors:** :ref:`cfg=smpi/IB-penalty-factors`
161 - **smpi/iprobe:** :ref:`cfg=smpi/iprobe`
162 - **smpi/iprobe-cpu-usage:** :ref:`cfg=smpi/iprobe-cpu-usage`
163 - **smpi/init:** :ref:`cfg=smpi/init`
164 - **smpi/keep-temps:** :ref:`cfg=smpi/keep-temps`
165 - **smpi/ois:** :ref:`cfg=smpi/ois`
166 - **smpi/or:** :ref:`cfg=smpi/or`
167 - **smpi/os:** :ref:`cfg=smpi/os`
168 - **smpi/papi-events:** :ref:`cfg=smpi/papi-events`
169 - **smpi/pedantic:** :ref:`cfg=smpi/pedantic`
170 - **smpi/privatization:** :ref:`cfg=smpi/privatization`
171 - **smpi/privatize-libs:** :ref:`cfg=smpi/privatize-libs`
172 - **smpi/send-is-detached-thresh:** :ref:`cfg=smpi/send-is-detached-thresh`
173 - **smpi/shared-malloc:** :ref:`cfg=smpi/shared-malloc`
174 - **smpi/shared-malloc-hugepage:** :ref:`cfg=smpi/shared-malloc-hugepage`
175 - **smpi/simulate-computation:** :ref:`cfg=smpi/simulate-computation`
176 - **smpi/test:** :ref:`cfg=smpi/test`
177 - **smpi/wtime:** :ref:`cfg=smpi/wtime`
178 - **smpi/list-leaks** :ref:`cfg=smpi/list-leaks`
180 - **Tracing configuration options** can be found in Section :ref:`tracing_tracing_options`
182 - **storage/model:** :ref:`options_model_select`
184 - **vm/model:** :ref:`options_model_select`
188 Configuring the Platform Models
189 -------------------------------
191 .. _options_model_select:
193 Choosing the Platform Models
194 ............................
196 SimGrid comes with several network, CPU and disk models built in,
197 and you can change the used model at runtime by changing the passed
198 configuration. The three main configuration items are given below.
199 For each of these items, passing the special ``help`` value gives you
200 a short description of all possible values (for example,
201 ``--cfg=network/model:help`` will present all provided network
202 models). Also, ``--help-models`` should provide information about all
203 models for all existing resources.
205 - ``network/model``: specify the used network model. Possible values:
207 - **LV08 (default one):** Realistic network analytic model
208 (slow-start modeled by multiplying latency by 13.01, bandwidth by
209 .97; bottleneck sharing uses a payload of S=20537 for evaluating
210 RTT). Described in `Accuracy Study and Improvement of Network
211 Simulation in the SimGrid Framework
212 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
213 - **Constant:** Simplistic network model where all communication
214 take a constant time (one second). This model provides the lowest
215 realism, but is (marginally) faster.
216 - **SMPI:** Realistic network model specifically tailored for HPC
217 settings (accurate modeling of slow start with correction factors on
218 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). This model can be
219 :ref:`further configured <options_model_network>`.
220 - **IB:** Realistic network model specifically tailored for HPC
221 settings with InfiniBand networks (accurate modeling contention
222 behavior, based on the model explained in `this PhD work
223 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
224 This model can be :ref:`further configured <options_model_network>`.
225 - **CM02:** Legacy network analytic model. Very similar to LV08, but
226 without corrective factors. The timings of small messages are thus
227 poorly modeled. This model is described in `A Network Model for
228 Simulation of Grid Application
229 <https://hal.inria.fr/inria-00071989/document>`_.
230 - **ns-3** (only available if you compiled SimGrid accordingly):
231 Use the packet-level network
232 simulators as network models (see :ref:`models_ns3`).
233 This model can be :ref:`further configured <options_pls>`.
235 - ``cpu/model``: specify the used CPU model. We have only one model for now:
237 - **Cas01:** Simplistic CPU model (time=size/speed)
239 - ``host/model``: we have two such models for now.
241 - **default:** Default host model. It simply uses the otherwise configured models for cpu, disk and network (i.e. CPU:Cas01,
242 disk:S19 and network:LV08 by default)
243 - **ptask_L07:** This model is mandatory if you plan to use parallel tasks (and useless otherwise). ptasks are intended to
244 model the moldable tasks of the grid scheduling literature. A specific host model is necessary because each such activity
245 has a both compute and communicate components, so the CPU and network models must be mixed together.
247 - ``storage/model``: specify the used storage model. Only one model is
249 - ``vm/model``: specify the model for virtual machines. Only one model
252 .. todo: make 'compound' the default host model.
254 .. _options_model_solver:
259 The different models rely on a linear inequalities solver to share
260 the underlying resources. SimGrid allows you to change the solver, but
261 be cautious, **don't change it unless you are 100% sure**.
263 - items ``cpu/solver``, ``network/solver``, ``disk/solver`` and ``host/solver``
264 allow you to change the solver for each model:
266 - **maxmin:** The default solver for all models except ptask. Provides a
267 max-min fairness allocation.
268 - **fairbottleneck:** The default solver for ptasks. Extends max-min to
269 allow heterogeneous resources.
270 - **bmf:** More realistic solver for heterogeneous resource sharing.
271 Implements BMF (Bottleneck max fairness) fairness. To be used with
272 parallel tasks instead of fair-bottleneck.
274 .. _options_model_optim:
279 The network and CPU models that are based on linear inequalities solver (that
280 is, all our analytical models) accept specific optimization
283 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
285 - **Lazy:** Lazy action management (partial invalidation in lmm +
286 heap in action remaining).
287 - **TI:** Trace integration. Highly optimized mode when using
288 availability traces (only available for the Cas01 CPU model for
290 - **Full:** Full update of remaining and variables. Slow but may be
291 useful when debugging.
293 - items ``network/maxmin-selective-update`` and
294 ``cpu/maxmin-selective-update``: configure whether the underlying
295 should be lazily updated or not. It should have no impact on the
296 computed timings, but should speed up the computation. |br| It is
297 still possible to disable this feature because it can reveal
298 counter-productive in very specific scenarios where the
299 interaction level is high. In particular, if all your
300 communication share a given backbone link, you should disable it:
301 without it, a simple regular loop is used to update each
302 communication. With it, each of them is still updated (because of
303 the dependency induced by the backbone), but through a complicated
304 and slow pattern that follows the actual dependencies.
306 .. _cfg=bmf/precision:
307 .. _cfg=precision/timing:
308 .. _cfg=precision/work-amount:
313 **Option** ``precision/timing`` **Default:** 1e-9 (in seconds) |br|
314 **Option** ``precision/work-amount`` **Default:** 1e-5 (in flops or bytes) |br|
315 **Option** ``bmf/precision`` **Default:** 1e-12 (no unit)
317 The analytical models handle a lot of floating point values. It is
318 possible to change the epsilon used to update and compare them through
319 this configuration item. Changing it may speedup the simulation by
320 discarding very small actions, at the price of a reduced numerical
321 precision. You can modify separately the precision used to manipulate
322 timings (in seconds) and the one used to manipulate amounts of work
325 .. _cfg=maxmin/concurrency-limit:
330 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
332 The maximum number of variables per resource can be tuned through this
333 option. You can have as many simultaneous actions per resources as you
334 want. If your simulation presents a very high level of concurrency, it
335 may help to use e.g. 100 as a value here. It means that at most 100
336 actions can consume a resource at a given time. The extraneous actions
337 are queued and wait until the amount of concurrency of the considered
338 resource lowers under the given boundary.
340 Such limitations help both to the simulation speed and simulation accuracy
341 on highly constrained scenarios, but the simulation speed suffers of this
342 setting on regular (less constrained) scenarios so it is off by default.
