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:
37 my_simulator --cfg=Item:Value (other arguments)
39 Several ``--cfg`` command line arguments can naturally be used. If you
40 need to include spaces in the argument, don't forget to quote the
41 argument. You can even escape the included quotes (write ``@'`` for ``'`` if
42 you have your argument between simple quotes).
44 Another solution is to use the ``<config>`` tag in the platform file. The
45 only restriction is that this tag must occur before the first
46 platform element (be it ``<zone>``, ``<cluster>``, ``<peer>`` or whatever).
47 The ``<config>`` tag takes an ``id`` attribute, but it is currently
48 ignored so you don't really need to pass it. The important part is that
49 within that tag, you can pass one or several ``<prop>`` tags to specify
50 the configuration to use. For example, setting ``Item`` to ``Value``
51 can be done by adding the following to the beginning of your platform
57 <prop id="Item" value="Value"/>
60 A last solution is to pass your configuration directly in your program
61 with :cpp:func:`simgrid::s4u::Engine::set_config` or :cpp:func:`MSG_config`.
65 #include <simgrid/s4u.hpp>
67 int main(int argc, char *argv[]) {
68 simgrid::s4u::Engine e(&argc, argv);
70 simgrid::s4u::Engine::set_config("Item:Value");
77 Existing Configuration Items
78 ----------------------------
81 The full list can be retrieved by passing ``--help`` and
82 ``--help-cfg`` to an executable that uses SimGrid. Try passing
83 ``help`` as a value to get the list of values accepted by a given
84 option. For example, ``--cfg=plugin:help`` will give you the list
85 of plugins available in your installation of SimGrid.
87 - **contexts/factory:** :ref:`cfg=contexts/factory`
88 - **contexts/guard-size:** :ref:`cfg=contexts/guard-size`
89 - **contexts/nthreads:** :ref:`cfg=contexts/nthreads`
90 - **contexts/stack-size:** :ref:`cfg=contexts/stack-size`
91 - **contexts/synchro:** :ref:`cfg=contexts/synchro`
93 - **cpu/maxmin-selective-update:** :ref:`Cpu Optimization Level <options_model_optim>`
94 - **cpu/model:** :ref:`options_model_select`
95 - **cpu/optim:** :ref:`Cpu Optimization Level <options_model_optim>`
97 - **debug/breakpoint:** :ref:`cfg=debug/breakpoint`
98 - **debug/clean-atexit:** :ref:`cfg=debug/clean-atexit`
99 - **debug/verbose-exit:** :ref:`cfg=debug/verbose-exit`
101 - **exception/cutpath:** :ref:`cfg=exception/cutpath`
103 - **host/model:** :ref:`options_model_select`
105 - **maxmin/precision:** :ref:`cfg=maxmin/precision`
106 - **maxmin/concurrency-limit:** :ref:`cfg=maxmin/concurrency-limit`
108 - **msg/debug-multiple-use:** :ref:`cfg=msg/debug-multiple-use`
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/termination:** :ref:`cfg=model-check/termination`
120 - **model-check/timeout:** :ref:`cfg=model-check/timeout`
121 - **model-check/visited:** :ref:`cfg=model-check/visited`
123 - **network/bandwidth-factor:** :ref:`cfg=network/bandwidth-factor`
124 - **network/crosstraffic:** :ref:`cfg=network/crosstraffic`
125 - **network/latency-factor:** :ref:`cfg=network/latency-factor`
126 - **network/loopback-lat:** :ref:`cfg=network/loopback`
127 - **network/loopback-bw:** :ref:`cfg=network/loopback`
128 - **network/maxmin-selective-update:** :ref:`Network Optimization Level <options_model_optim>`
129 - **network/model:** :ref:`options_model_select`
130 - **network/optim:** :ref:`Network Optimization Level <options_model_optim>`
131 - **network/TCP-gamma:** :ref:`cfg=network/TCP-gamma`
132 - **network/weight-S:** :ref:`cfg=network/weight-S`
134 - **ns3/TcpModel:** :ref:`options_pls`
135 - **ns3/seed:** :ref:`options_pls`
136 - **path:** :ref:`cfg=path`
137 - **plugin:** :ref:`cfg=plugin`
139 - **storage/max_file_descriptors:** :ref:`cfg=storage/max_file_descriptors`
141 - **surf/precision:** :ref:`cfg=surf/precision`
143 - **For collective operations of SMPI,** please refer to Section :ref:`cfg=smpi/coll-selector`
144 - **smpi/auto-shared-malloc-thresh:** :ref:`cfg=smpi/auto-shared-malloc-thresh`
145 - **smpi/async-small-thresh:** :ref:`cfg=smpi/async-small-thresh`
146 - **smpi/buffering:** :ref:`cfg=smpi/buffering`
147 - **smpi/bw-factor:** :ref:`cfg=smpi/bw-factor`
148 - **smpi/coll-selector:** :ref:`cfg=smpi/coll-selector`
149 - **smpi/comp-adjustment-file:** :ref:`cfg=smpi/comp-adjustment-file`
150 - **smpi/cpu-threshold:** :ref:`cfg=smpi/cpu-threshold`
151 - **smpi/display-allocs:** :ref:`cfg=smpi/display-allocs`
152 - **smpi/display-timing:** :ref:`cfg=smpi/display-timing`
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/lat-factor:** :ref:`cfg=smpi/lat-factor`
161 - **smpi/ois:** :ref:`cfg=smpi/ois`
162 - **smpi/or:** :ref:`cfg=smpi/or`
163 - **smpi/os:** :ref:`cfg=smpi/os`
164 - **smpi/papi-events:** :ref:`cfg=smpi/papi-events`
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:`model_ns3`).
228 This model can be :ref:`further configured <options_pls>`.
230 - ``cpu/model``: specify the used CPU model. We have only one model
233 - **Cas01:** Simplistic CPU model (time=size/speed)
235 - ``host/model``: The host concept is the aggregation of a CPU with a
236 network card. Three models exists, but actually, only 2 of them are
237 interesting. The "compound" one is simply due to the way our
238 internal code is organized, and can easily be ignored. So at the
239 end, you have two host models: The default one allows aggregation of
240 an existing CPU model with an existing network model, but does not
241 allow parallel tasks because these beasts need some collaboration
242 between the network and CPU model. That is why, ptask_07 is used by
243 default when using SimDag.
245 - **default:** Default host model. Currently, CPU:Cas01 and
246 network:LV08 (with cross traffic enabled)
247 - **compound:** Host model that is automatically chosen if
248 you change the network and CPU models
249 - **ptask_L07:** Host model somehow similar to Cas01+CM02 but
250 allowing "parallel tasks", that are intended to model the moldable
251 tasks of the grid scheduling literature.
253 - ``storage/model``: specify the used storage model. Only one model is
255 - ``vm/model``: specify the model for virtual machines. Only one model
258 .. todo: make 'compound' the default host model.
260 .. _options_model_optim:
265 The network and CPU models that are based on lmm_solve (that
266 is, all our analytical models) accept specific optimization
269 - items ``network/optim`` and ``cpu/optim`` (both default to 'Lazy'):
271 - **Lazy:** Lazy action management (partial invalidation in lmm +
272 heap in action remaining).
273 - **TI:** Trace integration. Highly optimized mode when using
274 availability traces (only available for the Cas01 CPU model for
276 - **Full:** Full update of remaining and variables. Slow but may be
277 useful when debugging.
279 - items ``network/maxmin-selective-update`` and
280 ``cpu/maxmin-selective-update``: configure whether the underlying
281 should be lazily updated or not. It should have no impact on the
282 computed timings, but should speed up the computation. |br| It is
283 still possible to disable this feature because it can reveal
284 counter-productive in very specific scenarios where the
285 interaction level is high. In particular, if all your
286 communication share a given backbone link, you should disable it:
287 without it, a simple regular loop is used to update each
288 communication. With it, each of them is still updated (because of
289 the dependency induced by the backbone), but through a complicated
290 and slow pattern that follows the actual dependencies.
