1 /*! \page options Step 2: Configure SimGrid
3 A number of options can be given at runtime to change the default
4 SimGrid behavior. For a complete list of all configuration options
5 accepted by the SimGrid version used in your simulator, simply pass
6 the --help configuration flag to your program. If some of the options
7 are not documented on this page, this is a bug that you should please
8 report so that we can fix it. Note that some of the options presented
9 here may not be available in your simulators, depending on the
10 @ref install_src_config "compile-time options" that you used.
14 \section options_using Passing configuration options to the simulators
16 There is several way to pass configuration options to the simulators.
17 The most common way is to use the \c --cfg command line argument. For
18 example, to set the item \c Item to the value \c Value, simply
19 type the following: \verbatim
20 my_simulator --cfg=Item:Value (other arguments)
23 Several \c `--cfg` command line arguments can naturally be used. If you
24 need to include spaces in the argument, don't forget to quote the
25 argument. You can even escape the included quotes (write \' for ' if
26 you have your argument between ').
28 Another solution is to use the \c \<config\> tag in the platform file. The
29 only restriction is that this tag must occure before the first
30 platform element (be it \c \<AS\>, \c \<cluster\>, \c \<peer\> or whatever).
31 The \c \<config\> tag takes an \c id attribute, but it is currently
32 ignored so you don't really need to pass it. The important par is that
33 within that tag, you can pass one or several \c \<prop\> tags to specify
34 the configuration to use. For example, setting \c Item to \c Value
35 can be done by adding the following to the beginning of your platform
39 <prop id="Item" value="Value"/>
43 A last solution is to pass your configuration directly using the C
44 interface. If you happen to use the MSG interface, this is very easy
45 with the MSG_config() function. If you do not use MSG, that's a bit
46 more complex, as you have to mess with the internal configuration set
47 directly as follows. Check the \ref XBT_config "relevant page" for
48 details on all the functions you can use in this context, \c
49 _sg_cfg_set being the only configuration set currently used in
53 #include <xbt/config.h>
55 extern xbt_cfg_t _sg_cfg_set;
57 int main(int argc, char *argv[]) {
60 /* Prefer MSG_config() if you use MSG!! */
61 xbt_cfg_set_parse(_sg_cfg_set,"Item:Value");
67 \section options_model Configuring the platform models
69 \anchor options_storage_model
70 \anchor options_vm_model
71 \subsection options_model_select Selecting the platform models
73 SimGrid comes with several network, CPU and storage models built in, and you
74 can change the used model at runtime by changing the passed
75 configuration. The three main configuration items are given below.
76 For each of these items, passing the special \c help value gives
77 you a short description of all possible values. Also, \c --help-models
78 should provide information about all models for all existing resources.
79 - \b network/model: specify the used network model
80 - \b cpu/model: specify the used CPU model
81 - \b host/model: specify the used host model
82 - \b storage/model: specify the used storage model (there is currently only one such model - this option is hence only useful for future releases)
83 - \b vm/model: specify the model for virtual machines (there is currently only one such model - this option is hence only useful for future releases)
85 %As of writing, the following network models are accepted. Over
86 the time new models can be added, and some experimental models can be
87 removed; check the values on your simulators for an uptodate
88 information. Note that the CM02 model is described in the research report
89 <a href="ftp://ftp.ens-lyon.fr/pub/LIP/Rapports/RR/RR2002/RR2002-40.ps.gz">A
90 Network Model for Simulation of Grid Application</a> while LV08 is
92 <a href="http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf">Accuracy Study and Improvement of Network Simulation in the SimGrid Framework</a>.
94 - \b LV08 (default one): Realistic network analytic model
95 (slow-start modeled by multiplying latency by 10.4, bandwidth by
96 .92; bottleneck sharing uses a payload of S=8775 for evaluating RTT)
97 - \anchor options_model_select_network_constant \b Constant: Simplistic network model where all communication
98 take a constant time (one second). This model provides the lowest
99 realism, but is (marginally) faster.
100 - \b SMPI: Realistic network model specifically tailored for HPC
101 settings (accurate modeling of slow start with correction factors on
102 three intervals: < 1KiB, < 64 KiB, >= 64 KiB). See also \ref
103 options_model_network_coefs "this section" for more info.
