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/communications-determinism:** :ref:`cfg=model-check/communications-determinism`
112 - **model-check/dot-output:** :ref:`cfg=model-check/dot-output`
113 - **model-check/max-depth:** :ref:`cfg=model-check/max-depth`
114 - **model-check/reduction:** :ref:`cfg=model-check/reduction`
115 - **model-check/replay:** :ref:`cfg=model-check/replay`
116 - **model-check/send-determinism:** :ref:`cfg=model-check/send-determinism`
117 - **model-check/setenv:** :ref:`cfg=model-check/setenv`
118 - **model-check/timeout:** :ref:`cfg=model-check/timeout`
120 - **network/bandwidth-factor:** :ref:`cfg=network/bandwidth-factor`
121 - **network/crosstraffic:** :ref:`cfg=network/crosstraffic`
122 - **network/latency-factor:** :ref:`cfg=network/latency-factor`
123 - **network/loopback-lat:** :ref:`cfg=network/loopback`
124 - **network/loopback-bw:** :ref:`cfg=network/loopback`
125 - **network/maxmin-selective-update:** :ref:`Network Optimization Level <options_model_optim>`
126 - **network/model:** :ref:`options_model_select`
127 - **network/optim:** :ref:`Network Optimization Level <options_model_optim>`
128 - **network/TCP-gamma:** :ref:`cfg=network/TCP-gamma`
129 - **network/weight-S:** :ref:`cfg=network/weight-S`
131 - **ns3/TcpModel:** :ref:`options_pls`
132 - **ns3/seed:** :ref:`options_pls`
133 - **path:** :ref:`cfg=path`
134 - **plugin:** :ref:`cfg=plugin`
136 - **storage/max_file_descriptors:** :ref:`cfg=storage/max_file_descriptors`
138 - **precision/timing:** :ref:`cfg=precision/timing`
139 - **precision/work-amount:** :ref:`cfg=precision/work-amount`
141 - **For collective operations of SMPI,** please refer to Section :ref:`cfg=smpi/coll-selector`
142 - **smpi/auto-shared-malloc-thresh:** :ref:`cfg=smpi/auto-shared-malloc-thresh`
143 - **smpi/async-small-thresh:** :ref:`cfg=smpi/async-small-thresh`
144 - **smpi/barrier-finalization:** :ref:`cfg=smpi/barrier-finalization`
145 - **smpi/barrier-collectives:** :ref:`cfg=smpi/barrier-collectives`
146 - **smpi/buffering:** :ref:`cfg=smpi/buffering`
147 - **smpi/coll-selector:** :ref:`cfg=smpi/coll-selector`
148 - **smpi/comp-adjustment-file:** :ref:`cfg=smpi/comp-adjustment-file`
149 - **smpi/cpu-threshold:** :ref:`cfg=smpi/cpu-threshold`
150 - **smpi/display-allocs:** :ref:`cfg=smpi/display-allocs`
151 - **smpi/display-timing:** :ref:`cfg=smpi/display-timing`
152 - **smpi/errors-are-fatal:** :ref:`cfg=smpi/errors-are-fatal`
153 - **smpi/grow-injected-times:** :ref:`cfg=smpi/grow-injected-times`
154 - **smpi/host-speed:** :ref:`cfg=smpi/host-speed`
155 - **smpi/IB-penalty-factors:** :ref:`cfg=smpi/IB-penalty-factors`
156 - **smpi/iprobe:** :ref:`cfg=smpi/iprobe`
157 - **smpi/iprobe-cpu-usage:** :ref:`cfg=smpi/iprobe-cpu-usage`
158 - **smpi/init:** :ref:`cfg=smpi/init`
159 - **smpi/keep-temps:** :ref:`cfg=smpi/keep-temps`
160 - **smpi/ois:** :ref:`cfg=smpi/ois`
161 - **smpi/or:** :ref:`cfg=smpi/or`
162 - **smpi/os:** :ref:`cfg=smpi/os`
163 - **smpi/papi-events:** :ref:`cfg=smpi/papi-events`
164 - **smpi/pedantic:** :ref:`cfg=smpi/pedantic`
165 - **smpi/privatization:** :ref:`cfg=smpi/privatization`
166 - **smpi/privatize-libs:** :ref:`cfg=smpi/privatize-libs`
167 - **smpi/send-is-detached-thresh:** :ref:`cfg=smpi/send-is-detached-thresh`
168 - **smpi/shared-malloc:** :ref:`cfg=smpi/shared-malloc`
169 - **smpi/shared-malloc-hugepage:** :ref:`cfg=smpi/shared-malloc-hugepage`
170 - **smpi/simulate-computation:** :ref:`cfg=smpi/simulate-computation`
171 - **smpi/test:** :ref:`cfg=smpi/test`
172 - **smpi/wtime:** :ref:`cfg=smpi/wtime`
173 - **smpi/list-leaks** :ref:`cfg=smpi/list-leaks`
175 - **Tracing configuration options** can be found in Section :ref:`tracing_tracing_options`
177 - **storage/model:** :ref:`options_model_select`
179 - **vm/model:** :ref:`options_model_select`
183 Configuring the Platform Models
184 -------------------------------
186 .. _options_model_select:
188 Choosing the Platform Models
189 ............................
191 SimGrid comes with several network, CPU and disk models built in,
192 and you can change the used model at runtime by changing the passed
193 configuration. The three main configuration items are given below.
194 For each of these items, passing the special ``help`` value gives you
195 a short description of all possible values (for example,
196 ``--cfg=network/model:help`` will present all provided network
197 models). Also, ``--help-models`` should provide information about all
198 models for all existing resources.
200 - ``network/model``: specify the used network model. Possible values:
202 - **LV08 (default one):** Realistic network analytic model
203 (slow-start modeled by multiplying latency by 13.01, bandwidth by
204 .97; bottleneck sharing uses a payload of S=20537 for evaluating
205 RTT). Described in `Accuracy Study and Improvement of Network
206 Simulation in the SimGrid Framework
207 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
208 - **Constant:** Simplistic network model where all communication
209 take a constant time (one second). This model provides the lowest
210 realism, but is (marginally) faster.
211 - **SMPI:** Realistic network model specifically tailored for HPC
212 settings (accurate modeling of slow start with correction factors on
213 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). This model can be
214 :ref:`further configured <options_model_network>`.
215 - **IB:** Realistic network model specifically tailored for HPC
216 settings with InfiniBand networks (accurate modeling contention
217 behavior, based on the model explained in `this PhD work
218 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_.
219 This model can be :ref:`further configured <options_model_network>`.
220 - **CM02:** Legacy network analytic model. Very similar to LV08, but
221 without corrective factors. The timings of small messages are thus
222 poorly modeled. This model is described in `A Network Model for
223 Simulation of Grid Application
224 <https://hal.inria.fr/inria-00071989/document>`_.
225 - **ns-3** (only available if you compiled SimGrid accordingly):
226 Use the packet-level network
227 simulators as network models (see :ref:`models_ns3`).
228 This model can be :ref:`further configured <options_pls>`.
230 - ``cpu/model``: specify the used CPU model. We have only one model for now:
232 - **Cas01:** Simplistic CPU model (time=size/speed)
234 - ``host/model``: we have two such models for now.
236 - **default:** Default host model. It simply uses the otherwise configured models for cpu, disk and network (i.e. CPU:Cas01,
237 disk:S19 and network:LV08 by default)
238 - **ptask_L07:** This model is mandatory if you plan to use parallel tasks (and useless otherwise). ptasks are intended to
239 model the moldable tasks of the grid scheduling literature. A specific host model is necessary because each such activity
240 has a both compute and communicate components, so the CPU and network models must be mixed together.
242 - ``storage/model``: specify the used storage model. Only one model is
244 - ``vm/model``: specify the model for virtual machines. Only one model
247 .. todo: make 'compound' the default host model.
249 .. _options_model_solver:
254 The different models rely on a linear inequalities solver to share
255 the underlying resources. SimGrid allows you to change the solver, but
256 be cautious, **don't change it unless you are 100% sure**.
258 - items ``cpu/solver``, ``network/solver``, ``disk/solver`` and ``host/solver``
259 allow you to change the solver for each model:
261 - **maxmin:** The default solver for all models except ptask. Provides a
262 max-min fairness allocation.
263 - **fairbottleneck:** The default solver for ptasks. Extends max-min to
264 allow heterogeneous resources.
265 - **bmf:** More realistic solver for heterogeneous resource sharing.
266 Implements BMF (Bottleneck max fairness) fairness. To be used with
267 parallel tasks instead of fair-bottleneck.
269 .. _options_model_optim:
274 The network and CPU models that are based on linear inequalities solver (that
275 is, all our analytical models) accept specific optimization
278 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
280 - **Lazy:** Lazy action management (partial invalidation in lmm +
281 heap in action remaining).