292 .. _cfg=maxmin/precision:
293 .. _cfg=surf/precision:
298 **Option** ``maxmin/precision`` **Default:** 0.00001 (in flops or bytes) |br|
299 **Option** ``surf/precision`` **Default:** 0.00001 (in seconds)
301 The analytical models handle a lot of floating point values. It is
302 possible to change the epsilon used to update and compare them through
303 this configuration item. Changing it may speedup the simulation by
304 discarding very small actions, at the price of a reduced numerical
305 precision. You can modify separately the precision used to manipulate
306 timings (in seconds) and the one used to manipulate amounts of work
309 .. _cfg=maxmin/concurrency-limit:
314 **Option** ``maxmin/concurrency-limit`` **Default:** -1 (no limit)
316 The maximum number of variables per resource can be tuned through this
317 option. You can have as many simultaneous actions per resources as you
318 want. If your simulation presents a very high level of concurrency, it
319 may help to use e.g. 100 as a value here. It means that at most 100
320 actions can consume a resource at a given time. The extraneous actions
321 are queued and wait until the amount of concurrency of the considered
322 resource lowers under the given boundary.
324 Such limitations help both to the simulation speed and simulation accuracy
325 on highly constrained scenarios, but the simulation speed suffers of this
326 setting on regular (less constrained) scenarios so it is off by default.
328 .. _options_model_network:
330 Configuring the Network Model
331 .............................
333 .. _cfg=network/TCP-gamma:
335 Maximal TCP Window Size
336 ^^^^^^^^^^^^^^^^^^^^^^^
338 **Option** ``network/TCP-gamma`` **Default:** 4194304
340 The analytical models need to know the maximal TCP window size to take
341 the TCP congestion mechanism into account. On Linux, this value can
342 be retrieved using the following commands. Both give a set of values,
343 and you should use the last one, which is the maximal size.
345 .. code-block:: shell
347 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
348 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
350 .. _cfg=network/bandwidth-factor:
351 .. _cfg=network/latency-factor:
352 .. _cfg=network/weight-S:
354 Correcting Important Network Parameters
355 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
357 SimGrid can take network irregularities such as a slow startup or
358 changing behavior depending on the message size into account. You
359 should not change these values unless you really know what you're
360 doing. The corresponding values were computed through data fitting
361 one the timings of packet-level simulators, as described in `Accuracy
362 Study and Improvement of Network Simulation in the SimGrid Framework
363 <http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf>`_.
366 If you are using the SMPI model, these correction coefficients are
367 themselves corrected by constant values depending on the size of the
368 exchange. By default SMPI uses factors computed on the Stampede
369 Supercomputer at TACC, with optimal deployment of processes on
370 nodes. Again, only hardcore experts should bother about this fact.
373 .. todo:: This section should be rewritten, and actually explain the
374 options network/bandwidth-factor, network/latency-factor,
377 .. _cfg=smpi/IB-penalty-factors:
382 InfiniBand network behavior can be modeled through 3 parameters
383 ``smpi/IB-penalty-factors:"βe;βs;γs"``, as explained in `this PhD
385 <http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf>`_ (in French)
386 or more concisely in `this paper <https://hal.inria.fr/hal-00953618/document>`_,
387 even if that paper does only describe models for myrinet and ethernet.
388 You can see in Fig 2 some results for Infiniband, for example. This model
389 may be outdated by now for modern infiniband, anyway, so a new
390 validation would be good.
392 The three paramaters are defined as follows:
394 - βs: penalty factor for outgoing messages, computed by running a simple send to
395 two nodes and checking slowdown compared to a single send to one node,
397 - βe: penalty factor for ingoing messages, same computation method but with one
398 node receiving several messages
399 - γr: slowdown factor when communication buffer memory is saturated. It needs a
400 more complicated pattern to run in order to be computed (5.3 in the thesis,
401 page 107), and formula in the end is γr = time(c)/(3×βe×time(ref)), where
402 time(ref) is the time of a single comm with no contention).
404 Once these values are computed, a penalty is assessed for each message (this is
405 the part implemented in the simulator) as shown page 106 of the thesis. Here is
406 a simple translation of this text. First, some notations:
408 - ∆e(e) which corresponds to the incoming degree of node e, that is to say the number of communications having as destination node e.
409 - ∆s (s) which corresponds to the degree outgoing from node s, that is to say the number of communications sent by node s.
410 - Φ (e) which corresponds to the number of communications destined for the node e but coming from a different node.
411 - 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
413 To determine the penalty for a communication, two values need to be calculated. First, the penalty caused by the conflict in transmission, noted ps.
416 - if ∆s (i) = 1 then ps = 1.
417 - if ∆s (i) ≥ 2 and ∆e (i) ≥ 3 then ps = ∆s (i) × βs × γr
418 - else, ps = ∆s (i) × βs
421 Then, the penalty caused by the conflict in reception (noted pe) should be computed as follows:
423 - if ∆e (i) = 1 then pe = 1
424 - else, pe = Φ (e) × βe × Ω (s, e)
426 Finally, the penalty associated with the communication is:
427 p = max (ps ∈ s, pe)
429 .. _cfg=network/crosstraffic:
431 Simulating Cross-Traffic
432 ^^^^^^^^^^^^^^^^^^^^^^^^
434 Since SimGrid v3.7, cross-traffic effects can be taken into account in
435 analytical simulations. It means that ongoing and incoming
436 communication flows are treated independently. In addition, the LV08
437 model adds 0.05 of usage on the opposite direction for each new
438 created flow. This can be useful to simulate some important TCP
439 phenomena such as ack compression.
441 For that to work, your platform must have two links for each
442 pair of interconnected hosts. An example of usable platform is
443 available in ``examples/platforms/crosstraffic.xml``.
445 This is activated through the ``network/crosstraffic`` item, that
446 can be set to 0 (disable this feature) or 1 (enable it).
448 Note that with the default host model this option is activated by default.
450 .. _cfg=network/loopback:
452 Configuring loopback link
453 ^^^^^^^^^^^^^^^^^^^^^^^^^
455 Several network model provide an implicit loopback link to account for local
456 communication on a host. By default it has a 10GBps bandwidth and a null latency.
457 This can be changed with ``network/loopback-lat`` and ``network/loopback-bw``
460 .. _cfg=smpi/async-small-thresh:
462 Simulating Asynchronous Send
463 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
465 (this configuration item is experimental and may change or disappear)
467 It is possible to specify that messages below a certain size (in bytes) will be
468 sent as soon as the call to MPI_Send is issued, without waiting for
469 the correspondent receive. This threshold can be configured through
470 the ``smpi/async-small-thresh`` item. The default value is 0. This
471 behavior can also be manually set for mailboxes, by setting the
472 receiving mode of the mailbox with a call to
473 :cpp:func:`MSG_mailbox_set_async`. After this, all messages sent to
474 this mailbox will have this behavior regardless of the message size.
476 This value needs to be smaller than or equals to the threshold set at
477 :ref:`cfg=smpi/send-is-detached-thresh`, because asynchronous messages
478 are meant to be detached as well.
485 **Option** ``ns3/TcpModel`` **Default:** "default" (ns-3 default)
487 When using ns-3, there is an extra item ``ns3/TcpModel``, corresponding
488 to the ``ns3::TcpL4Protocol::SocketType`` configuration item in
489 ns-3. The only valid values (enforced on the SimGrid side) are
490 'default' (no change to the ns-3 configuration), 'NewReno' or 'Reno' or
493 **Option** ``ns3/seed`` **Default:** "" (don't set the seed in ns-3)
495 This option is the random seed to provide to ns-3 with
496 ``ns3::RngSeedManager::SetSeed`` and ``ns3::RngSeedManager::SetRun``.