104 - \b IB: Realistic network model specifically tailored for HPC
105 settings with InfiniBand networks (accurate modeling contention
106 behavior, based on the model explained in
107 http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf).
108 See also \ref options_model_network_coefs "this section" for more info.
109 - \b CM02: Legacy network analytic model (Very similar to LV08, but
110 without corrective factors. The timings of small messages are thus
112 - \b Reno: Model from Steven H. Low using lagrange_solve instead of
113 lmm_solve (experts only; check the code for more info).
114 - \b Reno2: Model from Steven H. Low using lagrange_solve instead of
115 lmm_solve (experts only; check the code for more info).
116 - \b Vegas: Model from Steven H. Low using lagrange_solve instead of
117 lmm_solve (experts only; check the code for more info).
119 If you compiled SimGrid accordingly, you can use packet-level network
120 simulators as network models (see \ref pls). In that case, you have
121 two extra models, described below, and some \ref options_pls "specific
122 additional configuration flags".
123 - \b NS3: Network pseudo-model using the NS3 tcp model instead of an
126 Concerning the CPU, we have only one model for now:
127 - \b Cas01: Simplistic CPU model (time=size/power)
129 The host concept is the aggregation of a CPU with a network
130 card. Three models exists, but actually, only 2 of them are
131 interesting. The "compound" one is simply due to the way our internal
132 code is organized, and can easily be ignored. So at the end, you have
133 two host models: The default one allows to aggregate an
134 existing CPU model with an existing network model, but does not allow
135 parallel tasks because these beasts need some collaboration between
136 the network and CPU model. That is why, ptask_07 is used by default
138 - \b default: Default host model. Currently, CPU:Cas01 and
139 network:LV08 (with cross traffic enabled)
140 - \b compound: Host model that is automatically chosen if
141 you change the network and CPU models
142 - \b ptask_L07: Host model somehow similar to Cas01+CM02 but
143 allowing parallel tasks
145 \subsection options_generic_plugin Plugins
147 SimGrid supports the use of plugins; currently, no known plugins
148 can be activated but there are use-cases where you may want to write
149 your own plugin (for instance, for logging).
151 Plugins can for instance define own classes that inherit from
152 existing classes (for instance, a class "CpuEnergy" inherits from
153 "Cpu" to assess energy consumption).
155 The plugin connects to the code by registering callbacks using
156 ``signal.connect(callback)`` (see file ``src/surf/plugins/energy.cpp`` for
164 This option is case-sensitive: Energy and energy are not the same!
166 \subsection options_model_optim Optimization level of the platform models
168 The network and CPU models that are based on lmm_solve (that
169 is, all our analytical models) accept specific optimization
171 - items \b network/optim and \b CPU/optim (both default to 'Lazy'):
172 - \b Lazy: Lazy action management (partial invalidation in lmm +
173 heap in action remaining).
174 - \b TI: Trace integration. Highly optimized mode when using
175 availability traces (only available for the Cas01 CPU model for
177 - \b Full: Full update of remaining and variables. Slow but may be
178 useful when debugging.
179 - items \b network/maxmin_selective_update and
180 \b cpu/maxmin_selective_update: configure whether the underlying
181 should be lazily updated or not. It should have no impact on the
182 computed timings, but should speed up the computation.
184 It is still possible to disable the \c maxmin_selective_update feature
185 because it can reveal counter-productive in very specific scenarios
186 where the interaction level is high. In particular, if all your
187 communication share a given backbone link, you should disable it:
188 without \c maxmin_selective_update, every communications are updated
189 at each step through a simple loop over them. With that feature
190 enabled, every communications will still get updated in this case
191 (because of the dependency induced by the backbone), but through a
192 complicated pattern aiming at following the actual dependencies.
194 \subsection options_model_precision Numerical precision of the platform models
196 The analytical models handle a lot of floating point values. It is
197 possible to change the epsilon used to update and compare them through
198 the \b maxmin/precision item (default value: 0.00001). Changing it
199 may speedup the simulation by discarding very small actions, at the
200 price of a reduced numerical precision.