344 .. _cfg=bmf/max-iterations:
349 **Option** ``bmf/max-iterations`` **Default:** 1000
351 It may happen in some settings that the BMF solver fails to converge to
352 a solution, so there is a hard limit on the amount of iteration count to
353 avoid infinite loops.
355 .. _options_model_network:
357 Configuring the Network Model
358 .............................
360 .. _cfg=network/TCP-gamma:
362 Maximal TCP Window Size
363 ^^^^^^^^^^^^^^^^^^^^^^^
365 **Option** ``network/TCP-gamma`` **Default:** 4194304
367 The analytical models need to know the maximal TCP window size to take the TCP congestion mechanism into account (see
368 :ref:`this page <understanding_cm02>` for details). On Linux, this value can be retrieved using the following commands.
369 Both give a set of values, and you should use the last one, which is the maximal size.
371 .. code-block:: console
373 $ cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
374 $ cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
376 If you want to disable the TCP windowing mechanism, set this parameter to 0.
378 .. _cfg=network/bandwidth-factor:
379 .. _cfg=network/latency-factor:
380 .. _cfg=network/weight-S:
382 Manual calibration factors
383 ^^^^^^^^^^^^^^^^^^^^^^^^^^
385 SimGrid can take network irregularities such as a slow startup or changing behavior depending on the message size into account.
386 The values provided by default were computed a long time ago through data fitting one the timings of either packet-level
387 simulators or direct experiments on real platforms. These default values should be OK for most users, but if simulation realism
388 is really important to you, you probably want to recalibrate the models (i.e., devise sensible values for your specific
389 settings). This section only describes how to pass new values to the models while the calibration process involved in the
390 computation of these values is described :ref:`in the relevant chapter <models_calibration>`.
392 We found out that many networking effects can be realistically accounted for with the three following correction factors. They
393 were shown to be enough to capture slow-start effects, the different transmission modes of MPI systems (eager vs. rendez-vous
394 mode), or the non linear effects of wifi sharing.
396 **Option** ``network/latency-factor`` **Default:** 1.0, but overridden by most models
398 This option specifies a multiplier to apply to the *physical* latency (i.e., the one described in the platform) of the set of
399 links involved in a communication. The factor can either be a constant to apply to any communication, or it may depend on the
400 message size. The ``CM02`` model does not use any correction factor, so the latency-factor remains to 1. The ``LV08`` model sets
401 it to 13.01 to model slow-start, while the ``SMPI`` model has several possible values depending on the interval in which the
402 message size falls. The default SMPI setting given below specifies for example that a message smaller than 257 bytes will get a
403 latency multiplier of 2.01467 while a message whose size is in [15424, 65472] will get a latency multiplier of 3.48845. The
404 ``wifi`` model goes further and uses a callback in the program to compute the factor that must be non-linear in this case.
406 This multiplier is applied to the latency computed from the platform, that is the sum of all link *physical* latencies over the
407 :ref:`network path <platform_routing>` used by the considered communication, to derive the *effective* end-to-end latency.
409 Constant factors are easy to express, but the interval-based syntax used in SMPI is somewhat complex. It expects a set of
410 factors separated by semicolons, each of the form ``boundary:factor``. For example if your specification is
411 ``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
412 5000 and beyond. If your first interval does include size=0, then the default value of 1 is used before. Changing the factor
413 callback is not possible from the command line and must be done from your code, as shown in `this example
414 <https://framagit.org/simgrid/simgrid/tree/master/examples/cpp/network-factors/s4u-network-factors.cpp>`_. Note that the chosen
415 model only provides some default settings. You may pick a ``LV08`` model to get some of the settings, and override the latency
416 with interval-based values.
418 SMPI default value: 65472:11.6436; 15424:3.48845; 9376:2.59299; 5776:2.18796; 3484:1.88101; 1426:1.61075; 732:1.9503;
419 257:1.95341;0:2.01467 (interval boundaries are sorted automatically). These values were computed by data fitting on the Stampede
420 Supercomputer at TACC, with optimal deployment of processes on nodes. To accurately model your settings, you should redo the
421 :ref:`calibration <models_calibration>`.
423 **Option** ``network/bandwidth-factor`` **Default:** 1.0, but overridden by most models
425 Setting this option automatically adjusts the *effective* bandwidth (i.e., the one perceived by the application) used by any
426 given communication. As with latency-factor above, the value can be a constant (``CM02`` uses 1 -- no correction -- while
427 ``LV08`` uses 0.97 to discount TCP headers while computing the payload bandwidth), interval-based (as the default provided by
428 the ``SMPI``), or using in-program callbacks (as with ``wifi``).
430 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
431 This was also computed on the Stampede Supercomputer.
433 **Option** ``network/weight-S`` **Default:** depends on the model
435 Value used to account for RTT-unfairness when sharing a bottleneck (network connections with a large RTT are generally penalized
436 against those with a small one). See :ref:`models_TCP` and also this scientific paper: `Accuracy Study and Improvement of Network
437 Simulation in the SimGrid Framework <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_
439 Default values for ``CM02`` is 0. ``LV08`` sets it to 20537 while both ``SMPI`` and ``IB`` set it to 8775.
441 .. _cfg=network/loopback:
443 Configuring loopback link
444 ^^^^^^^^^^^^^^^^^^^^^^^^^
446 Several network models provide an implicit loopback link to account for local
447 communication on a host. By default it has a 10GBps bandwidth and a null latency.
448 This can be changed with ``network/loopback-lat`` and ``network/loopback-bw``
449 items. Note that this loopback is conveniently modeled with a :ref:`single FATPIPE link <pf_loopback>`
450 for the whole platform. If modeling contention inside nodes is important then you should
451 rather add such loopback links (one for each host) yourself.
453 .. _cfg=smpi/IB-penalty-factors:
458 InfiniBand network behavior can be modeled through 3 parameters
459 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `the PhD
460 thesis of Jérôme Vienne
461 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_ (in French)
462 or more concisely in `this paper <https://hal.inria.fr/hal-00953618/document>`_,
463 even if that paper does only describe models for myrinet and ethernet.
464 You can see in Fig 2 some results for Infiniband, for example. This model
465 may be outdated by now for modern infiniband, anyway, so a new
466 validation would be good.
468 The three paramaters are defined as follows:
470 - βs: penalty factor for outgoing messages, computed by running a simple send to
471 two nodes and checking slowdown compared to a single send to one node,
473 - βe: penalty factor for ingoing messages, same computation method but with one
474 node receiving several messages
475 - γr: slowdown factor when communication buffer memory is saturated. It needs a
476 more complicated pattern to run in order to be computed (5.3 in the thesis,
477 page 107), and formula in the end is γr = time(c)/(3×βe×time(ref)), where
478 time(ref) is the time of a single comm with no contention).
480 Once these values are computed, a penalty is assessed for each message (this is
481 the part implemented in the simulator) as shown page 106 of the thesis. Here is
482 a simple translation of this text. First, some notations:
484 - ∆e(e) which corresponds to the incoming degree of node e, that is to say the number of communications having as destination node e.
485 - ∆s (s) which corresponds to the degree outgoing from node s, that is to say the number of communications sent by node s.
486 - Φ (e) which corresponds to the number of communications destined for the node e but coming from a different node.
487 - 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
489 To determine the penalty for a communication, two values need to be calculated. First, the penalty caused by the conflict in transmission, noted ps.
492 - if ∆s (i) = 1 then ps = 1.