498 If left blank, no seed is set in ns-3. If the value 'time' is
499 provided, the current amount of seconds since epoch is used as a seed.
500 Otherwise, the provided value must be a number to use as a seed.
502 Configuring the Storage model
503 .............................
505 .. _cfg=storage/max_file_descriptors:
507 File Descriptor Count per Host
508 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
510 **Option** ``storage/max_file_descriptors`` **Default:** 1024
512 Each host maintains a fixed-size array of its file descriptors. You
513 can change its size through this item to either enlarge it if your
514 application requires it or to reduce it to save memory space.
521 SimGrid plugins allow one to extend the framework without changing its
522 source code directly. Read the source code of the existing plugins to
523 learn how to do so (in ``src/plugins``), and ask your questions to the
524 usual channels (Stack Overflow, Mailing list, IRC). The basic idea is
525 that plugins usually register callbacks to some signals of interest.
526 If they need to store some information about a given object (Link, CPU
527 or Actor), they do so through the use of a dedicated object extension.
529 Some of the existing plugins can be activated from the command line,
530 meaning that you can activate them from the command line without any
531 modification to your simulation code. For example, you can activate
532 the host energy plugin by adding ``--cfg=plugin:host_energy`` to your
535 Here is a partial list of plugins that can be activated this way. You can get
536 the full list by passing ``--cfg=plugin:help`` to your simulator.
538 - :ref:`Host Energy <plugin_host_energy>`: models the energy dissipation of the compute units.
539 - :ref:`Link Energy <plugin_link_energy>`: models the energy dissipation of the network.
540 - :ref:`Host Load <plugin_host_load>`: monitors the load of the compute units.
542 .. _options_modelchecking:
544 Configuring the Model-Checking
545 ------------------------------
547 To enable SimGrid's model-checking support, the program should
548 be executed using the simgrid-mc wrapper:
550 .. code-block:: shell
552 simgrid-mc ./my_program
554 Safety properties are expressed as assertions using the function
555 :cpp:func:`void MC_assert(int prop)`.
557 .. _cfg=smpi/buffering:
559 Specifying the MPI buffering behavior
560 .....................................
562 **Option** ``smpi/buffering`` **Default:** infty
564 Buffering in MPI has a huge impact on the communication semantic. For example,
565 standard blocking sends are synchronous calls when the system buffers are full
566 while these calls can complete immediately without even requiring a matching
567 receive call for small messages sent when the system buffers are empty.
569 In SMPI, this depends on the message size, that is compared against two thresholds:
571 - if (size < :ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>`) then
572 MPI_Send returns immediately, even if the corresponding receive has not be issued yet.
573 - if (:ref:`smpi/async-small-thresh <cfg=smpi/async-small-thresh>` < size < :ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>`) then
574 MPI_Send returns as soon as the corresponding receive has been issued. This is known as the eager mode.
575 - if (:ref:`smpi/send-is-detached-thresh <cfg=smpi/send-is-detached-thresh>` < size) then
576 MPI_Send returns only when the message has actually been sent over the network. This is known as the rendez-vous mode.
578 The ``smpi/buffering`` (only valid with MC) option gives an easier interface to choose between these semantics. It can take two values:
580 - **zero:** means that buffering should be disabled. All communications are actually blocking.
581 - **infty:** means that buffering should be made infinite. All communications are non-blocking.
583 .. _cfg=model-check/property:
585 Specifying a liveness property
586 ..............................
588 **Option** ``model-check/property`` **Default:** unset
590 If you want to specify liveness properties, you have to pass them on
591 the command line, specifying the name of the file containing the
592 property, as formatted by the `ltl2ba <https://github.com/utwente-fmt/ltl2ba>`_ program.
593 Note that ltl2ba is not part of SimGrid and must be installed separately.
595 .. code-block:: shell
597 simgrid-mc ./my_program --cfg=model-check/property:<filename>
599 .. _cfg=model-check/checkpoint:
601 Going for Stateful Verification
602 ...............................
604 By default, the system is backtracked to its initial state to explore
605 another path, instead of backtracking to the exact step before the fork
606 that we want to explore (this is called stateless verification). This
607 is done this way because saving intermediate states can rapidly
608 exhaust the available memory. If you want, you can change the value of
609 the ``model-check/checkpoint`` item. For example,
610 ``--cfg=model-check/checkpoint:1`` asks to take a checkpoint every
611 step. Beware, this will certainly explode your memory. Larger values
612 are probably better, make sure to experiment a bit to find the right
613 setting for your specific system.
615 .. _cfg=model-check/reduction:
617 Specifying the kind of reduction
618 ................................
620 The main issue when using the model-checking is the state space
621 explosion. You can activate some reduction technique with
622 ``--cfg=model-check/reduction:<technique>``. For now, this
623 configuration variable can take 2 values:
625 - **none:** Do not apply any kind of reduction (mandatory for
626 liveness properties, as our current DPOR algorithm breaks cycles)
627 - **dpor:** Apply Dynamic Partial Ordering Reduction. Only valid if
628 you verify local safety properties (default value for safety
631 Another way to mitigate the state space explosion is to search for
632 cycles in the exploration with the :ref:`cfg=model-check/visited`
633 configuration. Note that DPOR and state-equality reduction may not
634 play well together. You should choose between them.
636 Our current DPOR implementation could be improved in may ways. We are
637 currently improving its efficiency (both in term of reduction ability
638 and computational speed), and future work could make it compatible
639 with liveness properties.
641 .. _cfg=model-check/visited:
643 Size of Cycle Detection Set (state equality reduction)
644 ......................................................
646 Mc SimGrid can be asked to search for cycles during the exploration,
647 i.e. situations where a new explored state is in fact the same state
648 than a previous one.. This can prove useful to mitigate the state
649 space explosion with safety properties, and this is the crux when
650 searching for counter-examples to the liveness properties.
652 Note that this feature may break the current implementation of the
653 DPOR reduction technique.
655 The ``model-check/visited`` item is the maximum number of states, which
656 are stored in memory. If the maximum number of snapshotted state is
657 reached, some states will be removed from the memory and some cycles
658 might be missed. Small values can lead to incorrect verifications, but
659 large values can exhaust your memory and be CPU intensive as each new
660 state must be compared to that amount of older saved states.
662 The default settings depend on the kind of exploration. With safety
663 checking, no state is snapshotted and cycles cannot be detected. With
664 liveness checking, all states are snapshotted because missing a cycle
665 could hinder the exploration soundness.
667 .. _cfg=model-check/termination:
669 Non-Termination Detection
670 .........................
672 The ``model-check/termination`` configuration item can be used to
673 report if a non-termination execution path has been found. This is a
674 path with a cycle, which means that the program might never terminate.
676 This only works in safety mode, not in liveness mode.
678 This options is disabled by default.
680 .. _cfg=model-check/dot-output:
685 If set, the ``model-check/dot-output`` configuration item is the name
686 of a file in which to write a dot file of the path leading to the
687 property violation discovered (safety or liveness violation), as well
688 as the cycle for liveness properties. This dot file can then be fed to the
689 graphviz dot tool to generate a corresponding graphical representation.
691 .. _cfg=model-check/max-depth:
693 Exploration Depth Limit
694 .......................
696 The ``model-check/max-depth`` can set the maximum depth of the
697 exploration graph of the model checker. If this limit is reached, a
698 logging message is sent and the results might not be exact.
700 By default, the exploration is limited to the depth of 1000.
702 .. _cfg=model-check/timeout:
707 By default, the model checker does not handle timeout conditions: the `wait`
708 operations never time out. With the ``model-check/timeout`` configuration item
709 set to **yes**, the model checker will explore timeouts of `wait` operations.