202 \subsection options_model_nthreads Parallel threads for model updates
204 By default, Surf computes the analytical models sequentially to share their
205 resources and update their actions. It is possible to run them in parallel,
206 using the \b surf/nthreads item (default value: 1). If you use a
207 negative or null value, the amount of available cores is automatically
208 detected and used instead.
210 Depending on the workload of the models and their complexity, you may get a
211 speedup or a slowdown because of the synchronization costs of threads.
213 \subsection options_model_network Configuring the Network model
215 \subsubsection options_model_network_gamma Maximal TCP window size
217 The analytical models need to know the maximal TCP window size to take
218 the TCP congestion mechanism into account. This is set to 20000 by
219 default, but can be changed using the \b network/TCP_gamma item.
221 On linux, this value can be retrieved using the following
222 commands. Both give a set of values, and you should use the last one,
223 which is the maximal size.\verbatim
224 cat /proc/sys/net/ipv4/tcp_rmem # gives the sender window
225 cat /proc/sys/net/ipv4/tcp_wmem # gives the receiver window
228 \subsubsection options_model_network_coefs Correcting important network parameters
230 SimGrid can take network irregularities such as a slow startup or
231 changing behavior depending on the message size into account.
232 You should not change these values unless you really know what you're doing.
234 The corresponding values were computed through data fitting one the
235 timings of packet-level simulators.
238 <a href="http://mescal.imag.fr/membres/arnaud.legrand/articles/simutools09.pdf">Accuracy Study and Improvement of Network Simulation in the SimGrid Framework</a>
239 for more information about these parameters.
241 If you are using the SMPI model, these correction coefficients are
242 themselves corrected by constant values depending on the size of the
243 exchange. Again, only hardcore experts should bother about this fact.
245 InfiniBand network behavior can be modeled through 3 parameters, as explained in
246 <a href="http://mescal.imag.fr/membres/jean-marc.vincent/index.html/PhD/Vienne.pdf">this PhD thesis</a>.
247 These factors can be changed through the following option:
250 smpi/IB_penalty_factors:"βe;βs;γs"
253 By default SMPI uses factors computed on the Stampede Supercomputer at TACC, with optimal
254 deployment of processes on nodes.
256 \subsubsection options_model_network_crosstraffic Simulating cross-traffic
258 %As of SimGrid v3.7, cross-traffic effects can be taken into account in
259 analytical simulations. It means that ongoing and incoming
260 communication flows are treated independently. In addition, the LV08
261 model adds 0.05 of usage on the opposite direction for each new
262 created flow. This can be useful to simulate some important TCP
263 phenomena such as ack compression.
265 For that to work, your platform must have two links for each
266 pair of interconnected hosts. An example of usable platform is
267 available in <tt>examples/platforms/crosstraffic.xml</tt>.
269 This is activated through the \b network/crosstraffic item, that
270 can be set to 0 (disable this feature) or 1 (enable it).
272 Note that with the default host model this option is activated by default.
274 \subsubsection options_model_network_coord Coordinated-based network models
276 When you want to use network coordinates, as it happens when you use
277 an \<AS\> in your platform file with \c Vivaldi as a routing (see also
278 Section \ref pf_routing_model_vivaldi "Vivaldi Routing Model"), you must
279 set the \b network/coordinates to \c yes so that all mandatory
280 initialization are done in the simulator.
282 \subsubsection options_model_network_sendergap Simulating sender gap
284 (this configuration item is experimental and may change or disapear)
286 It is possible to specify a timing gap between consecutive emission on
287 the same network card through the \b network/sender_gap item. This
288 is still under investigation as of writting, and the default value is
289 to wait 10 microseconds (1e-5 seconds) between emissions.
291 \subsubsection options_model_network_asyncsend Simulating asyncronous send
293 (this configuration item is experimental and may change or disapear)
295 It is possible to specify that messages below a certain size will be sent
296 as soon as the call to MPI_Send is issued, without waiting for the
297 correspondant receive. This threshold can be configured through the
298 \b smpi/async_small_thresh item. The default value is 0. This behavior can also be
299 manually set for MSG mailboxes, by setting the receiving mode of the mailbox
300 with a call to \ref MSG_mailbox_set_async . For MSG, all messages sent to this
301 mailbox will have this behavior, so consider using two mailboxes if needed.