493 - if ∆s (i) ≥ 2 and ∆e (i) ≥ 3 then ps = ∆s (i) × βs × γr
494 - else, ps = ∆s (i) × βs
497 Then, the penalty caused by the conflict in reception (noted pe) should be computed as follows:
499 - if ∆e (i) = 1 then pe = 1
500 - else, pe = Φ (e) × βe × Ω (s, e)
502 Finally, the penalty associated with the communication is:
503 p = max (ps ∈ s, pe)
505 .. _cfg=network/crosstraffic:
507 Simulating Cross-Traffic
508 ^^^^^^^^^^^^^^^^^^^^^^^^
510 Since SimGrid v3.7, cross-traffic effects can be taken into account in
511 analytical simulations. It means that ongoing and incoming
512 communication flows are treated independently. In addition, the LV08
513 model adds 0.05 of usage on the opposite direction for each new
514 created flow. This can be useful to simulate some important TCP
515 phenomena such as ack compression.
517 For that to work, your platform must have two links for each
518 pair of interconnected hosts. An example of usable platform is
519 available in ``examples/platforms/crosstraffic.xml``.
521 This is activated through the ``network/crosstraffic`` item, that
522 can be set to 0 (disable this feature) or 1 (enable it).
524 Note that with the default host model this option is activated by default.
526 .. _cfg=smpi/async-small-thresh:
528 Simulating Asynchronous Send
529 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
531 (this configuration item is experimental and may change or disappear)
533 It is possible to specify that messages below a certain size (in bytes) will be
534 sent as soon as the call to MPI_Send is issued, without waiting for
535 the correspondent receive. This threshold can be configured through
536 the ``smpi/async-small-thresh`` item. The default value is 0. This
537 behavior can also be manually set for mailboxes, by setting the
538 receiving mode of the mailbox with a call to
539 :cpp:func:`sg_mailbox_set_receiver`. After this, all messages sent to
540 this mailbox will have this behavior regardless of the message size.
542 This value needs to be smaller than or equals to the threshold set at
543 :ref:`cfg=smpi/send-is-detached-thresh`, because asynchronous messages
544 are meant to be detached as well.
551 **Option** ``ns3/NetworkModel`` **Default:** "default" (ns-3 default TCP)
553 When using ns-3, the item ``ns3/NetworkModel`` can be used to switch between TCP or UDP, and switch the used TCP variante. If
554 the item is left unchanged, ns-3 uses the default TCP implementation. With a value of "UDP", ns-3 is set to use UDP instead.
555 With the value of either 'NewReno' or 'Cubic', the ``ns3::TcpL4Protocol::SocketType`` configuration item in ns-3 is set to the
556 corresponding protocol.
558 **Option** ``ns3/seed`` **Default:** "" (don't set the seed in ns-3)
560 This option is the random seed to provide to ns-3 with
561 ``ns3::RngSeedManager::SetSeed`` and ``ns3::RngSeedManager::SetRun``.
563 If left blank, no seed is set in ns-3. If the value 'time' is
564 provided, the current amount of seconds since epoch is used as a seed.
565 Otherwise, the provided value must be a number to use as a seed.
567 Configuring the Storage model
568 .............................
570 .. _cfg=storage/max_file_descriptors:
572 File Descriptor Count per Host
573 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
575 **Option** ``storage/max_file_descriptors`` **Default:** 1024
577 Each host maintains a fixed-size array of its file descriptors. You
578 can change its size through this item to either enlarge it if your
579 application requires it or to reduce it to save memory space.
586 SimGrid plugins allow one to extend the framework without changing its
587 source code directly. Read the source code of the existing plugins to
588 learn how to do so (in ``src/plugins``), and ask your questions to the
589 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
590 that plugins usually register callbacks to some signals of interest.
591 If they need to store some information about a given object (Link, CPU
592 or Actor), they do so through the use of a dedicated object extension.
594 Some of the existing plugins can be activated from the command line,
595 meaning that you can activate them from the command line without any
596 modification to your simulation code. For example, you can activate
597 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
600 Here is a partial list of plugins that can be activated this way. You can get
601 the full list by passing ``--cfg=plugin:help`` to your simulator.
603 - :ref:`Host Energy <plugin_host_energy>`: models the energy dissipation of the compute units.
604 - :ref:`Link Energy <plugin_link_energy>`: models the energy dissipation of the network.
605 - :ref:`Host Load <plugin_host_load>`: monitors the load of the compute units.
607 .. _options_modelchecking:
609 Configuring the Model-Checking
610 ------------------------------
612 To enable SimGrid's model-checking support, the program should
613 be executed using the simgrid-mc wrapper:
615 .. code-block:: console
617 $ simgrid-mc ./my_program
619 Safety properties are expressed as assertions using the function
620 :cpp:func:`void MC_assert(int prop)`.
622 .. _cfg=smpi/buffering:
624 Specifying the MPI buffering behavior
625 .....................................
627 **Option** ``smpi/buffering`` **Default:** infty
629 Buffering in MPI has a huge impact on the communication semantic. For example,
630 standard blocking sends are synchronous calls when the system buffers are full
631 while these calls can complete immediately without even requiring a matching
632 receive call for small messages sent when the system buffers are empty.
634 In SMPI, this depends on the message size, that is compared against two thresholds:
636 - if (size < :ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>`) then
637 MPI_Send returns immediately, and the message is sent even if the
638 corresponding receive has not be issued yet. This is known as the eager mode.
639 - if (:ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>` < size <
640 :ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>`) then
641 MPI_Send also returns immediately, but SMPI waits for the corresponding
642 receive to be posted, in order to perform the communication operation.
643 - if (:ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>` < size) then
644 MPI_Send returns only when the message has actually been sent over the network. This is known as the rendez-vous mode.
646 The ``smpi/buffering`` (only valid with MC) option gives an easier interface to choose between these semantics. It can take two values:
648 - **zero:** means that buffering should be disabled. All communications are actually blocking.
649 - **infty:** means that buffering should be made infinite. All communications are non-blocking.
651 .. _cfg=model-check/property:
653 Specifying a liveness property
654 ..............................
656 **Option** ``model-check/property`` **Default:** unset
658 If you want to specify liveness properties, you have to pass them on
659 the command line, specifying the name of the file containing the
660 property, as formatted by the `ltl2ba <https://github.com/utwente-fmt/ltl2ba>`_ program.
661 Note that ltl2ba is not part of SimGrid and must be installed separately.
663 .. code-block:: console
665 $ simgrid-mc ./my_program --cfg=model-check/property:<filename>
667 .. _cfg=model-check/checkpoint:
669 Going for Stateful Verification
670 ...............................
672 By default, the system is backtracked to its initial state to explore
673 another path, instead of backtracking to the exact step before the fork
674 that we want to explore (this is called stateless verification). This
675 is done this way because saving intermediate states can rapidly
676 exhaust the available memory. If you want, you can change the value of
677 the ``model-check/checkpoint`` item. For example,
678 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
679 step. Beware, this will certainly explode your memory. Larger values
680 are probably better, make sure to experiment a bit to find the right
681 setting for your specific system.
683 .. _cfg=model-check/reduction:
685 Specifying the kind of reduction
686 ................................
688 The main issue when using the model-checking is the state space
689 explosion. You can activate some reduction technique with
690 ``--cfg=model-check/reduction:<technique>``. For now, this
691 configuration variable can take 2 values:
693 - **none:** Do not apply any kind of reduction (mandatory for liveness properties, as our current DPOR algorithm breaks cycles)
694 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if you verify local safety properties (default value for
696 - **sdpor:** Source-set DPOR, as described in "Source Sets: A Foundation for Optimal Dynamic Partial Order Reduction"
698 - **odpor:** Optimal DPOR, as described in "Source Sets: A Foundation for Optimal Dynamic Partial Order Reduction"
701 Another way to mitigate the state space explosion is to search for
702 cycles in the exploration with the :ref:`cfg=model-check/visited`
703 configuration. Note that DPOR and state-equality reduction may not
704 play well together. You should choose between them.
706 Our current DPOR implementation could be improved in may ways. We are
707 currently improving its efficiency (both in term of reduction ability
708 and computational speed), and future work could make it compatible
709 with liveness properties.