282 - **TI:** Trace integration. Highly optimized mode when using
283 availability traces (only available for the Cas01 CPU model for
285 - **Full:** Full update of remaining and variables. Slow but may be
286 useful when debugging.
288 - items ``network/maxmin-selective-update`` and
289 ``cpu/maxmin-selective-update``: configure whether the underlying
290 should be lazily updated or not. It should have no impact on the
291 computed timings, but should speed up the computation. |br| It is
292 still possible to disable this feature because it can reveal
293 counter-productive in very specific scenarios where the
294 interaction level is high. In particular, if all your
295 communication share a given backbone link, you should disable it:
296 without it, a simple regular loop is used to update each
297 communication. With it, each of them is still updated (because of
298 the dependency induced by the backbone), but through a complicated
299 and slow pattern that follows the actual dependencies.
301 .. _cfg=bmf/precision:
302 .. _cfg=precision/timing:
303 .. _cfg=precision/work-amount:
308 **Option** ``precision/timing`` **Default:** 1e-9 (in seconds) |br|
309 **Option** ``precision/work-amount`` **Default:** 1e-5 (in flops or bytes) |br|
310 **Option** ``bmf/precision`` **Default:** 1e-12 (no unit)
312 The analytical models handle a lot of floating point values. It is
313 possible to change the epsilon used to update and compare them through
314 this configuration item. Changing it may speedup the simulation by
315 discarding very small actions, at the price of a reduced numerical
316 precision. You can modify separately the precision used to manipulate
317 timings (in seconds) and the one used to manipulate amounts of work
320 .. _cfg=maxmin/concurrency-limit:
325 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
327 The maximum number of variables per resource can be tuned through this
328 option. You can have as many simultaneous actions per resources as you
329 want. If your simulation presents a very high level of concurrency, it
330 may help to use e.g. 100 as a value here. It means that at most 100
331 actions can consume a resource at a given time. The extraneous actions
332 are queued and wait until the amount of concurrency of the considered
333 resource lowers under the given boundary.
335 Such limitations help both to the simulation speed and simulation accuracy
336 on highly constrained scenarios, but the simulation speed suffers of this
337 setting on regular (less constrained) scenarios so it is off by default.
339 .. _cfg=bmf/max-iterations:
344 **Option** ``bmf/max-iterations`` **Default:** 1000
346 It may happen in some settings that the BMF solver fails to converge to
347 a solution, so there is a hard limit on the amount of iteration count to
348 avoid infinite loops.
350 .. _options_model_network:
352 Configuring the Network Model
353 .............................
355 .. _cfg=network/TCP-gamma:
357 Maximal TCP Window Size
358 ^^^^^^^^^^^^^^^^^^^^^^^
360 **Option** ``network/TCP-gamma`` **Default:** 4194304
362 The analytical models need to know the maximal TCP window size to take the TCP congestion mechanism into account (see
363 :ref:`this page <understanding_cm02>` for details). On Linux, this value can be retrieved using the following commands.
364 Both give a set of values, and you should use the last one, which is the maximal size.
366 .. code-block:: console
368 $ cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
369 $ cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
371 If you want to disable the TCP windowing mechanism, set this parameter to 0.
373 .. _cfg=network/bandwidth-factor:
374 .. _cfg=network/latency-factor:
375 .. _cfg=network/weight-S:
377 Manual calibration factors
378 ^^^^^^^^^^^^^^^^^^^^^^^^^^
380 SimGrid can take network irregularities such as a slow startup or changing behavior depending on the message size into account.
381 The values provided by default were computed a long time ago through data fitting one the timings of either packet-level
382 simulators or direct experiments on real platforms. These default values should be OK for most users, but if simulation realism
383 is really important to you, you probably want to recalibrate the models (i.e., devise sensible values for your specific
384 settings). This section only describes how to pass new values to the models while the calibration process involved in the
385 computation of these values is described :ref:`in the relevant chapter <models_calibration>`.
387 We found out that many networking effects can be realistically accounted for with the three following correction factors. They
388 were shown to be enough to capture slow-start effects, the different transmission modes of MPI systems (eager vs. rendez-vous
389 mode), or the non linear effects of wifi sharing.
391 **Option** ``network/latency-factor`` **Default:** 1.0, but overridden by most models
393 This option specifies a multiplier to apply to the *physical* latency (i.e., the one described in the platform) of the set of
394 links involved in a communication. The factor can either be a constant to apply to any communication, or it may depend on the
395 message size. The ``CM02`` model does not use any correction factor, so the latency-factor remains to 1. The ``LV08`` model sets
396 it to 13.01 to model slow-start, while the ``SMPI`` model has several possible values depending on the interval in which the
397 message size falls. The default SMPI setting given below specifies for example that a message smaller than 257 bytes will get a
398 latency multiplier of 2.01467 while a message whose size is in [15424, 65472] will get a latency multiplier of 3.48845. The
399 ``wifi`` model goes further and uses a callback in the program to compute the factor that must be non-linear in this case.
401 This multiplier is applied to the latency computed from the platform, that is the sum of all link *physical* latencies over the
402 :ref:`network path <platform_routing>` used by the considered communication, to derive the *effective* end-to-end latency.
404 Constant factors are easy to express, but the interval-based syntax used in SMPI is somewhat complex. It expects a set of
405 factors separated by semicolons, each of the form ``boundary:factor``. For example if your specification is
406 ``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
407 5000 and beyond. If your first interval does include size=0, then the default value of 1 is used before. Changing the factor
408 callback is not possible from the command line and must be done from your code, as shown in `this example
409 <https://framagit.org/simgrid/simgrid/tree/master/examples/cpp/network-factors/s4u-network-factors.cpp>`_. Note that the chosen
410 model only provides some default settings. You may pick a ``LV08`` model to get some of the settings, and override the latency
411 with interval-based values.
413 SMPI default value: 65472:11.6436; 15424:3.48845; 9376:2.59299; 5776:2.18796; 3484:1.88101; 1426:1.61075; 732:1.9503;
414 257:1.95341;0:2.01467 (interval boundaries are sorted automatically). These values were computed by data fitting on the Stampede
415 Supercomputer at TACC, with optimal deployment of processes on nodes. To accurately model your settings, you should redo the
416 :ref:`calibration <models_calibration>`.
418 **Option** ``network/bandwidth-factor`` **Default:** 1.0, but overridden by most models
420 Setting this option automatically adjusts the *effective* bandwidth (i.e., the one perceived by the application) used by any
421 given communication. As with latency-factor above, the value can be a constant (``CM02`` uses 1 -- no correction -- while
422 ``LV08`` uses 0.97 to discount TCP headers while computing the payload bandwidth), interval-based (as the default provided by
423 the ``SMPI``), or using in-program callbacks (as with ``wifi``).
425 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
426 This was also computed on the Stampede Supercomputer.
428 **Option** ``network/weight-S`` **Default:** depends on the model
430 Value used to account for RTT-unfairness when sharing a bottleneck (network connections with a large RTT are generally penalized
431 against those with a small one). See :ref:`models_TCP` and also this scientific paper: `Accuracy Study and Improvement of Network
432 Simulation in the SimGrid Framework <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_
434 Default values for ``CM02`` is 0. ``LV08`` sets it to 20537 while both ``SMPI`` and ``IB`` set it to 8775.
436 .. _cfg=network/loopback:
438 Configuring loopback link
439 ^^^^^^^^^^^^^^^^^^^^^^^^^
441 Several network models provide an implicit loopback link to account for local
442 communication on a host. By default it has a 10GBps bandwidth and a null latency.
443 This can be changed with ``network/loopback-lat`` and ``network/loopback-bw``
444 items. Note that this loopback is conveniently modeled with a :ref:`single FATPIPE link <pf_loopback>`
445 for the whole platform. If modeling contention inside nodes is important then you should
446 rather add such loopback links (one for each host) yourself.
448 .. _cfg=smpi/IB-penalty-factors:
453 InfiniBand network behavior can be modeled through 3 parameters
454 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `the PhD
455 thesis of Jérôme Vienne
456 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_ (in French)
457 or more concisely in `this paper <https://hal.inria.fr/hal-00953618/document>`_,
458 even if that paper does only describe models for myrinet and ethernet.
459 You can see in Fig 2 some results for Infiniband, for example. This model
460 may be outdated by now for modern infiniband, anyway, so a new
461 validation would be good.
463 The three paramaters are defined as follows:
465 - βs: penalty factor for outgoing messages, computed by running a simple send to
466 two nodes and checking slowdown compared to a single send to one node,
468 - βe: penalty factor for ingoing messages, same computation method but with one
469 node receiving several messages
470 - γr: slowdown factor when communication buffer memory is saturated. It needs a
471 more complicated pattern to run in order to be computed (5.3 in the thesis,
472 page 107), and formula in the end is γr = time(c)/(3×βe×time(ref)), where
473 time(ref) is the time of a single comm with no contention).