711 .. _cfg=model-check/communications-determinism:
712 .. _cfg=model-check/send-determinism:
714 Communication Determinism
715 .........................
717 The ``model-check/communications-determinism`` and
718 ``model-check/send-determinism`` items can be used to select the
719 communication determinism mode of the model checker, which checks
720 determinism properties of the communications of an application.
724 Verification Performance Considerations
725 .......................................
727 The size of the stacks can have a huge impact on the memory
728 consumption when using model-checking. By default, each snapshot will
729 save a copy of the whole stacks and not only of the part that is
730 really meaningful: you should expect the contribution of the memory
731 consumption of the snapshots to be:
732 :math:`\text{number of processes} \times \text{stack size} \times \text{number of states}`.
734 When compiled against the model checker, the stacks are not
735 protected with guards: if the stack size is too small for your
736 application, the stack will silently overflow into other parts of the
737 memory (see :ref:`contexts/guard-size <cfg=contexts/guard-size>`).
739 .. _cfg=model-check/replay:
741 Replaying buggy execution paths from the model checker
742 ......................................................
744 Debugging the problems reported by the model checker is challenging:
745 First, the application under verification cannot be debugged with gdb
746 because the model checker already traces it. Then, the model checker may
747 explore several execution paths before encountering the issue, making it
748 very difficult to understand the output. Fortunately, SimGrid provides
749 the execution path leading to any reported issue so that you can replay
750 this path reported by the model checker, enabling the usage of classical
753 When the model checker finds an interesting path in the application
754 execution graph (where a safety or liveness property is violated), it
755 generates an identifier for this path. Here is an example of the output:
757 .. code-block:: shell
759 [ 0.000000] (0:@) Check a safety property
760 [ 0.000000] (0:@) **************************
761 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
762 [ 0.000000] (0:@) **************************
763 [ 0.000000] (0:@) Counter-example execution trace:
764 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
765 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
766 [ 0.000000] (0:@) Path = 1/3;1/4
767 [ 0.000000] (0:@) Expanded states = 27
768 [ 0.000000] (0:@) Visited states = 68
769 [ 0.000000] (0:@) Executed transitions = 46
771 The interesting line is ``Path = 1/3;1/4``, which means that you should use
772 ``--cfg=model-check/replay:1/3;1/4`` to replay your application on the buggy
773 execution path. All options (but the model checker related ones) must
774 remain the same. In particular, if you ran your application with
775 ``smpirun -wrapper simgrid-mc``, then do it again. Remove all
776 MC-related options, keep non-MC-related ones and add
777 ``--cfg=model-check/replay:???``.
779 Currently, if the path is of the form ``X;Y;Z``, each number denotes
780 the actor's pid that is selected at each indecision point. If it's of
781 the form ``X/a;Y/b``, the X and Y are the selected pids while the a
782 and b are the return values of their simcalls. In the previous
783 example, ``1/3;1/4``, you can see from the full output that the actor
784 1 is doing MC_RANDOM simcalls, so the 3 and 4 simply denote the values
785 that these simcall return on the execution branch leading to the
788 Configuring the User Code Virtualization
789 ----------------------------------------
791 .. _cfg=contexts/factory:
793 Selecting the Virtualization Factory
794 ....................................
796 **Option** contexts/factory **Default:** "raw"
798 In SimGrid, the user code is virtualized in a specific mechanism that
799 allows the simulation kernel to control its execution: when a user
800 process requires a blocking action (such as sending a message), it is
801 interrupted, and only gets released when the simulated clock reaches
802 the point where the blocking operation is done. This is explained
803 graphically in the `relevant tutorial, available online
804 <https://simgrid.org/tutorials/simgrid-simix-101.pdf>`_.
806 In SimGrid, the containers in which user processes are virtualized are
807 called contexts. Several context factory are provided, and you can
808 select the one you want to use with the ``contexts/factory``
809 configuration item. Some of the following may not exist on your
810 machine because of portability issues. In any case, the default one
811 should be the most effcient one (please report bugs if the
812 auto-detection fails for you). They are approximately sorted here from
813 the slowest to the most efficient:
815 - **thread:** very slow factory using full featured threads (either
816 pthreads or windows native threads). They are slow but very
817 standard. Some debuggers or profilers only work with this factory.
818 - **java:** Java applications are virtualized onto java threads (that
819 are regular pthreads registered to the JVM)
820 - **ucontext:** fast factory using System V contexts (Linux and FreeBSD only)
821 - **boost:** This uses the `context
822 implementation <http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html>`_
823 of the boost library for a performance that is comparable to our
825 |br| Install the relevant library (e.g. with the
826 libboost-contexts-dev package on Debian/Ubuntu) and recompile
828 - **raw:** amazingly fast factory using a context switching mechanism
829 of our own, directly implemented in assembly (only available for x86
830 and amd64 platforms for now) and without any unneeded system call.
832 The main reason to change this setting is when the debugging tools become
833 fooled by the optimized context factories. Threads are the most
834 debugging-friendly contexts, as they allow one to set breakpoints
835 anywhere with gdb and visualize backtraces for all processes, in order
836 to debug concurrency issues. Valgrind is also more comfortable with
837 threads, but it should be usable with all factories (Exception: the
838 callgrind tool really dislikes raw and ucontext factories).
840 .. _cfg=contexts/stack-size:
842 Adapting the Stack Size
843 .......................
845 **Option** ``contexts/stack-size`` **Default:** 8192 KiB
847 Each virtualized used process is executed using a specific system
848 stack. The size of this stack has a huge impact on the simulation
849 scalability, but its default value is rather large. This is because
850 the error messages that you get when the stack size is too small are
851 rather disturbing: this leads to stack overflow (overwriting other
852 stacks), leading to segfaults with corrupted stack traces.
854 If you want to push the scalability limits of your code, you might
855 want to reduce the ``contexts/stack-size`` item. Its default value is
856 8192 (in KiB), while our Chord simulation works with stacks as small
857 as 16 KiB, for example. You can ensure that some actors have a specific
858 size by simply changing the value of this configuration item before
859 creating these actors. The :cpp:func:`simgrid::s4u::Engine::set_config`
860 functions are handy for that.
862 This *setting is ignored* when using the thread factory (because there
863 is no way to modify the stack size with C++ system threads). Instead,
864 you should compile SimGrid and your application with
865 ``-fsplit-stack``. Note that this compilation flag is not compatible
866 with the model checker right now.
868 The operating system should only allocate memory for the pages of the
869 stack which are actually used and you might not need to use this in
870 most cases. However, this setting is very important when using the
871 model checker (see :ref:`options_mc_perf`).
873 .. _cfg=contexts/guard-size:
875 Disabling Stack Guard Pages
876 ...........................
878 **Option** ``contexts/guard-size`` **Default** 1 page in most case (0 pages on Windows or with MC)
880 Unless you use the threads context factory (see
881 :ref:`cfg=contexts/factory`), a stack guard page is usually used
882 which prevents the stack of a given actor from overflowing on another
883 stack. But the performance impact may become prohibitive when the
884 amount of actors increases. The option ``contexts/guard-size`` is the
885 number of stack guard pages used. By setting it to 0, no guard pages
886 will be used: in this case, you should avoid using small stacks (with
887 :ref:`contexts/stack-size <cfg=contexts/stack-size>`) as the stack
888 will silently overflow on other parts of the memory.
890 When no stack guard page is created, stacks may then silently overflow
891 on other parts of the memory if their size is too small for the
894 .. _cfg=contexts/nthreads:
895 .. _cfg=contexts/synchro:
897 Running User Code in Parallel
898 .............................
900 Parallel execution of the user code is only considered stable in
901 SimGrid v3.7 and higher, and mostly for MSG simulations. SMPI
902 simulations may well fail in parallel mode. It is described in
903 `INRIA RR-7653 <http://hal.inria.fr/inria-00602216/>`_.