303 This value needs to be smaller than or equals to the threshold set at
304 \ref options_model_smpi_detached , because asynchronous messages are
305 meant to be detached as well.
307 \subsubsection options_pls Configuring packet-level pseudo-models
309 When using the packet-level pseudo-models, several specific
310 configuration flags are provided to configure the associated tools.
311 There is by far not enough such SimGrid flags to cover every aspects
312 of the associated tools, since we only added the items that we
313 needed ourselves. Feel free to request more items (or even better:
314 provide patches adding more items).
316 When using NS3, the only existing item is \b ns3/TcpModel,
317 corresponding to the ns3::TcpL4Protocol::SocketType configuration item
318 in NS3. The only valid values (enforced on the SimGrid side) are
319 'NewReno' or 'Reno' or 'Tahoe'.
321 \section options_modelchecking Configuring the Model-Checking
323 To enable the SimGrid model-checking support the program should
324 be executed using the simgrid-mc wrapper:
326 simgrid-mc ./my_program
329 Safety properties are expressed as assertions using the function
331 void MC_assert(int prop);
334 \subsection options_modelchecking_liveness Specifying a liveness property
336 If you want to specify liveness properties (beware, that's
337 experimental), you have to pass them on the command line, specifying
338 the name of the file containing the property, as formatted by the
342 --cfg=model-check/property:<filename>
345 \subsection options_modelchecking_steps Going for stateful verification
347 By default, the system is backtracked to its initial state to explore
348 another path instead of backtracking to the exact step before the fork
349 that we want to explore (this is called stateless verification). This
350 is done this way because saving intermediate states can rapidly
351 exhaust the available memory. If you want, you can change the value of
352 the <tt>model-check/checkpoint</tt> variable. For example, the
353 following configuration will ask to take a checkpoint every step.
354 Beware, this will certainly explode your memory. Larger values are
355 probably better, make sure to experiment a bit to find the right
356 setting for your specific system.
359 --cfg=model-check/checkpoint:1
362 \subsection options_modelchecking_reduction Specifying the kind of reduction
364 The main issue when using the model-checking is the state space
365 explosion. To counter that problem, several exploration reduction
366 techniques can be used. There is unfortunately no silver bullet here,
367 and the most efficient reduction techniques cannot be applied to any
368 properties. In particular, the DPOR method cannot be applied on
369 liveness properties since it may break some cycles in the exploration
370 that are important to the property validity.
373 --cfg=model-check/reduction:<technique>
376 For now, this configuration variable can take 2 values:
377 * none: Do not apply any kind of reduction (mandatory for now for
379 * dpor: Apply Dynamic Partial Ordering Reduction. Only valid if you
380 verify local safety properties.
382 \subsection options_modelchecking_visited model-check/visited, Cycle detection
384 In order to detect cycles, the model-checker needs to check if a new explored
385 state is in fact the same state than a previous one. In order to do this,
386 the model-checker can take a snapshot of each visited state: this snapshot is
387 then used to compare it with subsequent states in the exploration graph.
389 The \b model-check/visited is the maximum number of states which are stored in
390 memory. If the maximum number of snapshotted state is reached some states will
391 be removed from the memory and some cycles might be missed.
393 By default, no state is snapshotted and cycles cannot be detected.
395 \subsection options_modelchecking_termination model-check/termination, Non termination detection
397 The \b model-check/termination configuration item can be used to report if a
398 non-termination execution path has been found. This is a path with a cycle
399 which means that the program might never terminate.
401 This only works in safety mode.
403 This options is disabled by default.
405 \subsection options_modelchecking_dot_output model-check/dot_output, Dot output
407 If set, the \b model-check/dot_output configuration item is the name of a file
408 in which to write a dot file of the path leading the found property (safety or
409 liveness violation) as well as the cycle for liveness properties. This dot file
410 can then fed to the graphviz dot tool to generate an corresponding graphical
413 \subsection options_modelchecking_max_depth model-check/max_depth, Depth limit
415 The \b model-checker/max_depth can set the maximum depth of the exploration
416 graph of the model-checker. If this limit is reached, a logging message is
417 sent and the results might not be exact.