475 Once these values are computed, a penalty is assessed for each message (this is
476 the part implemented in the simulator) as shown page 106 of the thesis. Here is
477 a simple translation of this text. First, some notations:
479 - ∆e(e) which corresponds to the incoming degree of node e, that is to say the number of communications having as destination node e.
480 - ∆s (s) which corresponds to the degree outgoing from node s, that is to say the number of communications sent by node s.
481 - Φ (e) which corresponds to the number of communications destined for the node e but coming from a different node.
482 - 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
484 To determine the penalty for a communication, two values need to be calculated. First, the penalty caused by the conflict in transmission, noted ps.
487 - if ∆s (i) = 1 then ps = 1.
488 - if ∆s (i) ≥ 2 and ∆e (i) ≥ 3 then ps = ∆s (i) × βs × γr
489 - else, ps = ∆s (i) × βs
492 Then, the penalty caused by the conflict in reception (noted pe) should be computed as follows:
494 - if ∆e (i) = 1 then pe = 1
495 - else, pe = Φ (e) × βe × Ω (s, e)
497 Finally, the penalty associated with the communication is:
498 p = max (ps ∈ s, pe)
500 .. _cfg=network/crosstraffic:
502 Simulating Cross-Traffic
503 ^^^^^^^^^^^^^^^^^^^^^^^^
505 Since SimGrid v3.7, cross-traffic effects can be taken into account in
506 analytical simulations. It means that ongoing and incoming
507 communication flows are treated independently. In addition, the LV08
508 model adds 0.05 of usage on the opposite direction for each new
509 created flow. This can be useful to simulate some important TCP
510 phenomena such as ack compression.
512 For that to work, your platform must have two links for each
513 pair of interconnected hosts. An example of usable platform is
514 available in ``examples/platforms/crosstraffic.xml``.
516 This is activated through the ``network/crosstraffic`` item, that
517 can be set to 0 (disable this feature) or 1 (enable it).
519 Note that with the default host model this option is activated by default.
521 .. _cfg=smpi/async-small-thresh:
523 Simulating Asynchronous Send
524 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
526 It is possible to specify that messages below a certain size (in bytes) will be
527 sent as soon as the call to MPI_Send is issued, without waiting for
528 the correspondent receive. This threshold can be configured through
529 the ``smpi/async-small-thresh`` item. The default value is 0. This
530 behavior can also be manually set for mailboxes, by setting the
531 receiving mode of the mailbox with a call to
532 :cpp:func:`sg_mailbox_set_receiver`. After this, all messages sent to
533 this mailbox will have this behavior regardless of the message size.
535 This value needs to be smaller than or equals to the threshold set at
536 :ref:`cfg=smpi/send-is-detached-thresh`, because asynchronous messages
537 are meant to be detached as well.
544 **Option** ``ns3/NetworkModel`` **Default:** "default" (ns-3 default TCP)
546 When using ns-3, the item ``ns3/NetworkModel`` can be used to switch between TCP or UDP, and switch the used TCP variante. If
547 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.
548 With the value of either 'NewReno' or 'Cubic', the ``ns3::TcpL4Protocol::SocketType`` configuration item in ns-3 is set to the
549 corresponding protocol.
551 **Option** ``ns3/seed`` **Default:** "" (don't set the seed in ns-3)
553 This option is the random seed to provide to ns-3 with
554 ``ns3::RngSeedManager::SetSeed`` and ``ns3::RngSeedManager::SetRun``.
556 If left blank, no seed is set in ns-3. If the value 'time' is
557 provided, the current amount of seconds since epoch is used as a seed.
558 Otherwise, the provided value must be a number to use as a seed.
560 Configuring the Storage model
561 .............................
563 .. _cfg=storage/max_file_descriptors:
565 File Descriptor Count per Host
566 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
568 **Option** ``storage/max_file_descriptors`` **Default:** 1024
570 Each host maintains a fixed-size array of its file descriptors. You
571 can change its size through this item to either enlarge it if your
572 application requires it or to reduce it to save memory space.
579 SimGrid plugins allow one to extend the framework without changing its
580 source code directly. Read the source code of the existing plugins to
581 learn how to do so (in ``src/plugins``), and ask your questions to the
582 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
583 that plugins usually register callbacks to some signals of interest.
584 If they need to store some information about a given object (Link, CPU
585 or Actor), they do so through the use of a dedicated object extension.
587 Some of the existing plugins can be activated from the command line,
588 meaning that you can activate them from the command line without any
589 modification to your simulation code. For example, you can activate
590 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
593 Here is a partial list of plugins that can be activated this way. You can get
594 the full list by passing ``--cfg=plugin:help`` to your simulator.
596 - :ref:`Host Energy <plugin_host_energy>`: models the energy dissipation of the compute units.
597 - :ref:`Link Energy <plugin_link_energy>`: models the energy dissipation of the network.
598 - :ref:`Host Load <plugin_host_load>`: monitors the load of the compute units.
600 .. _options_modelchecking:
602 Configuring the Model-Checking
603 ------------------------------
605 To enable SimGrid's model-checking support, the program should
606 be executed using the simgrid-mc wrapper:
608 .. code-block:: console
610 $ simgrid-mc ./my_program
612 Safety properties are expressed as assertions using the function
613 :cpp:func:`void MC_assert(int prop)`.
615 .. _cfg=smpi/buffering:
617 Specifying the MPI buffering behavior
618 .....................................
620 **Option** ``smpi/buffering`` **Default:** infty
622 Buffering in MPI has a huge impact on the communication semantic. For example,
623 standard blocking sends are synchronous calls when the system buffers are full
624 while these calls can complete immediately without even requiring a matching
625 receive call for small messages sent when the system buffers are empty.
627 In SMPI, this depends on the message size, that is compared against two thresholds:
629 - if (size < :ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>`) then
630 MPI_Send returns immediately, and the message is sent even if the
631 corresponding receive has not be issued yet. This is known as the eager mode.
632 - if (:ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>` < size <
633 :ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>`) then
634 MPI_Send also returns immediately, but SMPI waits for the corresponding
635 receive to be posted, in order to perform the communication operation.
636 - if (:ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>` < size) then
637 MPI_Send returns only when the message has actually been sent over the network. This is known as the rendez-vous mode.
639 The ``smpi/buffering`` (only valid with MC) option gives an easier interface to choose between these semantics. It can take two values:
641 - **zero:** means that buffering should be disabled. All communications are actually blocking.
642 - **infty:** means that buffering should be made infinite. All communications are non-blocking.
644 .. _cfg=model-check/reduction:
646 Specifying the kind of reduction
647 ................................
649 **Option** model-check/reduction **Default:** "dpor"
651 The main issue when using the model-checking is the state space
652 explosion. You can activate some reduction technique with
653 ``--cfg=model-check/reduction:<technique>``. For now, this
654 configuration variable can take 2 values:
656 - **none:** Do not apply any kind of reduction
657 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if you verify local safety properties (default value for
659 - **sdpor:** Source-set DPOR, as described in "Source Sets: A Foundation for Optimal Dynamic Partial Order Reduction"
661 - **odpor:** Optimal DPOR, as described in "Source Sets: A Foundation for Optimal Dynamic Partial Order Reduction"
664 Our current DPOR implementation could be improved in may ways. We are
665 currently improving its efficiency (both in term of reduction ability
666 and computational speed).
668 .. _cfg=model-check/strategy:
673 **Option** model-check/strategy **Default:** "none"
675 Even after the DPOR's reduction, the state space that we have to explore remains huge. SimGrid provides several guiding
676 strategies aiming at converging faster toward bugs. By default, none of these strategy is enabled, and SimGrid does a regular
679 - **max_match_comm**: Try to minimize the number of in-fly communication by appairing matching send and receive. This tend to
680 produce nicer backtraces, in particular when a user-level ``send`` is broken down internally into a ``send_async`` + ``wait``.
681 This strategy will ensure that the ``wait`` occures as soon as possible, easing the understanding of the user who do not
682 expect her ``send`` to be split.
683 - **min_match_comm**: Try to maximize the number of in-fly communication by not appairing matching send and receive. This is
684 the exact opposite strategy, but it is still useful as it tend to explore first the branches where the risk of deadlock is
686 - **uniform**: this is a boring random strategy where choices are based on a uniform sampling of possible choices.
687 Surprisingly, it gives really really good results.
689 .. _cfg=model-check/dot-output:
694 If set, the ``model-check/dot-output`` configuration item is the name
695 of a file in which to write a dot file of the path leading to the
696 property violation discovered (safety violation). This dot file can then be fed to the
697 graphviz dot tool to generate a corresponding graphical representation.