711 .. _cfg=model-check/sleep-set:
716 The performance of the DPOR algorithm is greatly improved by using sleep sets, but if you want, you can still disable it with
719 .. _cfg=model-check/visited:
721 Size of Cycle Detection Set (state equality reduction)
722 ......................................................
724 Mc SimGrid can be asked to search for cycles during the exploration,
725 i.e. situations where a new explored state is in fact the same state
726 than a previous one.. This can prove useful to mitigate the state
727 space explosion with safety properties, and this is the crux when
728 searching for counter-examples to the liveness properties.
730 Note that this feature may break the current implementation of the
731 DPOR reduction technique.
733 The ``model-check/visited`` item is the maximum number of states, which
734 are stored in memory. If the maximum number of snapshotted state is
735 reached, some states will be removed from the memory and some cycles
736 might be missed. Small values can lead to incorrect verifications, but
737 large values can exhaust your memory and be CPU intensive as each new
738 state must be compared to that amount of older saved states.
740 The default settings depend on the kind of exploration. With safety
741 checking, no state is snapshotted and cycles cannot be detected. With
742 liveness checking, all states are snapshotted because missing a cycle
743 could hinder the exploration soundness.
745 .. _cfg=model-check/termination:
747 Non-Termination Detection
748 .........................
750 The ``model-check/termination`` configuration item can be used to
751 report if a non-termination execution path has been found. This is a
752 path with a cycle, which means that the program might never terminate.
754 This only works in safety mode, not in liveness mode.
756 This options is disabled by default.
758 .. _cfg=model-check/dot-output:
763 If set, the ``model-check/dot-output`` configuration item is the name
764 of a file in which to write a dot file of the path leading to the
765 property violation discovered (safety or liveness violation), as well
766 as the cycle for liveness properties. This dot file can then be fed to the
767 graphviz dot tool to generate a corresponding graphical representation.
769 .. _cfg=model-check/max-depth:
771 Exploration Depth Limit
772 .......................
774 The ``model-check/max-depth`` can set the maximum depth of the
775 exploration graph of the model checker. If this limit is reached, a
776 logging message is sent and the results might not be exact.
778 By default, the exploration is limited to the depth of 1000.
780 .. _cfg=model-check/timeout:
785 By default, the model checker does not handle timeout conditions: the `wait`
786 operations never time out. With the ``model-check/timeout`` configuration item
787 set to **yes**, the model checker will explore timeouts of `wait` operations.
789 .. _cfg=model-check/communications-determinism:
790 .. _cfg=model-check/send-determinism:
792 Communication Determinism
793 .........................
795 The ``model-check/communications-determinism`` and
796 ``model-check/send-determinism`` items can be used to select the
797 communication determinism mode of the model checker, which checks
798 determinism properties of the communications of an application.
800 .. _cfg=model-check/setenv:
802 Passing environment variables
803 .............................
805 You can specify extra environment variables to be set in the verified application
806 with ``model-check/setenv``. For example, you can preload a library as follows:
807 ``-cfg=model-check/setenv:LD_PRELOAD=toto;LD_LIBRARY_PATH=/tmp``.
811 Verification Performance Considerations
812 .......................................
814 The size of the stacks can have a huge impact on the memory
815 consumption when using model-checking. By default, each snapshot will
816 save a copy of the whole stacks and not only of the part that is
817 really meaningful: you should expect the contribution of the memory
818 consumption of the snapshots to be:
819 :math:`\text{number of processes} \times \text{stack size} \times \text{number of states}`.
821 When compiled against the model checker, the stacks are not
822 protected with guards: if the stack size is too small for your
823 application, the stack will silently overflow into other parts of the
824 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
826 .. _cfg=model-check/replay:
828 Replaying buggy execution paths from the model checker
829 ......................................................
831 Debugging the problems reported by the model checker is challenging:
832 First, the application under verification cannot be debugged with gdb
833 because the model checker already traces it. Then, the model checker may
834 explore several execution paths before encountering the issue, making it
835 very difficult to understand the output. Fortunately, SimGrid provides
836 the execution path leading to any reported issue so that you can replay
837 this path reported by the model checker, enabling the usage of classical
840 When the model checker finds an interesting path in the application
841 execution graph (where a safety or liveness property is violated), it
842 generates an identifier for this path. Here is an example of the output:
844 .. code-block:: console
846 [ 0.000000] (0:@) Check a safety property
847 [ 0.000000] (0:@) **************************
848 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
849 [ 0.000000] (0:@) **************************
850 [ 0.000000] (0:@) Counter-example execution trace:
851 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
852 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
853 [ 0.000000] (0:@) Path = 1/3;1/4
854 [ 0.000000] (0:@) Expanded states = 27
855 [ 0.000000] (0:@) Visited states = 68
856 [ 0.000000] (0:@) Executed transitions = 46
858 The interesting line is ``Path = 1/3;1/4``, which means that you should use
859 ``--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
860 execution path. All options (but the model checker related ones) must
861 remain the same. In particular, if you ran your application with
862 ``smpirun -wrapper simgrid-mc``, then do it again. Remove all
863 MC-related options, keep non-MC-related ones and add
864 ``--cfg=model-check/replay:???``.
866 Currently, if the path is of the form ``X;Y;Z``, each number denotes
867 the actor's pid that is selected at each indecision point. If it's of
868 the form ``X/a;Y/b``, the X and Y are the selected pids while the a
869 and b are the return values of their simcalls. In the previous
870 example, ``1/3;1/4``, you can see from the full output that the actor
871 1 is doing MC_RANDOM simcalls, so the 3 and 4 simply denote the values
872 that these simcall return on the execution branch leading to the
875 Configuring the User Code Virtualization
876 ----------------------------------------
878 .. _cfg=contexts/factory:
880 Selecting the Virtualization Factory
881 ....................................
883 **Option** contexts/factory **Default:** "raw"
885 In SimGrid, the user code is virtualized in a specific mechanism that
886 allows the simulation kernel to control its execution: when a user
887 process requires a blocking action (such as sending a message), it is
888 interrupted, and only gets released when the simulated clock reaches
889 the point where the blocking operation is done. This is explained
890 graphically in the `relevant tutorial, available online
891 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
893 In SimGrid, the containers in which user processes are virtualized are
894 called contexts. Several context factory are provided, and you can
895 select the one you want to use with the ``contexts/factory``
896 configuration item. Some of the following may not exist on your
897 machine because of portability issues. In any case, the default one
898 should be the most effcient one (please report bugs if the
899 auto-detection fails for you). They are approximately sorted here from
900 the slowest to the most efficient:
902 - **thread:** very slow factory using full featured, standard threads.
903 They are slow but very standard. Some debuggers or profilers only work with this factory.
904 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
905 - **boost:** This uses the `context
906 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
907 of the boost library for a performance that is comparable to our
909 |br| Install the relevant library (e.g. with the
910 libboost-contexts-dev package on Debian/Ubuntu) and recompile
912 - **raw:** amazingly fast factory using a context switching mechanism
913 of our own, directly implemented in assembly (only available for x86
914 and amd64 platforms for now) and without any unneeded system call.
916 The main reason to change this setting is when the debugging tools become
917 fooled by the optimized context factories. Threads are the most
918 debugging-friendly contexts, as they allow one to set breakpoints
919 anywhere with gdb and visualize backtraces for all processes, in order
920 to debug concurrency issues. Valgrind is also more comfortable with
921 threads, but it should be usable with all factories (Exception: the
922 callgrind tool really dislikes raw and ucontext factories).
924 .. _cfg=contexts/stack-size:
926 Adapting the Stack Size
927 .......................
929 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
931 Each virtualized used process is executed using a specific system
932 stack. The size of this stack has a huge impact on the simulation
933 scalability, but its default value is rather large. This is because
934 the error messages that you get when the stack size is too small are
935 rather disturbing: this leads to stack overflow (overwriting other
936 stacks), leading to segfaults with corrupted stack traces.