419 By default, there is not depth limit.
421 \subsection options_modelchecking_timeout Handling of timeout
423 By default, the model-checker does not handle timeout conditions: the `wait`
424 operations never time out. With the \b model-check/timeout configuration item
425 set to \b yes, the model-checker will explore timeouts of `wait` operations.
427 \subsection options_modelchecking_comm_determinism Communication determinism
429 The \b model-check/communications_determinism and
430 \b model-check/send_determinism items can be used to select the communication
431 determinism mode of the model-checker which checks determinism properties of
432 the communications of an application.
434 \subsection options_modelchecking_sparse_checkpoint Per page checkpoints
436 When the model-checker is configured to take a snapshot of each explored state
437 (with the \b model-checker/visited item), the memory consumption can rapidly
438 reach GiB ou Tib of memory. However, for many workloads, the memory does not
439 change much between different snapshots and taking a complete copy of each
440 snapshot is a waste of memory.
442 The \b model-check/sparse_checkpoint option item can be set to \b yes in order
443 to avoid making a complete copy of each snapshot: instead, each snapshot will be
444 decomposed in blocks which will be stored separately.
445 If multiple snapshots share the same block (or if the same block
446 is used in the same snapshot), the same copy of the block will be shared leading
447 to a reduction of the memory footprint.
449 For many applications, this option considerably reduces the memory consumption.
450 In somes cases, the model-checker might be slightly slower because of the time
451 taken to manage the metadata about the blocks. In other cases however, this
452 snapshotting strategy will be much faster by reducing the cache consumption.
453 When the memory consumption is important, by avoiding to hit the swap or
454 reducing the swap usage, this option might be much faster than the basic
455 snapshotting strategy.
457 This option is currently disabled by default.
459 \subsection options_mc_perf Performance considerations for the model checker
461 The size of the stacks can have a huge impact on the memory
462 consumption when using model-checking. By default, each snapshot will
463 save a copy of the whole stacks and not only of the part which is
464 really meaningful: you should expect the contribution of the memory
465 consumption of the snapshots to be \f$ \mbox{number of processes}
466 \times \mbox{stack size} \times \mbox{number of states} \f$.
468 The \b model-check/sparse_checkpoint can be used to reduce the memory
469 consumption by trying to share memory between the different snapshots.
471 When compiled against the model checker, the stacks are not
472 protected with guards: if the stack size is too small for your
473 application, the stack will silently overflow on other parts of the
476 \subsection options_modelchecking_hash Hashing of the state (experimental)
478 Usually most of the time of the model-checker is spent comparing states. This
479 process is complicated and consumes a lot of bandwidth and cache.
480 In order to speedup the state comparison, the experimental \b model-checker/hash
481 configuration item enables the computation of a hash summarizing as much
482 information of the state as possible into a single value. This hash can be used
483 to avoid most of the comparisons: the costly comparison is then only used when
484 the hashes are identical.
486 Currently most of the state is not included in the hash because the
487 implementation was found to be buggy and this options is not as useful as
488 it could be. For this reason, it is currently disabled by default.
490 \subsection options_modelchecking_recordreplay Record/replay (experimental)
492 As the model-checker keeps jumping at different places in the execution graph,
493 it is difficult to understand what happens when trying to debug an application
494 under the model-checker. Event the output of the program is difficult to
495 interpret. Moreover, the model-checker does not behave nicely with advanced
496 debugging tools such as valgrind. For those reason, to identify a trajectory
497 in the execution graph with the model-checker and replay this trajcetory and
498 without the model-checker black-magic but with more standard tools
499 (such as a debugger, valgrind, etc.). For this reason, Simgrid implements an
500 experimental record/replay functionnality in order to record a trajectory with
501 the model-checker and replay it without the model-checker.