699 .. _cfg=model-check/max-depth:
701 Exploration Depth Limit
702 .......................
704 The ``model-check/max-depth`` can set the maximum depth of the
705 exploration graph of the model checker. If this limit is reached, a
706 logging message is sent and the results might not be exact.
708 By default, the exploration is limited to the depth of 1000.
710 .. _cfg=model-check/timeout:
715 By default, the model checker does not handle timeout conditions: the `wait`
716 operations never time out. With the ``model-check/timeout`` configuration item
717 set to **yes**, the model checker will explore timeouts of `wait` operations.
719 .. _cfg=model-check/communications-determinism:
720 .. _cfg=model-check/send-determinism:
722 Communication Determinism
723 .........................
725 The ``model-check/communications-determinism`` and
726 ``model-check/send-determinism`` items can be used to select the
727 communication determinism mode of the model checker, which checks
728 determinism properties of the communications of an application.
730 .. _cfg=model-check/setenv:
732 Passing environment variables
733 .............................
735 You can specify extra environment variables to be set in the verified application
736 with ``model-check/setenv``. For example, you can preload a library as follows:
737 ``-cfg=model-check/setenv:LD_PRELOAD=toto;LD_LIBRARY_PATH=/tmp``.
741 Verification Performance Considerations
742 .......................................
744 The size of the stacks can have a huge impact on the memory
745 consumption when using model-checking. By default, each snapshot will
746 save a copy of the whole stacks and not only of the part that is
747 really meaningful: you should expect the contribution of the memory
748 consumption of the snapshots to be:
749 :math:`\text{number of processes} \times \text{stack size} \times \text{number of states}`.
751 When compiled against the model checker, the stacks are not
752 protected with guards: if the stack size is too small for your
753 application, the stack will silently overflow into other parts of the
754 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
756 .. _cfg=model-check/replay:
758 Replaying buggy execution paths from the model checker
759 ......................................................
761 Debugging the problems reported by the model checker is challenging because
762 the model checker may explore several execution paths before encountering the issue,
763 the output very difficult to understand. Fortunately, SimGrid provides
764 the execution path leading to any reported issue so that you can replay
765 this path reported by the model checker, restoring a classical experience of debugging.
767 When the model checker finds an interesting path in the application
768 execution graph (where a safety property is violated), it
769 generates an identifier for this path. Here is an example of the output:
771 .. code-block:: console
773 [ 0.000000] (0:@) Check a safety property
774 [ 0.000000] (0:@) **************************
775 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
776 [ 0.000000] (0:@) **************************
777 [ 0.000000] (0:@) Counter-example execution trace:
778 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
779 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
780 [ 0.000000] (0:@) Path = 1/3;1/4
781 [ 0.000000] (0:@) Expanded states = 27
782 [ 0.000000] (0:@) Visited states = 68
783 [ 0.000000] (0:@) Executed transitions = 46
785 The interesting line is ``Path = 1/3;1/4``, which means that you should use
786 ``--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
787 execution path. All options (but the model checker related ones) must
788 remain the same. In particular, if you ran your application with
789 ``smpirun -wrapper simgrid-mc``, then remove the ``-wrapper simgrid-mc`` part
790 (you may want to use valgrind or gdb as wrappers instead). Also remove all
791 MC-related options, keep non-MC-related ones and add
792 ``--cfg=model-check/replay:???``.
794 Things are very similar if you are using sthread. Simply drop ``simgrid-mc`` from your command line, as follows:
796 .. code-block:: console
798 $ LD_PRELOAD=../../lib/libsthread.so ./pthread-mutex-simpledeadlock --cfg=model-check/replay:'2;2;3;2;3;3'
799 sthread is intercepting the execution of ./pthread-mutex-simpledeadlock
800 [0.000000] [xbt_cfg/INFO] Configuration change: Set 'model-check/replay' to '2;2;3;2;3;3'
801 [0.000000] [mc_record/INFO] path=2;2;3;2;3;3
802 All threads are started.
803 [0.000000] [mc_record/INFO] ***********************************************************************************
804 [0.000000] [mc_record/INFO] * Path chunk #1 '2/0' Actor thread 1(pid:2): MUTEX_ASYNC_LOCK(mutex_id:0 owner:none)
805 [0.000000] [mc_record/INFO] ***********************************************************************************
806 Backtrace (displayed in actor thread 1):
807 -> #0 simgrid::s4u::Mutex::lock() at ../../src/s4u/s4u_Mutex.cpp:26
808 -> #1 sthread_mutex_lock at ../../src/sthread/sthread_impl.cpp:188
809 -> #2 pthread_mutex_lock at ../../src/sthread/sthread.c:141
810 -> #3 thread_fun1 at ../../examples/sthread/pthread-mutex-simpledeadlock.c:21
812 [0.000000] [mc_record/INFO] ***********************************************************************************
813 [0.000000] [mc_record/INFO] * Path chunk #2 '2/0' Actor thread 1(pid:2): MUTEX_WAIT(mutex_id:0 owner:2)
814 [0.000000] [mc_record/INFO] ***********************************************************************************
815 Backtrace (displayed in actor thread 1):
816 -> #0 simgrid::s4u::Mutex::lock() at ../../src/s4u/s4u_Mutex.cpp:29
817 -> #1 sthread_mutex_lock at ../../src/sthread/sthread_impl.cpp:188
818 -> #2 pthread_mutex_lock at ../../src/sthread/sthread.c:141
819 -> #3 thread_fun1 at ../../examples/sthread/pthread-mutex-simpledeadlock.c:21
821 [0.000000] [mc_record/INFO] ***********************************************************************************
822 [0.000000] [mc_record/INFO] * Path chunk #3 '3/0' Actor thread 2(pid:3): MUTEX_ASYNC_LOCK(mutex_id:1 owner:none)
823 [0.000000] [mc_record/INFO] ***********************************************************************************
824 Backtrace (displayed in actor thread 2):
825 -> #0 simgrid::s4u::Mutex::lock() at ../../src/s4u/s4u_Mutex.cpp:26
826 -> #1 sthread_mutex_lock at ../../src/sthread/sthread_impl.cpp:188
827 -> #2 pthread_mutex_lock at ../../src/sthread/sthread.c:141
828 -> #3 thread_fun2 at ../../examples/sthread/pthread-mutex-simpledeadlock.c:31
830 [0.000000] [mc_record/INFO] ***********************************************************************************
831 [0.000000] [mc_record/INFO] * Path chunk #4 '2/0' Actor thread 1(pid:2): MUTEX_ASYNC_LOCK(mutex_id:1 owner:3)
832 [0.000000] [mc_record/INFO] ***********************************************************************************
833 Backtrace (displayed in actor thread 1):
834 -> #0 simgrid::s4u::Mutex::lock() at ../../src/s4u/s4u_Mutex.cpp:26
835 -> #1 sthread_mutex_lock at ../../src/sthread/sthread_impl.cpp:188
836 -> #2 pthread_mutex_lock at ../../src/sthread/sthread.c:141
837 -> #3 thread_fun1 at ../../examples/sthread/pthread-mutex-simpledeadlock.c:22
839 [0.000000] [mc_record/INFO] ***********************************************************************************
840 [0.000000] [mc_record/INFO] * Path chunk #5 '3/0' Actor thread 2(pid:3): MUTEX_WAIT(mutex_id:1 owner:3)
841 [0.000000] [mc_record/INFO] ***********************************************************************************
842 Backtrace (displayed in actor thread 2):
843 -> #0 simgrid::s4u::Mutex::lock() at ../../src/s4u/s4u_Mutex.cpp:29
844 -> #1 sthread_mutex_lock at ../../src/sthread/sthread_impl.cpp:188
845 -> #2 pthread_mutex_lock at ../../src/sthread/sthread.c:141
846 -> #3 thread_fun2 at ../../examples/sthread/pthread-mutex-simpledeadlock.c:31
848 [0.000000] [mc_record/INFO] ***********************************************************************************
849 [0.000000] [mc_record/INFO] * Path chunk #6 '3/0' Actor thread 2(pid:3): MUTEX_ASYNC_LOCK(mutex_id:0 owner:2)
850 [0.000000] [mc_record/INFO] ***********************************************************************************
851 Backtrace (displayed in actor thread 2):
852 -> #0 simgrid::s4u::Mutex::lock() at ../../src/s4u/s4u_Mutex.cpp:26
853 -> #1 sthread_mutex_lock at ../../src/sthread/sthread_impl.cpp:188
854 -> #2 pthread_mutex_lock at ../../src/sthread/sthread.c:141
855 -> #3 thread_fun2 at ../../examples/sthread/pthread-mutex-simpledeadlock.c:32
857 [0.000000] [mc_record/INFO] The replay of the trace is complete. DEADLOCK detected.