905 If you are using the **ucontext** or **raw** context factories, you can
906 request to execute the user code in parallel. Several threads are
907 launched, each of them handling the same number of user contexts at each
908 run. To activate this, set the ``contexts/nthreads`` item to the amount
909 of cores that you have in your computer (or lower than 1 to have the
910 amount of cores auto-detected).
912 When parallel execution is activated, you can choose the
913 synchronization schema used with the ``contexts/synchro`` item,
914 which value is either:
916 - **futex:** ultra optimized synchronisation schema, based on futexes
917 (fast user-mode mutexes), and thus only available on Linux systems.
918 This is the default mode when available.
919 - **posix:** slow but portable synchronisation using only POSIX
921 - **busy_wait:** not really a synchronisation: the worker threads
922 constantly request new contexts to execute. It should be the most
923 efficient synchronisation schema, but it loads all the cores of
924 your machine for no good reason. You probably prefer the other less
927 Configuring the Tracing
928 -----------------------
930 The :ref:`tracing subsystem <outcomes_vizu>` can be configured in
931 several different ways depending on the used interface (S4U, SMPI, SimDag)
932 and the kind of traces that needs to be obtained. See the
933 :ref:`Tracing Configuration Options subsection
934 <tracing_tracing_options>` for a full description of each
935 configuration option.
937 We detail here a simple way to get the traces working for you, even if
938 you never used the tracing API.
941 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
943 .. code-block:: shell
945 --cfg=tracing:yes --cfg=tracing/uncategorized:yes
947 The first parameter activates the tracing subsystem, and the second
948 tells it to trace host and link utilization (without any
951 - MSG or SimDag-based simulator and categorized traces (you need to
952 declare categories and classify your tasks according to them)
954 .. code-block:: shell
956 --cfg=tracing:yes --cfg=tracing/categorized:yes
958 The first parameter activates the tracing subsystem, and the second
959 tells it to trace host and link categorized utilization.
961 - SMPI simulator and traces for a space/time view:
963 .. code-block:: shell
967 The `-trace` parameter for the smpirun script runs the simulation
968 with ``--cfg=tracing:yes --cfg=tracing/smpi:yes``. Check the
969 smpirun's `-help` parameter for additional tracing options.
971 Sometimes you might want to put additional information on the trace to
972 correctly identify them later, or to provide data that can be used to
973 reproduce an experiment. You have two ways to do that:
975 - Add a string on top of the trace file as comment:
977 .. code-block:: shell
979 --cfg=tracing/comment:my_simulation_identifier
981 - Add the contents of a textual file on top of the trace file as comment:
983 .. code-block:: shell
985 --cfg=tracing/comment-file:my_file_with_additional_information.txt
987 Please, use these two parameters (for comments) to make reproducible
988 simulations. For additional details about this and all tracing
989 options, check See the :ref:`tracing_tracing_options`.
994 .. _cfg=msg/debug-multiple-use:
999 **Option** ``msg/debug-multiple-use`` **Default:** off
1001 Sometimes your application may try to send a task that is still being
1002 executed somewhere else, making it impossible to send this task. However,
1003 for debugging purposes, one may want to know what the other host is/was
1004 doing. This option shows a backtrace of the other process.
1009 The SMPI interface provides several specific configuration items.
1010 These are not easy to see, since the code is usually launched through the
1011 ``smiprun`` script directly.
1013 .. _cfg=smpi/host-speed:
1014 .. _cfg=smpi/cpu-threshold:
1015 .. _cfg=smpi/simulate-computation:
1017 Automatic Benchmarking of SMPI Code
1018 ...................................
1020 In SMPI, the sequential code is automatically benchmarked, and these
1021 computations are automatically reported to the simulator. That is to
1022 say that if you have a large computation between a ``MPI_Recv()`` and
1023 a ``MPI_Send()``, SMPI will automatically benchmark the duration of
1024 this code, and create an execution task within the simulator to take
1025 this into account. For that, the actual duration is measured on the
1026 host machine and then scaled to the power of the corresponding
1027 simulated machine. The variable ``smpi/host-speed`` allows one to
1028 specify the computational speed of the host machine (in flop/s by
1029 default) to use when scaling the execution times.
1031 The default value is ``smpi/host-speed=20kf`` (= 20,000 flop/s). This
1032 is probably underestimated for most machines, leading SimGrid to
1033 overestimate the amount of flops in the execution blocks that are
1034 automatically injected in the simulator. As a result, the execution
1035 time of the whole application will probably be overestimated until you
1036 use a realistic value.
1038 When the code consists of numerous consecutive MPI calls, the
1039 previous mechanism feeds the simulation kernel with numerous tiny
1040 computations. The ``smpi/cpu-threshold`` item becomes handy when this
1041 impacts badly on the simulation performance. It specifies a threshold (in
1042 seconds) below which the execution chunks are not reported to the
1043 simulation kernel (default value: 1e-6).
1045 .. note:: The option ``smpi/cpu-threshold`` ignores any computation
1046 time spent below this threshold. SMPI does not consider the
1047 `amount of time` of these computations; there is no offset for
1048 this. Hence, a value that is too small, may lead to unreliable
1051 In some cases, however, one may wish to disable simulation of
1052 the computation of an application. This is the case when SMPI is used not to
1053 simulate an MPI application, but instead an MPI code that performs
1054 "live replay" of another MPI app (e.g., ScalaTrace's replay tool, or
1055 various on-line simulators that run an app at scale). In this case the
1056 computation of the replay/simulation logic should not be simulated by
1057 SMPI. Instead, the replay tool or on-line simulator will issue
1058 "computation events", which correspond to the actual MPI simulation
1059 being replayed/simulated. At the moment, these computation events can
1060 be simulated using SMPI by calling internal smpi_execute*() functions.
1062 To disable the benchmarking/simulation of a computation in the simulated
1063 application, the variable ``smpi/simulate-computation`` should be set
1064 to **no**. This option just ignores the timings in your simulation; it
1065 still executes the computations itself. If you want to stop SMPI from
1066 doing that, you should check the SMPI_SAMPLE macros, documented in
1067 Section :ref:`SMPI_use_faster`.
1069 +------------------------------------+-------------------------+-----------------------------+
1070 | Solution | Computations executed? | Computations simulated? |
1071 +====================================+=========================+=============================+
1072 | --cfg=smpi/simulate-computation:no | Yes | Never |
1073 +------------------------------------+-------------------------+-----------------------------+
1074 | --cfg=smpi/cpu-threshold:42 | Yes, in all cases | If it lasts over 42 seconds |
1075 +------------------------------------+-------------------------+-----------------------------+
1076 | SMPI_SAMPLE() macro | Only once per loop nest | Always |
1077 +------------------------------------+-------------------------+-----------------------------+
1079 .. _cfg=smpi/comp-adjustment-file:
1081 Slow-down or speed-up parts of your code
1082 ........................................
1084 **Option** ``smpi/comp-adjustment-file:`` **Default:** unset
1086 This option allows you to pass a file that contains two columns: The
1087 first column defines the section that will be subject to a speedup;
1088 the second column is the speedup. For instance:
1090 .. code-block:: shell
1092 "start:stop","ratio"
1093 "exchange_1.f:30:exchange_1.f:130",1.18244559422142
1095 The first line is the header - you must include it. The following
1096 line means that the code between two consecutive MPI calls on line 30
1097 in exchange_1.f and line 130 in exchange_1.f should receive a speedup
1098 of 1.18244559422142. The value for the second column is therefore a
1099 speedup, if it is larger than 1 and a slowdown if it is smaller
1100 than 1. Nothing will be changed if it is equal to 1.
1102 Of course, you can set any arbitrary filenames you want (so the start
1103 and end don't have to be in the same file), but be aware that this
1104 mechanism only supports `consecutive calls!`
1106 Please note that you must pass the ``-trace-call-location`` flag to
1107 smpicc or smpiff, respectively. This flag activates some internal
1108 macro definitions that help with obtaining the call location.