503 When the model-checker finds an interesting path in the application execution
504 graph (where a safety or liveness property is violated), it can generate an
505 identifier for this path. In order to enable this behavious the
506 \b model-check/record must be set to \b yes. By default, this behaviour is not
509 This is an example of output:
512 [ 0.000000] (0:@) Check a safety property
513 [ 0.000000] (0:@) **************************
514 [ 0.000000] (0:@) *** PROPERTY NOT VALID ***
515 [ 0.000000] (0:@) **************************
516 [ 0.000000] (0:@) Counter-example execution trace:
517 [ 0.000000] (0:@) Path = 1/3;1/4
518 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(3)
519 [ 0.000000] (0:@) [(1)Tremblay (app)] MC_RANDOM(4)
520 [ 0.000000] (0:@) Expanded states = 27
521 [ 0.000000] (0:@) Visited states = 68
522 [ 0.000000] (0:@) Executed transitions = 46
525 This path can then be replayed outside of the model-checker (and even in
526 non-MC build of simgrid) by setting the \b model-check/replay item to the given
527 path. The other options should be the same (but the model-checker should
530 The format and meaning of the path may change between different releases so
531 the same release of Simgrid should be used for the record phase and the replay
534 \section options_virt Configuring the User Process Virtualization
536 \subsection options_virt_factory Selecting the virtualization factory
538 In SimGrid, the user code is virtualized in a specific mechanism
539 that allows the simulation kernel to control its execution: when a user
540 process requires a blocking action (such as sending a message), it is
541 interrupted, and only gets released when the simulated clock reaches
542 the point where the blocking operation is done.
544 In SimGrid, the containers in which user processes are virtualized are
545 called contexts. Several context factory are provided, and you can
546 select the one you want to use with the \b contexts/factory
547 configuration item. Some of the following may not exist on your
548 machine because of portability issues. In any case, the default one
549 should be the most effcient one (please report bugs if the
550 auto-detection fails for you). They are sorted here from the slowest
552 - \b thread: very slow factory using full featured threads (either
553 pthreads or windows native threads)
554 - \b ucontext: fast factory using System V contexts (or a portability
555 layer of our own on top of Windows fibers)
556 - \b raw: amazingly fast factory using a context switching mecanism
557 of our own, directly implemented in assembly (only available for x86
558 and amd64 platforms for now)
559 - \b boost: This uses the [context implementation](http://www.boost.org/doc/libs/1_59_0/libs/context/doc/html/index.html)
560 of the boost library; you must have this library installed before
561 you compile SimGrid. (On Debian GNU/Linux based systems, this is
562 provided by the libboost-contexts-dev package.)
564 The only reason to change this setting is when the debugging tools get
565 fooled by the optimized context factories. Threads are the most
566 debugging-friendly contextes, as they allow to set breakpoints anywhere with gdb
567 and visualize backtraces for all processes, in order to debug concurrency issues.
568 Valgrind is also more comfortable with threads, but it should be usable with all factories.
570 \subsection options_virt_stacksize Adapting the used stack size
572 Each virtualized used process is executed using a specific system
573 stack. The size of this stack has a huge impact on the simulation
574 scalability, but its default value is rather large. This is because
575 the error messages that you get when the stack size is too small are
576 rather disturbing: this leads to stack overflow (overwriting other
577 stacks), leading to segfaults with corrupted stack traces.
579 If you want to push the scalability limits of your code, you might
580 want to reduce the \b contexts/stack_size item. Its default value
581 is 8192 (in KiB), while our Chord simulation works with stacks as small
582 as 16 KiB, for example. For the thread factory, the default value
583 is the one of the system, if it is too large/small, it has to be set
586 The operating system should only allocate memory for the pages of the
587 stack which are actually used and you might not need to use this in
588 most cases. However, this setting is very important when using the
589 model checker (see \ref options_mc_perf).
591 In some cases, no stack guard page is used and the stack will silently
592 overflow on other parts of the memory if the stack size is too small
593 for your application. This happens :
595 - on Windows systems;
596 - when the model checker is enabled;
597 - when stack guard pages are explicitely disabled (see \ref options_perf_guard_size).