858 [0.000000] [ker_engine/INFO] 3 actors are still running, waiting for something.
859 [0.000000] [ker_engine/INFO] Legend of the following listing: "Actor <pid> (<name>@<host>): <status>"
860 [0.000000] [ker_engine/INFO] Actor 1 (main thread@Lilibeth) simcall ActorJoin(pid:2)
861 [0.000000] [ker_engine/INFO] Actor 2 (thread 1@Lilibeth) simcall MUTEX_WAIT(mutex_id:1 owner:3)
862 [0.000000] [ker_engine/INFO] Actor 3 (thread 2@Lilibeth) simcall MUTEX_WAIT(mutex_id:0 owner:2)
863 [0.000000] [sthread/INFO] All threads exited. Terminating the simulation.
864 [0.000000] ../../src/kernel/EngineImpl.cpp:265: [ker_engine/WARNING] Process called exit when leaving - Skipping cleanups
865 [0.000000] ../../src/kernel/EngineImpl.cpp:265: [ker_engine/WARNING] Process called exit when leaving - Skipping cleanups
867 Currently, if the path is of the form ``X;Y;Z``, each number denotes
868 the actor's pid that is selected at each indecision point. If it's of
869 the form ``X/a;Y/b``, the X and Y are the selected pids while the a
870 and b are the return values of their simcalls. In the previous
871 example, ``1/3;1/4``, you can see from the full output that the actor
872 1 is doing MC_RANDOM simcalls, so the 3 and 4 simply denote the values
873 that these simcall return on the execution branch leading to the
876 Configuring the User Code Virtualization
877 ----------------------------------------
879 .. _cfg=contexts/factory:
881 Selecting the Virtualization Factory
882 ....................................
884 **Option** contexts/factory **Default:** "raw"
886 In SimGrid, the user code is virtualized in a specific mechanism that
887 allows the simulation kernel to control its execution: when a user
888 process requires a blocking action (such as sending a message), it is
889 interrupted, and only gets released when the simulated clock reaches
890 the point where the blocking operation is done. This is explained
891 graphically in the `relevant tutorial, available online
892 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
894 In SimGrid, the containers in which user processes are virtualized are
895 called contexts. Several context factory are provided, and you can
896 select the one you want to use with the ``contexts/factory``
897 configuration item. Some of the following may not exist on your
898 machine because of portability issues. In any case, the default one
899 should be the most effcient one (please report bugs if the
900 auto-detection fails for you). They are approximately sorted here from
901 the slowest to the most efficient:
903 - **thread:** very slow factory using full featured, standard threads.
904 They are slow but very standard. Some debuggers or profilers only work with this factory.
905 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
906 - **boost:** This uses the `context
907 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
908 of the boost library for a performance that is comparable to our
910 |br| Install the relevant library (e.g. with the
911 libboost-contexts-dev package on Debian/Ubuntu) and recompile
913 - **raw:** amazingly fast factory using a context switching mechanism
914 of our own, directly implemented in assembly (only available for x86
915 and amd64 platforms for now) and without any unneeded system call.
917 The main reason to change this setting is when the debugging tools become
918 fooled by the optimized context factories. Threads are the most
919 debugging-friendly contexts, as they allow one to set breakpoints
920 anywhere with gdb and visualize backtraces for all processes, in order
921 to debug concurrency issues. Valgrind is also more comfortable with
922 threads, but it should be usable with all factories (Exception: the
923 callgrind tool really dislikes raw and ucontext factories).
925 .. _cfg=contexts/stack-size:
927 Adapting the Stack Size
928 .......................
930 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
932 Each virtualized used process is executed using a specific system
933 stack. The size of this stack has a huge impact on the simulation
934 scalability, but its default value is rather large. This is because
935 the error messages that you get when the stack size is too small are
936 rather disturbing: this leads to stack overflow (overwriting other
937 stacks), leading to segfaults with corrupted stack traces.
939 If you want to push the scalability limits of your code, you might
940 want to reduce the ``contexts/stack-size`` item. Its default value is
941 8192 (in KiB), while our Chord simulation works with stacks as small
942 as 16 KiB, for example. You can ensure that some actors have a specific
943 size by simply changing the value of this configuration item before
944 creating these actors. The :cpp:func:`simgrid::s4u::Engine::set_config`
945 functions are handy for that.
947 This *setting is ignored* when using the thread factory (because there
948 is no way to modify the stack size with C++ system threads). Instead,
949 you should compile SimGrid and your application with
950 ``-fsplit-stack``. Note that this compilation flag is not compatible
951 with the model checker right now.
953 The operating system should only allocate memory for the pages of the
954 stack which are actually used and you might not need to use this in
955 most cases. However, this setting is very important when using the
956 model checker (see :ref:`options_mc_perf`).
958 .. _cfg=contexts/guard-size:
960 Disabling Stack Guard Pages
961 ...........................
963 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages with MC)
965 Unless you use the threads context factory (see
966 :ref:`cfg=contexts/factory`), a stack guard page is usually used
967 which prevents the stack of a given actor from overflowing on another
968 stack. But the performance impact may become prohibitive when the
969 amount of actors increases. The option ``contexts/guard-size`` is the
970 number of stack guard pages used. By setting it to 0, no guard pages
971 will be used: in this case, you should avoid using small stacks (with
972 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
973 will silently overflow on other parts of the memory.
975 When no stack guard page is created, stacks may then silently overflow
976 on other parts of the memory if their size is too small for the
979 .. _cfg=contexts/nthreads:
980 .. _cfg=contexts/synchro:
982 Running User Code in Parallel
983 .............................
985 Parallel execution of the user code is only considered stable in
986 SimGrid v3.7 and higher, and mostly for S4U simulations. SMPI
987 simulations may well fail in parallel mode. It is described in
988 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
990 If you are using the **ucontext** or **raw** context factories, you can
991 request to execute the user code in parallel. Several threads are
992 launched, each of them handling the same number of user contexts at each
993 run. To activate this, set the ``contexts/nthreads`` item to the amount
994 of cores that you have in your computer (or lower than 1 to have the
995 amount of cores auto-detected).
997 When parallel execution is activated, you can choose the
998 synchronization schema used with the ``contexts/synchro`` item,
999 which value is either:
1001 - **futex:** ultra optimized synchronisation schema, based on futexes
1002 (fast user-mode mutexes), and thus only available on Linux systems.
1003 This is the default mode when available.
1004 - **posix:** slow but portable synchronisation using only POSIX
1006 - **busy_wait:** not really a synchronisation: the worker threads
1007 constantly request new contexts to execute. It should be the most
1008 efficient synchronisation schema, but it loads all the cores of
1009 your machine for no good reason. You probably prefer the other less
1012 Configuring the Tracing
1013 -----------------------
1015 The :ref:`tracing subsystem <outcome_vizu>` can be configured in
1016 several different ways depending on the used interface (S4U, SMPI)
1017 and the kind of traces that needs to be obtained. See the
1018 :ref:`Tracing Configuration Options subsection
1019 <tracing_tracing_options>` for a full description of each
1020 configuration option.
1022 We detail here a simple way to get the traces working for you, even if
1023 you never used the tracing API.
1026 - Any SimGrid-based simulator (S4U, SMPI, ...) and raw traces:
1028 .. code-block:: none
1030 --cfg=tracing:yes --cfg=tracing/uncategorized:yes
1032 The first parameter activates the tracing subsystem, and the second
1033 tells it to trace host and link utilization (without any
1036 - S4U-based simulator and categorized traces (you need to
1037 declare categories and classify your tasks according to them)
1039 .. code-block:: none
1041 --cfg=tracing:yes --cfg=tracing/categorized:yes
1043 The first parameter activates the tracing subsystem, and the second
1044 tells it to trace host and link categorized utilization.
1046 - SMPI simulator and traces for a space/time view:
1048 .. code-block:: console
1050 $ smpirun -trace ...
1052 The `-trace` parameter for the smpirun script runs the simulation
1053 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
1054 smpirun's `-help` parameter for additional tracing options.
1056 Sometimes you might want to put additional information on the trace to
1057 correctly identify them later, or to provide data that can be used to
1058 reproduce an experiment. You have two ways to do that:
1060 - Add a string on top of the trace file as comment:
1062 .. code-block:: none
1064 --cfg=tracing/comment:my_simulation_identifier
1066 - Add the contents of a textual file on top of the trace file as comment:
1068 .. code-block:: none
1070 --cfg=tracing/comment-file:my_file_with_additional_information.txt
1072 Please, use these two parameters (for comments) to make reproducible
1073 simulations. For additional details about this and all tracing
1074 options, check See the :ref:`tracing_tracing_options`.