938 If you want to push the scalability limits of your code, you might
939 want to reduce the ``contexts/stack-size`` item. Its default value is
940 8192 (in KiB), while our Chord simulation works with stacks as small
941 as 16 KiB, for example. You can ensure that some actors have a specific
942 size by simply changing the value of this configuration item before
943 creating these actors. The :cpp:func:`simgrid::s4u::Engine::set_config`
944 functions are handy for that.
946 This *setting is ignored* when using the thread factory (because there
947 is no way to modify the stack size with C++ system threads). Instead,
948 you should compile SimGrid and your application with
949 ``-fsplit-stack``. Note that this compilation flag is not compatible
950 with the model checker right now.
952 The operating system should only allocate memory for the pages of the
953 stack which are actually used and you might not need to use this in
954 most cases. However, this setting is very important when using the
955 model checker (see :ref:`options_mc_perf`).
957 .. _cfg=contexts/guard-size:
959 Disabling Stack Guard Pages
960 ...........................
962 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages with MC)
964 Unless you use the threads context factory (see
965 :ref:`cfg=contexts/factory`), a stack guard page is usually used
966 which prevents the stack of a given actor from overflowing on another
967 stack. But the performance impact may become prohibitive when the
968 amount of actors increases. The option ``contexts/guard-size`` is the
969 number of stack guard pages used. By setting it to 0, no guard pages
970 will be used: in this case, you should avoid using small stacks (with
971 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
972 will silently overflow on other parts of the memory.
974 When no stack guard page is created, stacks may then silently overflow
975 on other parts of the memory if their size is too small for the
978 .. _cfg=contexts/nthreads:
979 .. _cfg=contexts/synchro:
981 Running User Code in Parallel
982 .............................
984 Parallel execution of the user code is only considered stable in
985 SimGrid v3.7 and higher, and mostly for S4U simulations. SMPI
986 simulations may well fail in parallel mode. It is described in
987 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
989 If you are using the **ucontext** or **raw** context factories, you can
990 request to execute the user code in parallel. Several threads are
991 launched, each of them handling the same number of user contexts at each
992 run. To activate this, set the ``contexts/nthreads`` item to the amount
993 of cores that you have in your computer (or lower than 1 to have the
994 amount of cores auto-detected).
996 When parallel execution is activated, you can choose the
997 synchronization schema used with the ``contexts/synchro`` item,
998 which value is either:
1000 - **futex:** ultra optimized synchronisation schema, based on futexes
1001 (fast user-mode mutexes), and thus only available on Linux systems.
1002 This is the default mode when available.
1003 - **posix:** slow but portable synchronisation using only POSIX
1005 - **busy_wait:** not really a synchronisation: the worker threads
1006 constantly request new contexts to execute. It should be the most
1007 efficient synchronisation schema, but it loads all the cores of
1008 your machine for no good reason. You probably prefer the other less
1011 Configuring the Tracing
1012 -----------------------
1014 The :ref:`tracing subsystem <outcome_vizu>` can be configured in
1015 several different ways depending on the used interface (S4U, SMPI)
1016 and the kind of traces that needs to be obtained. See the
1017 :ref:`Tracing Configuration Options subsection
1018 <tracing_tracing_options>` for a full description of each
1019 configuration option.
1021 We detail here a simple way to get the traces working for you, even if
1022 you never used the tracing API.
1025 - Any SimGrid-based simulator (S4U, SMPI, ...) and raw traces:
1027 .. code-block:: none
1029 --cfg=tracing:yes --cfg=tracing/uncategorized:yes
1031 The first parameter activates the tracing subsystem, and the second
1032 tells it to trace host and link utilization (without any
1035 - S4U-based simulator and categorized traces (you need to
1036 declare categories and classify your tasks according to them)
1038 .. code-block:: none
1040 --cfg=tracing:yes --cfg=tracing/categorized:yes
1042 The first parameter activates the tracing subsystem, and the second
1043 tells it to trace host and link categorized utilization.
1045 - SMPI simulator and traces for a space/time view:
1047 .. code-block:: console
1049 $ smpirun -trace ...
1051 The `-trace` parameter for the smpirun script runs the simulation
1052 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
1053 smpirun's `-help` parameter for additional tracing options.
1055 Sometimes you might want to put additional information on the trace to
1056 correctly identify them later, or to provide data that can be used to
1057 reproduce an experiment. You have two ways to do that:
1059 - Add a string on top of the trace file as comment:
1061 .. code-block:: none
1063 --cfg=tracing/comment:my_simulation_identifier
1065 - Add the contents of a textual file on top of the trace file as comment:
1067 .. code-block:: none
1069 --cfg=tracing/comment-file:my_file_with_additional_information.txt
1071 Please, use these two parameters (for comments) to make reproducible
1072 simulations. For additional details about this and all tracing
1073 options, check See the :ref:`tracing_tracing_options`.
1078 The SMPI interface provides several specific configuration items.
1079 These are not easy to see with ``--help-cfg``, since SMPI binaries are usually launched through the ``smiprun`` script.
1081 .. _cfg=smpi/host-speed:
1082 .. _cfg=smpi/cpu-threshold:
1083 .. _cfg=smpi/simulate-computation:
1085 Automatic Benchmarking of SMPI Code
1086 ...................................
1088 In SMPI, the sequential code is automatically benchmarked, and these
1089 computations are automatically reported to the simulator. That is to
1090 say that if you have a large computation between a ``MPI_Recv()`` and
1091 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
1092 this code, and create an execution task within the simulator to take
1093 this into account. For that, the actual duration is measured on the
1094 host machine and then scaled to the power of the corresponding
1095 simulated machine. The variable ``smpi/host-speed`` allows one to
1096 specify the computational speed of the host machine (in flop/s by
1097 default) to use when scaling the execution times.
1099 The default value is ``smpi/host-speed=20kf`` (= 20,000 flop/s). This
1100 is probably underestimated for most machines, leading SimGrid to
1101 overestimate the amount of flops in the execution blocks that are
1102 automatically injected in the simulator. As a result, the execution
1103 time of the whole application will probably be overestimated until you
1104 use a realistic value.
1106 When the code consists of numerous consecutive MPI calls, the
1107 previous mechanism feeds the simulation kernel with numerous tiny
1108 computations. The ``smpi/cpu-threshold`` item becomes handy when this
1109 impacts badly on the simulation performance. It specifies a threshold (in
1110 seconds) below which the execution chunks are not reported to the
1111 simulation kernel (default value: 1e-6).
1113 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
1114 time spent below this threshold. SMPI does not consider the
1115 `amount of time` of these computations; there is no offset for
1116 this. Hence, a value that is too small, may lead to unreliable
1119 In some cases, however, one may wish to disable simulation of
1120 the computation of an application. This is the case when SMPI is used not to
1121 simulate an MPI application, but instead an MPI code that performs
1122 "live replay" of another MPI app (e.g., ScalaTrace's replay tool, or
1123 various on-line simulators that run an app at scale). In this case the
1124 computation of the replay/simulation logic should not be simulated by
1125 SMPI. Instead, the replay tool or on-line simulator will issue
1126 "computation events", which correspond to the actual MPI simulation
1127 being replayed/simulated. At the moment, these computation events can
1128 be simulated using SMPI by calling internal smpi_execute*() functions.