1110 .. _cfg=smpi/bw-factor:
1115 **Option** ``smpi/bw-factor``
1116 |br| **Default:** 65472:0.940694;15424:0.697866;9376:0.58729;5776:1.08739;3484:0.77493;1426:0.608902;732:0.341987;257:0.338112;0:0.812084
1118 The possible throughput of network links is often dependent on the
1119 message sizes, as protocols may adapt to different message sizes. With
1120 this option, a series of message sizes and factors are given, helping
1121 the simulation to be more realistic. For instance, the current default
1122 value means that messages with size 65472 bytes and more will get a total of
1123 MAX_BANDWIDTH*0.940694, messages of size 15424 to 65471 will get
1124 MAX_BANDWIDTH*0.697866, and so on (where MAX_BANDWIDTH denotes the
1125 bandwidth of the link).
1127 An experimental script to compute these factors is available online. See
1128 https://framagit.org/simgrid/platform-calibration/
1129 https://simgrid.org/contrib/smpi-saturation-doc.html
1131 .. _cfg=smpi/display-timing:
1133 Reporting Simulation Time
1134 .........................
1136 **Option** ``smpi/display-timing`` **Default:** 0 (false)
1138 Most of the time, you run MPI code with SMPI to compute the time it
1139 would take to run it on a platform. But since the code is run through
1140 the ``smpirun`` script, you don't have any control on the launcher
1141 code, making it difficult to report the simulated time when the
1142 simulation ends. If you enable the ``smpi/display-timing`` item,
1143 ``smpirun`` will display this information when the simulation
1145 SMPI will also display information about the amout of real time spent
1146 in application code and in SMPI internals, to provide hints about the
1147 need to use sampling to reduce simulation time.
1149 .. _cfg=smpi/display-allocs:
1151 Reporting memory allocations
1152 ............................
1154 **Option** ``smpi/display-allocs`` **Default:** 0 (false)
1156 SMPI intercepts malloc and calloc calls performed inside the running
1157 application, if it wasn't compiled with SMPI_NO_OVERRIDE_MALLOC.
1158 With this option, SMPI will show at the end of execution the amount of
1159 memory allocated through these calls, and locate the most expensive one.
1160 This helps finding the targets for manual memory sharing, or the threshold
1161 to use for smpi/auto-shared-malloc-thresh option (see :ref:`cfg=smpi/auto-shared-malloc-thresh`).
1163 .. _cfg=smpi/keep-temps:
1165 Keeping temporary files after simulation
1166 ........................................
1168 **Option** ``smpi/keep-temps`` **default:** 0 (false)
1170 SMPI usually generates a lot of temporary files that are cleaned after
1171 use. This option requests to preserve them, for example to debug or
1172 profile your code. Indeed, the binary files are removed very early
1173 under the dlopen privatization schema, which tends to fool the
1176 .. _cfg=smpi/lat-factor:
1181 **Option** ``smpi/lat-factor`` |br|
1182 **default:** 65472:11.6436;15424:3.48845;9376:2.59299;5776:2.18796;3484:1.88101;1426:1.61075;732:1.9503;257:1.95341;0:2.01467
1184 The motivation and syntax for this option is identical to the motivation/syntax
1185 of :ref:`cfg=smpi/bw-factor`.
1187 There is an important difference, though: While smpi/bw-factor `reduces` the
1188 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
1189 increase the latency, i.e., values larger than or equal to 1 are valid here.
1191 .. _cfg=smpi/papi-events:
1193 Trace hardware counters with PAPI
1194 .................................
1196 **Option** ``smpi/papi-events`` **default:** unset
1198 When the PAPI support is compiled into SimGrid, this option takes the
1199 names of PAPI counters and adds their respective values to the trace
1200 files (See Section :ref:`tracing_tracing_options`).
1204 This feature currently requires superuser privileges, as registers
1205 are queried. Only use this feature with code you trust! Call
1206 smpirun for instance via ``smpirun -wrapper "sudo "
1207 <your-parameters>`` or run ``sudo sh -c "echo 0 >
1208 /proc/sys/kernel/perf_event_paranoid"`` In the later case, sudo
1209 will not be required.
1211 It is planned to make this feature available on a per-process (or per-thread?) basis.
1212 The first draft, however, just implements a "global" (i.e., for all processes) set
1213 of counters, the "default" set.
1215 .. code-block:: shell
1217 --cfg=smpi/papi-events:"default:PAPI_L3_LDM:PAPI_L2_LDM"
1219 .. _cfg=smpi/privatization:
1221 Automatic Privatization of Global Variables
1222 ...........................................
1224 **Option** ``smpi/privatization`` **default:** "dlopen" (when using smpirun)
1226 MPI executables are usually meant to be executed in separate
1227 processes, but SMPI is executed in only one process. Global variables
1228 from executables will be placed in the same memory region and shared
1229 between processes, causing intricate bugs. Several options are
1230 possible to avoid this, as described in the main `SMPI publication
1231 <https://hal.inria.fr/hal-01415484>`_ and in the :ref:`SMPI
1232 documentation <SMPI_what_globals>`. SimGrid provides two ways of
1233 automatically privatizing the globals, and this option allows one to
1234 choose between them.
1236 - **no** (default when not using smpirun): Do not automatically
1237 privatize variables. Pass ``-no-privatize`` to smpirun to disable
1239 - **dlopen** or **yes** (default when using smpirun): Link multiple
1240 times against the binary.
1241 - **mmap** (slower, but maybe somewhat more stable):
1242 Runtime automatic switching of the data segments.
1245 This configuration option cannot be set in your platform file. You can only
1246 pass it as an argument to smpirun.
1248 .. _cfg=smpi/privatize-libs:
1250 Automatic privatization of global variables inside external libraries
1251 .....................................................................
1253 **Option** ``smpi/privatize-libs`` **default:** unset
1255 **Linux/BSD only:** When using dlopen (default) privatization,
1256 privatize specific shared libraries with internal global variables, if
1257 they can't be linked statically. For example libgfortran is usually
1258 used for Fortran I/O and indexes in files can be mixed up.
1260 Multiple libraries can be given, semicolon separated.
1262 This configuration option can only use either full paths to libraries,
1263 or full names. Check with ldd the name of the library you want to
1266 .. code-block:: shell
1270 libgfortran.so.3 => /usr/lib/x86_64-linux-gnu/libgfortran.so.3 (0x00007fbb4d91b000)
1273 Then you can use ``--cfg=smpi/privatize-libs:libgfortran.so.3``
1274 or ``--cfg=smpi/privatize-libs:/usr/lib/x86_64-linux-gnu/libgfortran.so.3``,
1275 but not ``libgfortran`` nor ``libgfortran.so``.
1277 .. _cfg=smpi/send-is-detached-thresh:
1279 Simulating MPI detached send
1280 ............................
1282 **Option** ``smpi/send-is-detached-thresh`` **default:** 65536
1284 This threshold specifies the size in bytes under which the send will
1285 return immediately. This is different from the threshold detailed in
1286 :ref:`cfg=smpi/async-small-thresh` because the message is not
1287 really sent when the send is posted. SMPI still waits for the
1288 corresponding receive to be posted, in order to perform the communication
1291 .. _cfg=smpi/coll-selector:
1293 Simulating MPI collective algorithms
1294 ....................................
1296 **Option** ``smpi/coll-selector`` **Possible values:** naive (default), ompi, mpich
1298 SMPI implements more than 100 different algorithms for MPI collective
1299 communication, to accurately simulate the behavior of most of the
1300 existing MPI libraries. The ``smpi/coll-selector`` item can be used to
1301 select the decision logic either of the OpenMPI or the MPICH libraries. (By
1302 default SMPI uses naive version of collective operations.)