1079 The SMPI interface provides several specific configuration items.
1080 These are not easy to see with ``--help-cfg``, since SMPI binaries are usually launched through the ``smiprun`` script.
1082 .. _cfg=smpi/host-speed:
1083 .. _cfg=smpi/cpu-threshold:
1084 .. _cfg=smpi/simulate-computation:
1086 Automatic Benchmarking of SMPI Code
1087 ...................................
1089 In SMPI, the sequential code is automatically benchmarked, and these
1090 computations are automatically reported to the simulator. That is to
1091 say that if you have a large computation between a ``MPI_Recv()`` and
1092 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
1093 this code, and create an execution task within the simulator to take
1094 this into account. For that, the actual duration is measured on the
1095 host machine and then scaled to the power of the corresponding
1096 simulated machine. The variable ``smpi/host-speed`` allows one to
1097 specify the computational speed of the host machine (in flop/s by
1098 default) to use when scaling the execution times.
1100 The default value is ``smpi/host-speed=20kf`` (= 20,000 flop/s). This
1101 is probably underestimated for most machines, leading SimGrid to
1102 overestimate the amount of flops in the execution blocks that are
1103 automatically injected in the simulator. As a result, the execution
1104 time of the whole application will probably be overestimated until you
1105 use a realistic value.
1107 When the code consists of numerous consecutive MPI calls, the
1108 previous mechanism feeds the simulation kernel with numerous tiny
1109 computations. The ``smpi/cpu-threshold`` item becomes handy when this
1110 impacts badly on the simulation performance. It specifies a threshold (in
1111 seconds) below which the execution chunks are not reported to the
1112 simulation kernel (default value: 1e-6).
1114 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
1115 time spent below this threshold. SMPI does not consider the
1116 `amount of time` of these computations; there is no offset for
1117 this. Hence, a value that is too small, may lead to unreliable
1120 In some cases, however, one may wish to disable simulation of
1121 the computation of an application. This is the case when SMPI is used not to
1122 simulate an MPI application, but instead an MPI code that performs
1123 "live replay" of another MPI app (e.g., ScalaTrace's replay tool, or
1124 various on-line simulators that run an app at scale). In this case the
1125 computation of the replay/simulation logic should not be simulated by
1126 SMPI. Instead, the replay tool or on-line simulator will issue
1127 "computation events", which correspond to the actual MPI simulation
1128 being replayed/simulated. At the moment, these computation events can
1129 be simulated using SMPI by calling internal smpi_execute*() functions.
1131 To disable the benchmarking/simulation of a computation in the simulated
1132 application, the variable ``smpi/simulate-computation`` should be set
1133 to **no**. This option just ignores the timings in your simulation; it
1134 still executes the computations itself. If you want to stop SMPI from
1135 doing that, you should check the SMPI_SAMPLE macros, documented in
1136 Section :ref:`SMPI_use_faster`.
1138 +------------------------------------+-------------------------+-----------------------------+
1139 | Solution | Computations executed? | Computations simulated? |
1140 +====================================+=========================+=============================+
1141 | --cfg=smpi/simulate-computation:no | Yes | Never |
1142 +------------------------------------+-------------------------+-----------------------------+
1143 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
1144 +------------------------------------+-------------------------+-----------------------------+
1145 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
1146 +------------------------------------+-------------------------+-----------------------------+
1148 .. _cfg=smpi/comp-adjustment-file:
1150 Slow-down or speed-up parts of your code
1151 ........................................
1153 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
1155 This option allows you to pass a file that contains two columns: The
1156 first column defines the section that will be subject to a speedup;
1157 the second column is the speedup. For instance:
1159 .. code-block:: none
1161 "start:stop","ratio"
1162 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1164 The first line is the header - you must include it. The following
1165 line means that the code between two consecutive MPI calls on line 30
1166 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1167 of 1.18244559422142. The value for the second column is therefore a
1168 speedup, if it is larger than 1 and a slowdown if it is smaller
1169 than 1. Nothing will be changed if it is equal to 1.
1171 Of course, you can set any arbitrary filenames you want (so the start
1172 and end don't have to be in the same file), but be aware that this
1173 mechanism only supports `consecutive calls!`
1175 Please note that you must pass the ``-trace-call-location`` flag to
1176 smpicc or smpiff, respectively. This flag activates some internal
1177 macro definitions that help with obtaining the call location.
1179 Bandwidth and latency factors
1180 .............................
1182 Adapting the bandwidth and latency acurately to the network conditions is of a paramount importance to get realistic results.
1183 This is done through the :ref:`network/bandwidth-factor <cfg=network/bandwidth-factor>` and :ref:`network/latency-factor
1184 <cfg=network/latency-factor>` items. You probably also want to read the following section: :ref:`models_calibration`.
1186 .. _cfg=smpi/display-timing:
1188 Reporting Simulation Time
1189 .........................
1191 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1193 Most of the time, you run MPI code with SMPI to compute the time it
1194 would take to run it on a platform. But since the code is run through
1195 the ``smpirun`` script, you don't have any control on the launcher
1196 code, making it difficult to report the simulated time when the
1197 simulation ends. If you enable the ``smpi/display-timing`` item,
1198 ``smpirun`` will display this information when the simulation
1200 SMPI will also display information about the amout of real time spent
1201 in application code and in SMPI internals, to provide hints about the
1202 need to use sampling to reduce simulation time.
1204 .. _cfg=smpi/display-allocs:
1206 Reporting memory allocations
1207 ............................
1209 **Option** ``smpi/display-allocs`` **Default:** 0 (false)
1211 SMPI intercepts malloc and calloc calls performed inside the running
1212 application, if it wasn't compiled with SMPI_NO_OVERRIDE_MALLOC.
1213 With this option, SMPI will show at the end of execution the amount of
1214 memory allocated through these calls, and locate the most expensive one.
1215 This helps finding the targets for manual memory sharing, or the threshold
1216 to use for smpi/auto-shared-malloc-thresh option (see :ref:`cfg=smpi/auto-shared-malloc-thresh`).
1218 .. _cfg=smpi/keep-temps:
1220 Keeping temporary files after simulation
1221 ........................................
1223 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1225 SMPI usually generates a lot of temporary files that are cleaned after
1226 use. This option requests to preserve them, for example to debug or
1227 profile your code. Indeed, the binary files are removed very early
1228 under the dlopen privatization schema, which tends to fool the
1231 .. _cfg=smpi/papi-events:
1233 Trace hardware counters with PAPI
1234 .................................
1236 **Option** ``smpi/papi-events`` **default:** unset
1238 When the PAPI support is compiled into SimGrid, this option takes the
1239 names of PAPI counters and adds their respective values to the trace
1240 files (See Section :ref:`tracing_tracing_options`).
1244 This feature currently requires superuser privileges, as registers
1245 are queried. Only use this feature with code you trust! Call
1246 smpirun for instance via ``smpirun -wrapper "sudo "
1247 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1248 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1249 will not be required.
1251 It is planned to make this feature available on a per-process (or per-thread?) basis.
1252 The first draft, however, just implements a "global" (i.e., for all processes) set
1253 of counters, the "default" set.
1255 .. code-block:: none
1257 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1259 .. _cfg=smpi/privatization:
1261 Automatic Privatization of Global Variables
1262 ...........................................
1264 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1266 MPI executables are usually meant to be executed in separate
1267 processes, but SMPI is executed in only one process. Global variables
1268 from executables will be placed in the same memory region and shared
1269 between processes, causing intricate bugs. Several options are
1270 possible to avoid this, as described in the main `SMPI publication
1271 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1272 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1273 automatically privatizing the globals, and this option allows one to
1274 choose between them.
1276 - **no** (default when not using smpirun): Do not automatically
1277 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1279 - **dlopen** or **yes** (default when using smpirun): Link multiple
1280 times against the binary.
1281 - **mmap** (slower, but maybe somewhat more stable):
1282 Runtime automatic switching of the data segments.
1285 This configuration option cannot be set in your platform file. You can only
1286 pass it as an argument to smpirun.
1288 .. _cfg=smpi/privatize-libs:
1290 Automatic privatization of global variables inside external libraries
1291 .....................................................................
1293 **Option** ``smpi/privatize-libs`` **default:** unset
1295 **Linux/BSD only:** When using dlopen (default) privatization,
1296 privatize specific shared libraries with internal global variables, if
1297 they can't be linked statically. For example libgfortran is usually
1298 used for Fortran I/O and indexes in files can be mixed up.
1300 Multiple libraries can be given, semicolon separated.
1302 This configuration option can only use either full paths to libraries,
1303 or full names. Check with ldd the name of the library you want to
1306 .. code-block:: console
1310 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1313 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1314 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1315 but not ``libgfortran`` nor ``libgfortran.so``.