1130 To disable the benchmarking/simulation of a computation in the simulated
1131 application, the variable ``smpi/simulate-computation`` should be set
1132 to **no**. This option just ignores the timings in your simulation; it
1133 still executes the computations itself. If you want to stop SMPI from
1134 doing that, you should check the SMPI_SAMPLE macros, documented in
1135 Section :ref:`SMPI_use_faster`.
1137 +------------------------------------+-------------------------+-----------------------------+
1138 | Solution | Computations executed? | Computations simulated? |
1139 +====================================+=========================+=============================+
1140 | --cfg=smpi/simulate-computation:no | Yes | Never |
1141 +------------------------------------+-------------------------+-----------------------------+
1142 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
1143 +------------------------------------+-------------------------+-----------------------------+
1144 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
1145 +------------------------------------+-------------------------+-----------------------------+
1147 .. _cfg=smpi/comp-adjustment-file:
1149 Slow-down or speed-up parts of your code
1150 ........................................
1152 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
1154 This option allows you to pass a file that contains two columns: The
1155 first column defines the section that will be subject to a speedup;
1156 the second column is the speedup. For instance:
1158 .. code-block:: none
1160 "start:stop","ratio"
1161 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1163 The first line is the header - you must include it. The following
1164 line means that the code between two consecutive MPI calls on line 30
1165 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1166 of 1.18244559422142. The value for the second column is therefore a
1167 speedup, if it is larger than 1 and a slowdown if it is smaller
1168 than 1. Nothing will be changed if it is equal to 1.
1170 Of course, you can set any arbitrary filenames you want (so the start
1171 and end don't have to be in the same file), but be aware that this
1172 mechanism only supports `consecutive calls!`
1174 Please note that you must pass the ``-trace-call-location`` flag to
1175 smpicc or smpiff, respectively. This flag activates some internal
1176 macro definitions that help with obtaining the call location.
1178 Bandwidth and latency factors
1179 .............................
1181 Adapting the bandwidth and latency acurately to the network conditions is of a paramount importance to get realistic results.
1182 This is done through the :ref:`network/bandwidth-factor <cfg=network/bandwidth-factor>` and :ref:`network/latency-factor
1183 <cfg=network/latency-factor>` items. You probably also want to read the following section: :ref:`models_calibration`.
1185 .. _cfg=smpi/display-timing:
1187 Reporting Simulation Time
1188 .........................
1190 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1192 Most of the time, you run MPI code with SMPI to compute the time it
1193 would take to run it on a platform. But since the code is run through
1194 the ``smpirun`` script, you don't have any control on the launcher
1195 code, making it difficult to report the simulated time when the
1196 simulation ends. If you enable the ``smpi/display-timing`` item,
1197 ``smpirun`` will display this information when the simulation
1199 SMPI will also display information about the amout of real time spent
1200 in application code and in SMPI internals, to provide hints about the
1201 need to use sampling to reduce simulation time.
1203 .. _cfg=smpi/display-allocs:
1205 Reporting memory allocations
1206 ............................
1208 **Option** ``smpi/display-allocs`` **Default:** 0 (false)
1210 SMPI intercepts malloc and calloc calls performed inside the running
1211 application, if it wasn't compiled with SMPI_NO_OVERRIDE_MALLOC.
1212 With this option, SMPI will show at the end of execution the amount of
1213 memory allocated through these calls, and locate the most expensive one.
1214 This helps finding the targets for manual memory sharing, or the threshold
1215 to use for smpi/auto-shared-malloc-thresh option (see :ref:`cfg=smpi/auto-shared-malloc-thresh`).
1217 .. _cfg=smpi/keep-temps:
1219 Keeping temporary files after simulation
1220 ........................................
1222 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1224 SMPI usually generates a lot of temporary files that are cleaned after
1225 use. This option requests to preserve them, for example to debug or
1226 profile your code. Indeed, the binary files are removed very early
1227 under the dlopen privatization schema, which tends to fool the
1230 .. _cfg=smpi/papi-events:
1232 Trace hardware counters with PAPI
1233 .................................
1235 **Option** ``smpi/papi-events`` **default:** unset
1237 When the PAPI support is compiled into SimGrid, this option takes the
1238 names of PAPI counters and adds their respective values to the trace
1239 files (See Section :ref:`tracing_tracing_options`).
1243 This feature currently requires superuser privileges, as registers
1244 are queried. Only use this feature with code you trust! Call
1245 smpirun for instance via ``smpirun -wrapper "sudo "
1246 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1247 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1248 will not be required.
1250 It is planned to make this feature available on a per-process (or per-thread?) basis.
1251 The first draft, however, just implements a "global" (i.e., for all processes) set
1252 of counters, the "default" set.
1254 .. code-block:: none
1256 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1258 .. _cfg=smpi/privatization:
1260 Automatic Privatization of Global Variables
1261 ...........................................
1263 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1265 MPI executables are usually meant to be executed in separate
1266 processes, but SMPI is executed in only one process. Global variables
1267 from executables will be placed in the same memory region and shared
1268 between processes, causing intricate bugs. Several options are
1269 possible to avoid this, as described in the main `SMPI publication
1270 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1271 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1272 automatically privatizing the globals, and this option allows one to
1273 choose between them.
1275 - **no** (default when not using smpirun): Do not automatically
1276 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1278 - **dlopen** or **yes** (default when using smpirun): Link multiple
1279 times against the binary.
1280 - **mmap** (slower, but maybe somewhat more stable):
1281 Runtime automatic switching of the data segments.
1284 This configuration option cannot be set in your platform file. You can only
1285 pass it as an argument to smpirun.
1287 .. _cfg=smpi/privatize-libs:
1289 Automatic privatization of global variables inside external libraries
1290 .....................................................................
1292 **Option** ``smpi/privatize-libs`` **default:** unset
1294 **Linux/BSD only:** When using dlopen (default) privatization,
1295 privatize specific shared libraries with internal global variables, if
1296 they can't be linked statically. For example libgfortran is usually
1297 used for Fortran I/O and indexes in files can be mixed up.
1299 Multiple libraries can be given, semicolon separated.
1301 This configuration option can only use either full paths to libraries,
1302 or full names. Check with ldd the name of the library you want to
1305 .. code-block:: console
1309 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1312 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1313 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1314 but not ``libgfortran`` nor ``libgfortran.so``.
1316 .. _cfg=smpi/send-is-detached-thresh:
1318 Simulating MPI detached send
1319 ............................
1321 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1323 This threshold specifies the size in bytes under which the send will
1324 return immediately. This is different from the threshold detailed in
1325 :ref:`cfg=smpi/async-small-thresh` because the message is not
1326 really sent when the send is posted. SMPI still waits for the
1327 corresponding receive to be posted, in order to perform the communication
1330 .. _cfg=smpi/coll-selector:
1332 Simulating MPI collective algorithms
1333 ....................................
1335 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1337 SMPI implements more than 100 different algorithms for MPI collective
1338 communication, to accurately simulate the behavior of most of the
1339 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1340 select the decision logic either of the OpenMPI or the MPICH libraries. (By
1341 default SMPI uses naive version of collective operations.)
1343 Each collective operation can be manually selected with a ``smpi/collective_name:algo_name``. For example, if you want to use
1344 the Bruck algorithm for the Alltoall algorithm, you should use ``--cfg=smpi/alltoall:bruck`` on the command-line of smpirun. The
1345 reference of all available algorithms are listed in :ref:`SMPI_use_colls`, and you can get the full list implemented in your
1346 version using ``smpirun --help-coll``.
1348 .. _cfg=smpi/barrier-collectives:
1350 Add a barrier in all collectives
1351 ................................
1353 **Option** ``smpi/barrier-collectives`` **default:** off
1355 This option adds a simple barrier in some collective operations to catch dangerous
1356 code that may or may not work depending on the MPI implementation: Bcast, Exscan,
1357 Gather, Gatherv, Scan, Scatter, Scatterv and Reduce.
1359 For example, the following code works with OpenMPI while it deadlocks in MPICH and
1360 Intel MPI. Broadcast seem to be "fire and forget" in OpenMPI while other
1361 implementations expect to receive a message.