1304 Each collective operation can be manually selected with a
1305 ``smpi/collective_name:algo_name``. Available algorithms are listed in
1306 :ref:`SMPI_use_colls`.
1308 .. TODO:: All available collective algorithms will be made available
1309 via the ``smpirun --help-coll`` command.
1311 .. _cfg=smpi/iprobe:
1313 Inject constant times for MPI_Iprobe
1314 ....................................
1316 **Option** ``smpi/iprobe`` **default:** 0.0001
1318 The behavior and motivation for this configuration option is identical
1319 with :ref:`smpi/test <cfg=smpi/test>`, but for the function
1322 .. _cfg=smpi/iprobe-cpu-usage:
1324 Reduce speed for iprobe calls
1325 .............................
1327 **Option** ``smpi/iprobe-cpu-usage`` **default:** 1 (no change)
1329 MPI_Iprobe calls can be heavily used in applications. To account
1330 correctly for the energy that cores spend probing, it is necessary to
1331 reduce the load that these calls cause inside SimGrid.
1333 For instance, we measured a maximum power consumption of 220 W for a
1334 particular application but only 180 W while this application was
1335 probing. Hence, the correct factor that should be passed to this
1336 option would be 180/220 = 0.81.
1340 Inject constant times for MPI_Init
1341 ..................................
1343 **Option** ``smpi/init`` **default:** 0
1345 The behavior and motivation for this configuration option is identical
1346 with :ref:`smpi/test <cfg=smpi/test>`, but for the function ``MPI_Init()``.
1350 Inject constant times for MPI_Isend()
1351 .....................................
1353 **Option** ``smpi/ois``
1355 The behavior and motivation for this configuration option is identical
1356 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Isend()``.
1360 Inject constant times for MPI_send()
1361 ....................................
1363 **Option** ``smpi/os``
1365 In several network models such as LogP, send (MPI_Send, MPI_Isend) and
1366 receive (MPI_Recv) operations incur costs (i.e., they consume CPU
1367 time). SMPI can factor these costs in as well, but the user has to
1368 configure SMPI accordingly as these values may vary by machine. This
1369 can be done by using ``smpi/os`` for MPI_Send operations; for MPI_Isend
1370 and MPI_Recv, use ``smpi/ois`` and ``smpi/or``, respectively. These work
1371 exactly as ``smpi/ois``.
1373 This item can consist of multiple sections; each section takes three
1374 values, for example ``1:3:2;10:5:1``. The sections are divided by ";"
1375 so this example contains two sections. Furthermore, each section
1376 consists of three values.
1378 1. The first value denotes the minimum size in bytes for this section to take effect;
1379 read it as "if message size is greater than this value (and other section has a larger
1380 first value that is also smaller than the message size), use this".
1381 In the first section above, this value is "1".
1383 2. The second value is the startup time; this is a constant value that will always
1384 be charged, no matter what the size of the message. In the first section above,
1387 3. The third value is the `per-byte` cost. That is, it is charged for every
1388 byte of the message (incurring cost messageSize*cost_per_byte)
1389 and hence accounts also for larger messages. In the first
1390 section of the example above, this value is "2".
1392 Now, SMPI always checks which section it should use for a given
1393 message; that is, if a message of size 11 is sent with the
1394 configuration of the example above, only the second section will be
1395 used, not the first, as the first value of the second section is
1396 closer to the message size. Hence, when ``smpi/os=1:3:2;10:5:1``, a
1397 message of size 11 incurs the following cost inside MPI_Send:
1398 ``5+11*1`` because 5 is the startup cost and 1 is the cost per byte.
1400 Note that the order of sections can be arbitrary; they will be ordered internally.
1404 Inject constant times for MPI_Recv()
1405 ....................................
1407 **Option** ``smpi/or``
1409 The behavior and motivation for this configuration option is identical
1410 with :ref:`smpi/os <cfg=smpi/os>`, but for the function ``MPI_Recv()``.
1413 .. _cfg=smpi/grow-injected-times:
1415 Inject constant times for MPI_Test
1416 ..................................
1418 **Option** ``smpi/test`` **default:** 0.0001
1420 By setting this option, you can control the amount of time a process
1421 sleeps when MPI_Test() is called; this is important, because SimGrid
1422 normally only advances the time while communication is happening and
1423 thus, MPI_Test will not add to the time, resulting in deadlock if it is
1424 used as a break-condition as in the following example:
1429 MPI_Test(request, flag, status);
1433 To speed up execution, we use a counter to keep track of how often we
1434 checked if the handle is now valid or not. Hence, we actually
1435 use counter*SLEEP_TIME, that is, the time MPI_Test() causes the
1436 process to sleep increases linearly with the number of previously
1437 failed tests. This behavior can be disabled by setting
1438 ``smpi/grow-injected-times`` to **no**. This will also disable this
1439 behavior for MPI_Iprobe.
1441 .. _cfg=smpi/shared-malloc:
1442 .. _cfg=smpi/shared-malloc-hugepage:
1447 **Option** ``smpi/shared-malloc`` **Possible values:** global (default), local
1449 If your simulation consumes too much memory, you may want to modify
1450 your code so that the working areas are shared by all MPI ranks. For
1451 example, in a block-cyclic matrix multiplication, you will only
1452 allocate one set of blocks, and all processes will share them.
1453 Naturally, this will lead to very wrong results, but this will save a
1454 lot of memory. So this is still desirable for some studies. For more on
1455 the motivation for that feature, please refer to the `relevant section
1456 <https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication>`_
1457 of the SMPI CourseWare (see Activity #2.2 of the pointed
1458 assignment). In practice, change the calls for malloc() and free() into
1459 SMPI_SHARED_MALLOC() and SMPI_SHARED_FREE().
1461 SMPI provides two algorithms for this feature. The first one, called
1462 ``local``, allocates one block per call to SMPI_SHARED_MALLOC()
1463 (each call site gets its own block) ,and this block is shared
1464 among all MPI ranks. This is implemented with the shm_* functions
1465 to create a new POSIX shared memory object (kept in RAM, in /dev/shm)
1466 for each shared block.
1468 With the ``global`` algorithm, each call to SMPI_SHARED_MALLOC()
1469 returns a new address, but it only points to a shadow block: its memory
1470 area is mapped on a 1 MiB file on disk. If the returned block is of size
1471 N MiB, then the same file is mapped N times to cover the whole block.
1472 At the end, no matter how many times you call SMPI_SHARED_MALLOC, this will
1473 only consume 1 MiB in memory.
1475 You can disable this behavior and come back to regular mallocs (for
1476 example for debugging purposes) using ``no`` as a value.
1478 If you want to keep private some parts of the buffer, for instance if these
1479 parts are used by the application logic and should not be corrupted, you
1480 can use SMPI_PARTIAL_SHARED_MALLOC(size, offsets, offsets_count). For example:
1484 mem = SMPI_PARTIAL_SHARED_MALLOC(500, {27,42 , 100,200}, 2);
1486 This will allocate 500 bytes to mem, such that mem[27..41] and
1487 mem[100..199] are shared while other area remain private.
1489 Then, it can be deallocated by calling SMPI_SHARED_FREE(mem).
1491 When smpi/shared-malloc:global is used, the memory consumption problem
1492 is solved, but it may induce too much load on the kernel's pages table.
1493 In this case, you should use huge pages so that the kernel creates only one
1494 entry per MB of malloced data instead of one entry per 4 kB.
1495 To activate this, you must mount a hugetlbfs on your system and allocate
1496 at least one huge page:
1498 .. code-block:: shell
1501 sudo mount none /home/huge -t hugetlbfs -o rw,mode=0777
1502 sudo sh -c 'echo 1 > /proc/sys/vm/nr_hugepages' # echo more if you need more
1504 Then, you can pass the option
1505 ``--cfg=smpi/shared-malloc-hugepage:/home/huge`` to smpirun to
1506 actually activate the huge page support in shared mallocs.