599 \subsection options_virt_parallel Running user code in parallel
601 Parallel execution of the user code is only considered stable in
602 SimGrid v3.7 and higher. It is described in
603 <a href="http://hal.inria.fr/inria-00602216/">INRIA RR-7653</a>.
605 If you are using the \c ucontext or \c raw context factories, you can
606 request to execute the user code in parallel. Several threads are
607 launched, each of them handling as much user contexts at each run. To
608 actiave this, set the \b contexts/nthreads item to the amount of
609 cores that you have in your computer (or lower than 1 to have
610 the amount of cores auto-detected).
612 Even if you asked several worker threads using the previous option,
613 you can request to start the parallel execution (and pay the
614 associated synchronization costs) only if the potential parallelism is
615 large enough. For that, set the \b contexts/parallel_threshold
616 item to the minimal amount of user contexts needed to start the
617 parallel execution. In any given simulation round, if that amount is
618 not reached, the contexts will be run sequentially directly by the
619 main thread (thus saving the synchronization costs). Note that this
620 option is mainly useful when the grain of the user code is very fine,
621 because our synchronization is now very efficient.
623 When parallel execution is activated, you can choose the
624 synchronization schema used with the \b contexts/synchro item,
625 which value is either:
626 - \b futex: ultra optimized synchronisation schema, based on futexes
627 (fast user-mode mutexes), and thus only available on Linux systems.
628 This is the default mode when available.
629 - \b posix: slow but portable synchronisation using only POSIX
631 - \b busy_wait: not really a synchronisation: the worker threads
632 constantly request new contexts to execute. It should be the most
633 efficient synchronisation schema, but it loads all the cores of your
634 machine for no good reason. You probably prefer the other less
637 \section options_tracing Configuring the tracing subsystem
639 The \ref tracing "tracing subsystem" can be configured in several
640 different ways depending on the nature of the simulator (MSG, SimDag,
641 SMPI) and the kind of traces that need to be obtained. See the \ref
642 tracing_tracing_options "Tracing Configuration Options subsection" to
643 get a detailed description of each configuration option.
645 We detail here a simple way to get the traces working for you, even if
646 you never used the tracing API.
649 - Any SimGrid-based simulator (MSG, SimDag, SMPI, ...) and raw traces:
651 --cfg=tracing:yes --cfg=tracing/uncategorized:yes --cfg=triva/uncategorized:uncat.plist
653 The first parameter activates the tracing subsystem, the second
654 tells it to trace host and link utilization (without any
655 categorization) and the third creates a graph configuration file
656 to configure Triva when analysing the resulting trace file.
658 - MSG or SimDag-based simulator and categorized traces (you need to declare categories and classify your tasks according to them)
660 --cfg=tracing:yes --cfg=tracing/categorized:yes --cfg=triva/categorized:cat.plist
662 The first parameter activates the tracing subsystem, the second
663 tells it to trace host and link categorized utilization and the
664 third creates a graph configuration file to configure Triva when
665 analysing the resulting trace file.
667 - SMPI simulator and traces for a space/time view:
671 The <i>-trace</i> parameter for the smpirun script runs the
672 simulation with --cfg=tracing:yes and --cfg=tracing/smpi:yes. Check the
673 smpirun's <i>-help</i> parameter for additional tracing options.
675 Sometimes you might want to put additional information on the trace to
676 correctly identify them later, or to provide data that can be used to
677 reproduce an experiment. You have two ways to do that:
679 - Add a string on top of the trace file as comment:
681 --cfg=tracing/comment:my_simulation_identifier
684 - Add the contents of a textual file on top of the trace file as comment:
686 --cfg=tracing/comment_file:my_file_with_additional_information.txt
689 Please, use these two parameters (for comments) to make reproducible
690 simulations. For additional details about this and all tracing
691 options, check See the \ref tracing_tracing_options.
693 \section options_msg Configuring MSG
695 \subsection options_msg_debug_multiple_use Debugging MSG
697 Sometimes your application may try to send a task that is still being
698 executed somewhere else, making it impossible to send this task. However,
699 for debugging purposes, one may want to know what the other host is/was
700 doing. This option shows a backtrace of the other process.