1366 MPI_Bcast(buf1, buff_size, MPI_CHAR, 0, newcom);
1367 MPI_Send(&buf2, buff_size, MPI_CHAR, 1, tag, newcom);
1368 } else if (rank==1) {
1369 MPI_Recv(&buf2, buff_size, MPI_CHAR, 0, tag, newcom, MPI_STATUS_IGNORE);
1370 MPI_Bcast(buf1, buff_size, MPI_CHAR, 0, newcom);
1373 The barrier is only simulated and does not involve any additional message (it is a S4U barrier).
1374 This option is disabled by default, and activated by the `-analyze` flag of smpirun.
1376 .. _cfg=smpi/barrier-finalization:
1378 Add a barrier in MPI_Finalize
1379 .............................
1381 **Option** ``smpi/finalization-barrier`` **default:** off
1383 By default, SMPI processes are destroyed as soon as soon as their code ends,
1384 so after a successful MPI_Finalize call returns. In some rare cases, some data
1385 might have been attached to MPI objects still active in the remaining processes,
1386 and can be destroyed eagerly by the finished process.
1387 If your code shows issues at finalization, such as segmentation fault, triggering
1388 this option will add an explicit MPI_Barrier(MPI_COMM_WORLD) call inside the
1389 MPI_Finalize, so that all processes will terminate at almost the same point.
1390 It might affect the total timing by the cost of a barrier.
1392 .. _cfg=smpi/errors-are-fatal:
1394 Disable MPI fatal errors
1395 ........................
1397 **Option** ``smpi/errors-are-fatal`` **default:** on
1399 By default, SMPI processes will crash if a MPI error code is returned. MPI allows
1400 to explicitely set MPI_ERRORS_RETURN errhandler to avoid this behaviour. This flag
1401 will turn on this behaviour by default (for all concerned types and errhandlers).
1402 This can ease debugging by going after the first reported error.
1404 .. _cfg=smpi/pedantic:
1406 Disable pedantic MPI errors
1407 ...........................
1409 **Option** ``smpi/pedantic`` **default:** on
1411 By default, SMPI will report all errors it finds in MPI codes. Some of these errors
1412 may not be considered as errors by all developers. This flag can be turned off to
1413 avoid reporting some usually harmless mistakes.
1414 Concerned errors list (will be expanded in the future):
1416 - Calling MPI_Win_fence only once in a program, hence just opening an epoch without
1419 .. _cfg=smpi/iprobe:
1421 Inject constant times for MPI_Iprobe
1422 ....................................
1424 **Option** ``smpi/iprobe`` **default:** 0.0001
1426 The behavior and motivation for this configuration option is identical
1427 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1430 .. _cfg=smpi/iprobe-cpu-usage:
1432 Reduce speed for iprobe calls
1433 .............................
1435 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1437 MPI_Iprobe calls can be heavily used in applications. To account
1438 correctly for the energy that cores spend probing, it is necessary to
1439 reduce the load that these calls cause inside SimGrid.
1441 For instance, we measured a maximum power consumption of 220 W for a
1442 particular application but only 180 W while this application was
1443 probing. Hence, the correct factor that should be passed to this
1444 option would be 180/220 = 0.81.
1448 Inject constant times for MPI_Init
1449 ..................................
1451 **Option** ``smpi/init`` **default:** 0
1453 The behavior and motivation for this configuration option is identical
1454 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1458 Inject constant times for MPI_Isend()
1459 .....................................
1461 **Option** ``smpi/ois``
1463 The behavior and motivation for this configuration option is identical
1464 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1468 Inject constant times for MPI_send()
1469 ....................................
1471 **Option** ``smpi/os``
1473 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1474 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1475 time). SMPI can factor these costs in as well, but the user has to
1476 configure SMPI accordingly as these values may vary by machine. This
1477 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1478 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1479 exactly as ``smpi/ois``.
1481 This item can consist of multiple sections; each section takes three
1482 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1483 so this example contains two sections. Furthermore, each section
1484 consists of three values.
1486 1. The first value denotes the minimum size in bytes for this section to take effect;
1487 read it as "if message size is greater than this value (and other section has a larger
1488 first value that is also smaller than the message size), use this".
1489 In the first section above, this value is "1".
1491 2. The second value is the startup time; this is a constant value that will always
1492 be charged, no matter what the size of the message. In the first section above,
1495 3. The third value is the `per-byte` cost. That is, it is charged for every
1496 byte of the message (incurring cost messageSize*cost_per_byte)
1497 and hence accounts also for larger messages. In the first
1498 section of the example above, this value is "2".
1500 Now, SMPI always checks which section it should use for a given
1501 message; that is, if a message of size 11 is sent with the
1502 configuration of the example above, only the second section will be
1503 used, not the first, as the first value of the second section is
1504 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1505 message of size 11 incurs the following cost inside MPI_Send:
1506 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1508 Note that the order of sections can be arbitrary; they will be ordered internally.
1512 Inject constant times for MPI_Recv()
1513 ....................................
1515 **Option** ``smpi/or``
1517 The behavior and motivation for this configuration option is identical
1518 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1521 .. _cfg=smpi/grow-injected-times:
1523 Inject constant times for MPI_Test
1524 ..................................
1526 **Option** ``smpi/test`` **default:** 0.0001
1528 By setting this option, you can control the amount of time a process
1529 sleeps when MPI_Test() is called; this is important, because SimGrid
1530 normally only advances the time while communication is happening and
1531 thus, MPI_Test will not add to the time, resulting in deadlock if it is
1532 used as a break-condition as in the following example:
1537 MPI_Test(request, flag, status);
1541 To speed up execution, we use a counter to keep track of how often we
1542 checked if the handle is now valid or not. Hence, we actually
1543 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1544 process to sleep increases linearly with the number of previously
1545 failed tests. This behavior can be disabled by setting
1546 ``smpi/grow-injected-times`` to **no**. This will also disable this
1547 behavior for MPI_Iprobe.
1549 .. _cfg=smpi/shared-malloc:
1550 .. _cfg=smpi/shared-malloc-hugepage:
1555 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1557 If your simulation consumes too much memory, you may want to modify
1558 your code so that the working areas are shared by all MPI ranks. For
1559 example, in a block-cyclic matrix multiplication, you will only
1560 allocate one set of blocks, and all processes will share them.
1561 Naturally, this will lead to very wrong results, but this will save a
1562 lot of memory. So this is still desirable for some studies. For more on
1563 the motivation for that feature, please refer to the `relevant section
1564 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1565 of the SMPI CourseWare (see Activity #2.2 of the pointed
1566 assignment). In practice, change the calls for malloc() and free() into
1567 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1569 SMPI provides two algorithms for this feature. The first one, called
1570 ``local``, allocates one block per call to SMPI_SHARED_MALLOC()
1571 (each call site gets its own block) ,and this block is shared
1572 among all MPI ranks. This is implemented with the shm_* functions
1573 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1574 for each shared block.
1576 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1577 returns a new address, but it only points to a shadow block: its memory
1578 area is mapped on a 1 MiB file on disk. If the returned block is of size
1579 N MiB, then the same file is mapped N times to cover the whole block.
1580 At the end, no matter how many times you call SMPI_SHARED_MALLOC, this will
1581 only consume 1 MiB in memory.
1583 You can disable this behavior and come back to regular mallocs (for
1584 example for debugging purposes) using ``no`` as a value.
1586 If you want to keep private some parts of the buffer, for instance if these
1587 parts are used by the application logic and should not be corrupted, you
1588 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). For example:
1592 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1594 This will allocate 500 bytes to mem, such that mem[27..41] and
1595 mem[100..199] are shared while other area remain private.
1597 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1599 When smpi/shared-malloc:global is used, the memory consumption problem
1600 is solved, but it may induce too much load on the kernel's pages table.