1317 .. _cfg=smpi/send-is-detached-thresh:
1319 Simulating MPI detached send
1320 ............................
1322 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1324 This threshold specifies the size in bytes under which the send will
1325 return immediately. This is different from the threshold detailed in
1326 :ref:`cfg=smpi/async-small-thresh` because the message is not
1327 really sent when the send is posted. SMPI still waits for the
1328 corresponding receive to be posted, in order to perform the communication
1331 .. _cfg=smpi/coll-selector:
1333 Simulating MPI collective algorithms
1334 ....................................
1336 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1338 SMPI implements more than 100 different algorithms for MPI collective
1339 communication, to accurately simulate the behavior of most of the
1340 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1341 select the decision logic either of the OpenMPI or the MPICH libraries. (By
1342 default SMPI uses naive version of collective operations.)
1344 Each collective operation can be manually selected with a ``smpi/collective_name:algo_name``. For example, if you want to use
1345 the Bruck algorithm for the Alltoall algorithm, you should use ``--cfg=smpi/alltoall:bruck`` on the command-line of smpirun. The
1346 reference of all available algorithms are listed in :ref:`SMPI_use_colls`, and you can get the full list implemented in your
1347 version using ``smpirun --help-coll``.
1349 .. _cfg=smpi/barrier-collectives:
1351 Add a barrier in all collectives
1352 ................................
1354 **Option** ``smpi/barrier-collectives`` **default:** off
1356 This option adds a simple barrier in some collective operations to catch dangerous
1357 code that may or may not work depending on the MPI implementation: Bcast, Exscan,
1358 Gather, Gatherv, Scan, Scatter, Scatterv and Reduce.
1360 For example, the following code works with OpenMPI while it deadlocks in MPICH and
1361 Intel MPI. Broadcast seem to be "fire and forget" in OpenMPI while other
1362 implementations expect to receive a message.
1367 MPI_Bcast(buf1, buff_size, MPI_CHAR, 0, newcom);
1368 MPI_Send(&buf2, buff_size, MPI_CHAR, 1, tag, newcom);
1369 } else if (rank==1) {
1370 MPI_Recv(&buf2, buff_size, MPI_CHAR, 0, tag, newcom, MPI_STATUS_IGNORE);
1371 MPI_Bcast(buf1, buff_size, MPI_CHAR, 0, newcom);
1374 The barrier is only simulated and does not involve any additional message (it is a S4U barrier).
1375 This option is disabled by default, and activated by the `-analyze` flag of smpirun.
1377 .. _cfg=smpi/barrier-finalization:
1379 Add a barrier in MPI_Finalize
1380 .............................
1382 **Option** ``smpi/finalization-barrier`` **default:** off
1384 By default, SMPI processes are destroyed as soon as soon as their code ends,
1385 so after a successful MPI_Finalize call returns. In some rare cases, some data
1386 might have been attached to MPI objects still active in the remaining processes,
1387 and can be destroyed eagerly by the finished process.
1388 If your code shows issues at finalization, such as segmentation fault, triggering
1389 this option will add an explicit MPI_Barrier(MPI_COMM_WORLD) call inside the
1390 MPI_Finalize, so that all processes will terminate at almost the same point.
1391 It might affect the total timing by the cost of a barrier.
1393 .. _cfg=smpi/errors-are-fatal:
1395 Disable MPI fatal errors
1396 ........................
1398 **Option** ``smpi/errors-are-fatal`` **default:** on
1400 By default, SMPI processes will crash if a MPI error code is returned. MPI allows
1401 to explicitely set MPI_ERRORS_RETURN errhandler to avoid this behaviour. This flag
1402 will turn on this behaviour by default (for all concerned types and errhandlers).
1403 This can ease debugging by going after the first reported error.
1405 .. _cfg=smpi/pedantic:
1407 Disable pedantic MPI errors
1408 ...........................
1410 **Option** ``smpi/pedantic`` **default:** on
1412 By default, SMPI will report all errors it finds in MPI codes. Some of these errors
1413 may not be considered as errors by all developers. This flag can be turned off to
1414 avoid reporting some usually harmless mistakes.
1415 Concerned errors list (will be expanded in the future):
1417 - Calling MPI_Win_fence only once in a program, hence just opening an epoch without
1420 .. _cfg=smpi/iprobe:
1422 Inject constant times for MPI_Iprobe
1423 ....................................
1425 **Option** ``smpi/iprobe`` **default:** 0.0001
1427 The behavior and motivation for this configuration option is identical
1428 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1431 .. _cfg=smpi/iprobe-cpu-usage:
1433 Reduce speed for iprobe calls
1434 .............................
1436 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1438 MPI_Iprobe calls can be heavily used in applications. To account
1439 correctly for the energy that cores spend probing, it is necessary to
1440 reduce the load that these calls cause inside SimGrid.
1442 For instance, we measured a maximum power consumption of 220 W for a
1443 particular application but only 180 W while this application was
1444 probing. Hence, the correct factor that should be passed to this
1445 option would be 180/220 = 0.81.
1449 Inject constant times for MPI_Init
1450 ..................................
1452 **Option** ``smpi/init`` **default:** 0
1454 The behavior and motivation for this configuration option is identical
1455 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1459 Inject constant times for MPI_Isend()
1460 .....................................
1462 **Option** ``smpi/ois``
1464 The behavior and motivation for this configuration option is identical
1465 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1469 Inject constant times for MPI_send()
1470 ....................................
1472 **Option** ``smpi/os``
1474 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1475 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1476 time). SMPI can factor these costs in as well, but the user has to
1477 configure SMPI accordingly as these values may vary by machine. This
1478 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1479 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1480 exactly as ``smpi/ois``.
1482 This item can consist of multiple sections; each section takes three
1483 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1484 so this example contains two sections. Furthermore, each section
1485 consists of three values.
1487 1. The first value denotes the minimum size in bytes for this section to take effect;
1488 read it as "if message size is greater than this value (and other section has a larger
1489 first value that is also smaller than the message size), use this".
1490 In the first section above, this value is "1".
1492 2. The second value is the startup time; this is a constant value that will always
1493 be charged, no matter what the size of the message. In the first section above,
1496 3. The third value is the `per-byte` cost. That is, it is charged for every
1497 byte of the message (incurring cost messageSize*cost_per_byte)
1498 and hence accounts also for larger messages. In the first
1499 section of the example above, this value is "2".
1501 Now, SMPI always checks which section it should use for a given
1502 message; that is, if a message of size 11 is sent with the
1503 configuration of the example above, only the second section will be
1504 used, not the first, as the first value of the second section is
1505 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1506 message of size 11 incurs the following cost inside MPI_Send:
1507 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1509 Note that the order of sections can be arbitrary; they will be ordered internally.
1513 Inject constant times for MPI_Recv()
1514 ....................................
1516 **Option** ``smpi/or``
1518 The behavior and motivation for this configuration option is identical
1519 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1522 .. _cfg=smpi/grow-injected-times:
1524 Inject constant times for MPI_Test
1525 ..................................
1527 **Option** ``smpi/test`` **default:** 0.0001
1529 By setting this option, you can control the amount of time a process
1530 sleeps when MPI_Test() is called; this is important, because SimGrid
1531 normally only advances the time while communication is happening and
1532 thus, MPI_Test will not add to the time, resulting in deadlock if it is
1533 used as a break-condition as in the following example:
1538 MPI_Test(request, flag, status);
1542 To speed up execution, we use a counter to keep track of how often we
1543 checked if the handle is now valid or not. Hence, we actually
1544 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1545 process to sleep increases linearly with the number of previously
1546 failed tests. This behavior can be disabled by setting
1547 ``smpi/grow-injected-times`` to **no**. This will also disable this
1548 behavior for MPI_Iprobe.
1550 .. _cfg=smpi/shared-malloc:
1551 .. _cfg=smpi/shared-malloc-hugepage:
1556 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1558 If your simulation consumes too much memory, you may want to modify
1559 your code so that the working areas are shared by all MPI ranks. For
1560 example, in a block-cyclic matrix multiplication, you will only
1561 allocate one set of blocks, and all processes will share them.
1562 Naturally, this will lead to very wrong results, but this will save a
1563 lot of memory. So this is still desirable for some studies. For more on
1564 the motivation for that feature, please refer to the `relevant section
1565 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1566 of the SMPI CourseWare (see Activity #2.2 of the pointed
1567 assignment). In practice, change the calls for malloc() and free() into
1568 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1570 SMPI provides two algorithms for this feature. The first one, called
1571 ``local``, allocates one block per call to SMPI_SHARED_MALLOC()
1572 (each call site gets its own block) ,and this block is shared
1573 among all MPI ranks. This is implemented with the shm_* functions
1574 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1575 for each shared block.