1508 .. _cfg=smpi/auto-shared-malloc-thresh:
1510 Automatically share allocations
1511 ...............................
1513 **Option** ``smpi/auto-shared-malloc-thresh:`` **Default:** 0 (false)
1514 This value in bytes represents the size above which all allocations
1515 will be "shared" by default (as if they were performed through
1516 SMPI_SHARED_MALLOC macros). Default = 0 = disabled feature.
1517 The value must be carefully chosen to only select data buffers which
1518 will not modify execution path or cause crash if their content is false.
1519 Option :ref:`cfg=smpi/display-allocs` can be used to locate the largest
1520 allocation detected in a run, and provide a good starting threshold.
1521 Note : malloc, calloc and free are overridden by smpicc/cxx by default.
1522 This can cause some troubles if codes are already overriding these. If this
1523 is the case, defining SMPI_NO_OVERRIDE_MALLOC in the compilation flags can
1524 help, but will make this feature unusable.
1528 Inject constant times for MPI_Wtime, gettimeofday and clock_gettime
1529 ...................................................................
1531 **Option** ``smpi/wtime`` **default:** 10 ns
1533 This option controls the amount of (simulated) time spent in calls to
1534 MPI_Wtime(), gettimeofday() and clock_gettime(). If you set this value
1535 to 0, the simulated clock is not advanced in these calls, which leads
1536 to issues if your application contains such a loop:
1540 while(MPI_Wtime() < some_time_bound) {
1541 /* some tests, with no communication nor computation */
1544 When the option smpi/wtime is set to 0, the time advances only on
1545 communications and computations. So the previous code results in an
1546 infinite loop: the current [simulated] time will never reach
1547 ``some_time_bound``. This infinite loop is avoided when that option
1548 is set to a small value, as it is by default since SimGrid v3.21.
1550 Note that if your application does not contain any loop depending on
1551 the current time only, then setting this option to a non-zero value
1552 will slow down your simulations by a tiny bit: the simulation loop has
1553 to be broken out of and reset each time your code asks for the current time.
1554 If the simulation speed really matters to you, you can avoid this
1555 extra delay by setting smpi/wtime to 0.
1557 .. _cfg=smpi/list-leaks:
1559 Report leaked MPI objects
1560 .........................
1562 **Option** ``smpi/list-leaks`` **default:** 0
1564 This option controls whether to report leaked MPI objects.
1565 The parameter is the number of leaks to report.
1567 Other Configurations
1568 --------------------
1570 .. _cfg=debug/clean-atexit:
1572 Cleanup at Termination
1573 ......................
1575 **Option** ``debug/clean-atexit`` **default:** on
1577 If your code is segfaulting during its finalization, it may help to
1578 disable this option to request that SimGrid not attempt any cleanups at
1579 the end of the simulation. Since the Unix process is ending anyway,
1580 the operating system will wipe it all.
1587 **Option** ``path`` **default:** . (current dir)
1589 It is possible to specify a list of directories to search in for the
1590 trace files (see :ref:`pf_trace`) by using this configuration
1591 item. To add several directory to the path, set the configuration
1592 item several times, as in ``--cfg=path:toto --cfg=path:tutu``
1594 .. _cfg=debug/breakpoint:
1599 **Option** ``debug/breakpoint`` **default:** unset
1601 This configuration option sets a breakpoint: when the simulated clock
1602 reaches the given time, a SIGTRAP is raised. This can be used to stop
1603 the execution and get a backtrace with a debugger.
1605 It is also possible to set the breakpoint from inside the debugger, by
1606 writing in global variable simgrid::simix::breakpoint. For example,
1609 .. code-block:: shell
1611 set variable simgrid::simix::breakpoint = 3.1416
1613 .. _cfg=debug/verbose-exit:
1618 **Option** ``debug/verbose-exit`` **default:** on
1620 By default, when Ctrl-C is pressed, the status of all existing actors
1621 is displayed before exiting the simulation. This is very useful to
1622 debug your code, but it can become troublesome if you have many
1623 actors. Set this configuration item to **off** to disable this
1626 .. _cfg=exception/cutpath:
1628 Truncate local path from exception backtrace
1629 ............................................
1631 **Option** ``exception/cutpath`` **default:** off
1633 This configuration option is used to remove the path from the
1634 backtrace shown when an exception is thrown. This is mainly useful for
1635 the tests: the full file path would makes the tests non-reproducible because
1636 the paths of source files depend of the build settings. That would
1637 break most of the tests since their output is continually compared.
1641 Logging configuration
1642 ---------------------
1644 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
1645 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
1646 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
1649 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
1650 messages from your code.
1652 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
1653 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
1654 practice, the following is equivalent to the above settings: ``--log=root.thresh:error --log=s4u_host.thresh:debug``.
1656 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
1657 your settings, as in ``--log="root.thresh:error s4u_host.thresh:debug"``. The parameters are interpreted in order, from left to right.
1660 Threshold configuration
1661 .......................
1663 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
1664 ``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
1665 see, ``threshold`` can be abbreviated here.
1667 Existing thresholds:
1669 - ``trace`` some functions display a message at this level when entering or returning
1670 - ``debug`` output that is mostly useful when debugging the corresponding module.
1671 - ``verbose`` verbose output that is only mildly interesting and can easily be ignored
1672 - ``info`` usual output (this is the default threshold of all categories)
1673 - ``warning`` minor issue encountered
1674 - ``error`` issue encountered
1675 - ``critical`` major issue encountered, such as assertions failures
1679 Format configuration
1680 ....................
1682 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
1683 as the date, or the actor ID, everything. Existing format directives:
1686 - %n: line separator (LOG4J compatible)
1687 - %e: plain old space (SimGrid extension)
1689 - %m: user-provided message
1691 - %c: Category name (LOG4J compatible)
1692 - %p: Priority name (LOG4J compatible)
1694 - %h: Hostname (SimGrid extension)
1695 - %a: Actor name (SimGrid extension -- note that with SMPI this is the integer value of the process rank)
1696 - %i: Actor PID (SimGrid extension -- this is a 'i' as in 'i'dea)
1697 - %t: Thread "name" (LOG4J compatible -- actually the address of the thread in memory)
1699 - %F: file name where the log event was raised (LOG4J compatible)
1700 - %l: location where the log event was raised (LOG4J compatible, like '%%F:%%L' -- this is a l as in 'l'etter)
1701 - %L: line number where the log event was raised (LOG4J compatible)
1702 - %M: function name (LOG4J compatible -- called method name here of course).
1704 - %d: date (UNIX-like epoch)
1705 - %r: application age (time elapsed since the beginning of the application)
1708 ``--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
1709 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
1710 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
1711 provided layout is used for every messages.
1713 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
1717 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
1718 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'"``.
1719 Another option is to use the ``%e`` directive for spaces, as in ``--log=root.fmt:%l:%e[%p/%c]:%e%m%n``.
1724 The keyword ``app`` controls the appended of a logging category. For example ``--log=root.app:file:mylogfile`` redirects every output to the file ``mylogfile``.
1726 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
1727 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.
1729 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``
1730 ensures that the log file ``mylog`` will never overpass 500 bytes in size.
1732 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
1733 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.
1738 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
1739 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
1740 ``on`` (or ``yes`` or ``1``), the produced messages will also be passed to the upper appender.
1742 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
1743 ``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
1744 will only be sent to ``all.log``.
1749 ``--help-logs`` displays a complete help message about logging in SimGrid.
1751 ``--help-log-categories`` displays the actual hierarchy of log categories for this binary.
1753 ``--log=no_loc`` hides the source locations (file names and line numbers) from the messages. This is useful to make tests reproducible.