1601 In this case, you should use huge pages so that the kernel creates only one
1602 entry per MB of malloced data instead of one entry per 4 kB.
1603 To activate this, you must mount a hugetlbfs on your system and allocate
1604 at least one huge page:
1606 .. code-block:: console
1609 $ sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1610 $ sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1612 Then, you can pass the option
1613 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1614 actually activate the huge page support in shared mallocs.
1616 .. _cfg=smpi/auto-shared-malloc-thresh:
1618 Automatically share allocations
1619 ...............................
1621 **Option** ``smpi/auto-shared-malloc-thresh:`` **Default:** 0 (false)
1622 This value in bytes represents the size above which all allocations
1623 will be "shared" by default (as if they were performed through
1624 SMPI_SHARED_MALLOC macros). Default = 0 = disabled feature.
1625 The value must be carefully chosen to only select data buffers which
1626 will not modify execution path or cause crash if their content is false.
1627 Option :ref:`cfg=smpi/display-allocs` can be used to locate the largest
1628 allocation detected in a run, and provide a good starting threshold.
1629 Note : malloc, calloc and free are overridden by smpicc/cxx by default.
1630 This can cause some troubles if codes are already overriding these. If this
1631 is the case, defining SMPI_NO_OVERRIDE_MALLOC in the compilation flags can
1632 help, but will make this feature unusable.
1636 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1637 ...................................................................
1639 **Option** ``smpi/wtime`` **default:** 10 ns
1641 This option controls the amount of (simulated) time spent in calls to
1642 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1643 to 0, the simulated clock is not advanced in these calls, which leads
1644 to issues if your application contains such a loop:
1648 while(MPI_Wtime() < some_time_bound) {
1649 /* some tests, with no communication nor computation */
1652 When the option smpi/wtime is set to 0, the time advances only on
1653 communications and computations. So the previous code results in an
1654 infinite loop: the current [simulated] time will never reach
1655 ``some_time_bound``. This infinite loop is avoided when that option
1656 is set to a small value, as it is by default since SimGrid v3.21.
1658 Note that if your application does not contain any loop depending on
1659 the current time only, then setting this option to a non-zero value
1660 will slow down your simulations by a tiny bit: the simulation loop has
1661 to be broken out of and reset each time your code asks for the current time.
1662 If the simulation speed really matters to you, you can avoid this
1663 extra delay by setting smpi/wtime to 0.
1665 .. _cfg=smpi/list-leaks:
1667 Report leaked MPI objects
1668 .........................
1670 **Option** ``smpi/list-leaks`` **default:** 0
1672 This option controls whether to report leaked MPI objects.
1673 The parameter is the number of leaks to report.
1675 Other Configurations
1676 --------------------
1678 .. _cfg=debug/clean-atexit:
1680 Cleanup at Termination
1681 ......................
1683 **Option** ``debug/clean-atexit`` **default:** on
1685 If your code is segfaulting during its finalization, it may help to
1686 disable this option to request that SimGrid not attempt any cleanups at
1687 the end of the simulation. Since the Unix process is ending anyway,
1688 the operating system will wipe it all.
1695 **Option** ``path`` **default:** . (current dir)
1697 It is possible to specify a list of directories to search in for the
1698 trace files (see :ref:`pf_trace`) by using this configuration
1699 item. To add several directory to the path, set the configuration
1700 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1702 .. _cfg=debug/breakpoint:
1707 **Option** ``debug/breakpoint`` **default:** unset
1709 This configuration option sets a breakpoint: when the simulated clock
1710 reaches the given time, a SIGTRAP is raised. This can be used to stop
1711 the execution and get a backtrace with a debugger.
1713 It is also possible to set the breakpoint from inside the debugger, by
1714 writing in global variable simgrid::kernel::cfg_breakpoint. For example,
1717 .. code-block:: none
1719 set variable simgrid::kernel::cfg_breakpoint = 3.1416
1721 .. _cfg=debug/verbose-exit:
1726 **Option** ``debug/verbose-exit`` **default:** on
1728 By default, when Ctrl-C is pressed, the status of all existing actors
1729 is displayed before exiting the simulation. This is very useful to
1730 debug your code, but it can become troublesome if you have many
1731 actors. Set this configuration item to **off** to disable this
1734 .. _cfg=exception/cutpath:
1736 Truncate local path from exception backtrace
1737 ............................................
1739 **Option** ``exception/cutpath`` **default:** off
1741 This configuration option is used to remove the path from the
1742 backtrace shown when an exception is thrown. This is mainly useful for
1743 the tests: the full file path would makes the tests non-reproducible because
1744 the paths of source files depend of the build settings. That would
1745 break most of the tests since their output is continually compared.
1749 Logging configuration
1750 ---------------------
1752 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
1753 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
1754 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
1757 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
1758 messages from your code.
1760 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
1761 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
1762 practice, the following is equivalent to the above settings: ``--log=root.thresh:error --log=s4u_host.thresh:debug``.
1764 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
1765 your settings, as in ``--log="root.thresh:error s4u_host.thresh:debug"``. The parameters are interpreted in order, from left to right.
1768 Threshold configuration
1769 .......................
1771 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
1772 ``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
1773 see, ``threshold`` can be abbreviated here.
1775 Existing thresholds:
1777 - ``trace`` some functions display a message at this level when entering or returning
1778 - ``debug`` output that is mostly useful when debugging the corresponding module.
1779 - ``verbose`` verbose output that is only mildly interesting and can easily be ignored
1780 - ``info`` usual output (this is the default threshold of all categories)
1781 - ``warning`` minor issue encountered
1782 - ``error`` issue encountered
1783 - ``critical`` major issue encountered, such as assertions failures
1787 Format configuration
1788 ....................
1790 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
1791 as the date, or the actor ID, everything. Existing format directives:
1794 - %n: line separator (LOG4J compatible)
1795 - %e: plain old space (SimGrid extension)
1797 - %m: user-provided message
1799 - %c: Category name (LOG4J compatible)
1800 - %p: Priority name (LOG4J compatible)
1802 - %h: Hostname (SimGrid extension)
1803 - %a: Actor name (SimGrid extension -- note that with SMPI this is the integer value of the process rank)
1804 - %i: Actor PID (SimGrid extension -- this is a 'i' as in 'i'dea)
1805 - %t: Thread "name" (LOG4J compatible -- actually the address of the thread in memory)
1807 - %F: file name where the log event was raised (LOG4J compatible)
1808 - %l: location where the log event was raised (LOG4J compatible, like '%%F:%%L' -- this is a l as in 'l'etter)
1809 - %L: line number where the log event was raised (LOG4J compatible)
1810 - %M: function name (LOG4J compatible -- called method name here of course).
1812 - %d: date (UNIX-like epoch)
1813 - %r: application age (time elapsed since the beginning of the application)
1816 ``--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
1817 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
1818 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
1819 provided layout is used for every messages.
1821 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
1825 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
1826 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'"``.
1827 Another option is to use the ``%e`` directive for spaces, as in ``--log=root.fmt:%l:%e[%p/%c]:%e%m%n``.
1832 The keyword ``app`` controls the appended of a logging category. For example ``--log=root.app:file:mylogfile`` redirects every output to the file ``mylogfile``.
1834 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
1835 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.
1837 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``
1838 ensures that the log file ``mylog`` will never overpass 500 bytes in size.
1840 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
1841 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.
1846 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
1847 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
1848 ``on`` (or ``yes`` or ``1``), the produced messages will also be passed to the upper appender.
1850 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
1851 ``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
1852 will only be sent to ``all.log``.
1857 ``--help-logs`` displays a complete help message about logging in SimGrid.
1859 ``--help-log-categories`` displays the actual hierarchy of log categories for this binary.
1861 ``--log=no_loc`` hides the source locations (file names and line numbers) from the messages. This is useful to make tests reproducible.