1577 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1578 returns a new address, but it only points to a shadow block: its memory
1579 area is mapped on a 1 MiB file on disk. If the returned block is of size
1580 N MiB, then the same file is mapped N times to cover the whole block.
1581 At the end, no matter how many times you call SMPI_SHARED_MALLOC, this will
1582 only consume 1 MiB in memory.
1584 You can disable this behavior and come back to regular mallocs (for
1585 example for debugging purposes) using ``no`` as a value.
1587 If you want to keep private some parts of the buffer, for instance if these
1588 parts are used by the application logic and should not be corrupted, you
1589 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). For example:
1593 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1595 This will allocate 500 bytes to mem, such that mem[27..41] and
1596 mem[100..199] are shared while other area remain private.
1598 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1600 When smpi/shared-malloc:global is used, the memory consumption problem
1601 is solved, but it may induce too much load on the kernel's pages table.
1602 In this case, you should use huge pages so that the kernel creates only one
1603 entry per MB of malloced data instead of one entry per 4 kB.
1604 To activate this, you must mount a hugetlbfs on your system and allocate
1605 at least one huge page:
1607 .. code-block:: console
1610 $ sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1611 $ sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1613 Then, you can pass the option
1614 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1615 actually activate the huge page support in shared mallocs.
1617 .. _cfg=smpi/auto-shared-malloc-thresh:
1619 Automatically share allocations
1620 ...............................
1622 **Option** ``smpi/auto-shared-malloc-thresh:`` **Default:** 0 (false)
1623 This value in bytes represents the size above which all allocations
1624 will be "shared" by default (as if they were performed through
1625 SMPI_SHARED_MALLOC macros). Default = 0 = disabled feature.
1626 The value must be carefully chosen to only select data buffers which
1627 will not modify execution path or cause crash if their content is false.
1628 Option :ref:`cfg=smpi/display-allocs` can be used to locate the largest
1629 allocation detected in a run, and provide a good starting threshold.
1630 Note : malloc, calloc and free are overridden by smpicc/cxx by default.
1631 This can cause some troubles if codes are already overriding these. If this
1632 is the case, defining SMPI_NO_OVERRIDE_MALLOC in the compilation flags can
1633 help, but will make this feature unusable.
1637 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1638 ...................................................................
1640 **Option** ``smpi/wtime`` **default:** 10 ns
1642 This option controls the amount of (simulated) time spent in calls to
1643 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1644 to 0, the simulated clock is not advanced in these calls, which leads
1645 to issues if your application contains such a loop:
1649 while(MPI_Wtime() < some_time_bound) {
1650 /* some tests, with no communication nor computation */
1653 When the option smpi/wtime is set to 0, the time advances only on
1654 communications and computations. So the previous code results in an
1655 infinite loop: the current [simulated] time will never reach
1656 ``some_time_bound``. This infinite loop is avoided when that option
1657 is set to a small value, as it is by default since SimGrid v3.21.
1659 Note that if your application does not contain any loop depending on
1660 the current time only, then setting this option to a non-zero value
1661 will slow down your simulations by a tiny bit: the simulation loop has
1662 to be broken out of and reset each time your code asks for the current time.
1663 If the simulation speed really matters to you, you can avoid this
1664 extra delay by setting smpi/wtime to 0.
1666 .. _cfg=smpi/list-leaks:
1668 Report leaked MPI objects
1669 .........................
1671 **Option** ``smpi/list-leaks`` **default:** 0
1673 This option controls whether to report leaked MPI objects.
1674 The parameter is the number of leaks to report.
1676 Other Configurations
1677 --------------------
1679 .. _cfg=debug/clean-atexit:
1681 Cleanup at Termination
1682 ......................
1684 **Option** ``debug/clean-atexit`` **default:** on
1686 If your code is segfaulting during its finalization, it may help to
1687 disable this option to request that SimGrid not attempt any cleanups at
1688 the end of the simulation. Since the Unix process is ending anyway,
1689 the operating system will wipe it all.
1696 **Option** ``path`` **default:** . (current dir)
1698 It is possible to specify a list of directories to search in for the
1699 trace files (see :ref:`pf_trace`) by using this configuration
1700 item. To add several directory to the path, set the configuration
1701 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1703 .. _cfg=debug/breakpoint:
1708 **Option** ``debug/breakpoint`` **default:** unset
1710 This configuration option sets a breakpoint: when the simulated clock
1711 reaches the given time, a SIGTRAP is raised. This can be used to stop
1712 the execution and get a backtrace with a debugger.
1714 It is also possible to set the breakpoint from inside the debugger, by
1715 writing in global variable simgrid::kernel::cfg_breakpoint. For example,
1718 .. code-block:: none
1720 set variable simgrid::kernel::cfg_breakpoint = 3.1416
1722 .. _cfg=debug/verbose-exit:
1727 **Option** ``debug/verbose-exit`` **default:** on
1729 By default, when Ctrl-C is pressed, the status of all existing actors
1730 is displayed before exiting the simulation. This is very useful to
1731 debug your code, but it can become troublesome if you have many
1732 actors. Set this configuration item to **off** to disable this
1735 .. _cfg=exception/cutpath:
1737 Truncate local path from exception backtrace
1738 ............................................
1740 **Option** ``exception/cutpath`` **default:** off
1742 This configuration option is used to remove the path from the
1743 backtrace shown when an exception is thrown. This is mainly useful for
1744 the tests: the full file path would makes the tests non-reproducible because
1745 the paths of source files depend of the build settings. That would
1746 break most of the tests since their output is continually compared.
1750 Logging configuration
1751 ---------------------
1753 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
1754 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
1755 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
1758 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
1759 messages from your code.
1761 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
1762 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
1763 practice, the following is equivalent to the above settings: ``--log=root.thresh:error --log=s4u_host.thresh:debug``.
1765 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
1766 your settings, as in ``--log="root.thresh:error s4u_host.thresh:debug"``. The parameters are interpreted in order, from left to right.
1769 Threshold configuration
1770 .......................
1772 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
1773 ``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
1774 see, ``threshold`` can be abbreviated here.
1776 Existing thresholds:
1778 - ``trace`` some functions display a message at this level when entering or returning
1779 - ``debug`` output that is mostly useful when debugging the corresponding module.
1780 - ``verbose`` verbose output that is only mildly interesting and can easily be ignored
1781 - ``info`` usual output (this is the default threshold of all categories)
1782 - ``warning`` minor issue encountered
1783 - ``error`` issue encountered
1784 - ``critical`` major issue encountered, such as assertions failures
1788 Format configuration
1789 ....................
1791 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
1792 as the date, or the actor ID, everything. Existing format directives:
1795 - %n: line separator (LOG4J compatible)
1796 - %e: plain old space (SimGrid extension)
1798 - %m: user-provided message
1800 - %c: Category name (LOG4J compatible)
1801 - %p: Priority name (LOG4J compatible)
1803 - %h: Hostname (SimGrid extension)
1804 - %a: Actor name (SimGrid extension -- note that with SMPI this is the integer value of the process rank)
1805 - %i: Actor PID (SimGrid extension -- this is a 'i' as in 'i'dea)
1806 - %t: Thread "name" (LOG4J compatible -- actually the address of the thread in memory)
1808 - %F: file name where the log event was raised (LOG4J compatible)
1809 - %l: location where the log event was raised (LOG4J compatible, like '%%F:%%L' -- this is a l as in 'l'etter)
1810 - %L: line number where the log event was raised (LOG4J compatible)
1811 - %M: function name (LOG4J compatible -- called method name here of course).
1813 - %d: date (UNIX-like epoch)
1814 - %r: application age (time elapsed since the beginning of the application)
1817 ``--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
1818 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
1819 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
1820 provided layout is used for every messages.
1822 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
1826 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
1827 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'"``.
1828 Another option is to use the ``%e`` directive for spaces, as in ``--log=root.fmt:%l:%e[%p/%c]:%e%m%n``.
1833 The keyword ``app`` controls the appended of a logging category. For example ``--log=root.app:file:mylogfile`` redirects every output to the file ``mylogfile``.
1835 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
1836 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.
1838 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``
1839 ensures that the log file ``mylog`` will never overpass 500 bytes in size.
1841 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
1842 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.
1847 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
1848 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
1849 ``on`` (or ``yes`` or ``1``), the produced messages will also be passed to the upper appender.
1851 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
1852 ``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
1853 will only be sent to ``all.log``.
1858 ``--help-logs`` displays a complete help message about logging in SimGrid.
1860 ``--help-log-categories`` displays the actual hierarchy of log categories for this binary.
1862 ``--log=no_loc`` hides the source locations (file names and line numbers) from the messages. This is useful to make tests reproducible.