702 Enable this option by adding
705 --cfg=msg/debug_multiple_use:on
708 \section options_smpi Configuring SMPI
710 The SMPI interface provides several specific configuration items.
711 These are uneasy to see since the code is usually launched through the
712 \c smiprun script directly.
714 \subsection options_smpi_bench smpi/bench: Automatic benchmarking of SMPI code
716 In SMPI, the sequential code is automatically benchmarked, and these
717 computations are automatically reported to the simulator. That is to
718 say that if you have a large computation between a \c MPI_Recv() and a
719 \c MPI_Send(), SMPI will automatically benchmark the duration of this
720 code, and create an execution task within the simulator to take this
721 into account. For that, the actual duration is measured on the host
722 machine and then scaled to the power of the corresponding simulated
723 machine. The variable \b smpi/running_power allows to specify the
724 computational power of the host machine (in flop/s) to use when
725 scaling the execution times. It defaults to 20000, but you really want
726 to update it to get accurate simulation results.
728 When the code is constituted of numerous consecutive MPI calls, the
729 previous mechanism feeds the simulation kernel with numerous tiny
730 computations. The \b smpi/cpu_threshold item becomes handy when this
731 impacts badly the simulation performance. It specifies a threshold (in
732 seconds) below which the execution chunks are not reported to the
733 simulation kernel (default value: 1e-6).
737 The option smpi/cpu_threshold ignores any computation time spent
738 below this threshold. SMPI does not consider the \a amount of these
739 computations; there is no offset for this. Hence, by using a
740 value that is too low, you may end up with unreliable simulation
743 In some cases, however, one may wish to disable simulation of
744 application computation. This is the case when SMPI is used not to
745 simulate an MPI applications, but instead an MPI code that performs
746 "live replay" of another MPI app (e.g., ScalaTrace's replay tool,
747 various on-line simulators that run an app at scale). In this case the
748 computation of the replay/simulation logic should not be simulated by
749 SMPI. Instead, the replay tool or on-line simulator will issue
750 "computation events", which correspond to the actual MPI simulation
751 being replayed/simulated. At the moment, these computation events can
752 be simulated using SMPI by calling internal smpi_execute*() functions.
754 To disable the benchmarking/simulation of computation in the simulated
755 application, the variable \b smpi/simulate_computation should be set to no.
756 Equivalently, setting \b smpi/cpu_threshold to -1 also ignores all
760 This option just ignores the timings in your simulation; it still executes
761 the computations itself. If you want to stop SMPI from doing that,
762 you should check the SMPI_SAMPLE macros, documented in the chapter
763 \ref SMPI_adapting_speed.
765 \subsection options_model_smpi_bw_factor smpi/bw_factor: Bandwidth factors
767 The possible throughput of network links is often dependent on the
768 message sizes, as protocols may adapt to different message sizes. With
769 this option, a series of message sizes and factors are given, helping
770 the simulation to be more realistic. For instance, the current
774 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
777 So, messages with size 65472 and more will get a total of MAX_BANDWIDTH*0.940694,
778 messages of size 15424 to 65471 will get MAX_BANDWIDTH*0.697866 and so on.
779 Here, MAX_BANDWIDTH denotes the bandwidth of the link.
782 The SimGrid-Team has developed a script to help you determine these
783 values. You can find more information and the download here:
784 1. http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
785 2. http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
787 \subsection options_smpi_timing smpi/display_timing: Reporting simulation time
789 \b Default: 0 (false)
791 Most of the time, you run MPI code with SMPI to compute the time it
792 would take to run it on a platform. But since the
793 code is run through the \c smpirun script, you don't have any control
794 on the launcher code, making it difficult to report the simulated time
795 when the simulation ends. If you set the \b smpi/display_timing item
796 to 1, \c smpirun will display this information when the simulation ends. \verbatim
797 Simulation time: 1e3 seconds.
800 \subsection options_model_smpi_lat_factor smpi/lat_factor: Latency factors
802 The motivation and syntax for this option is identical to the motivation/syntax
803 of smpi/bw_factor, see \ref options_model_smpi_bw_factor for details.
805 There is an important difference, though: While smpi/bw_factor \a reduces the
806 actual bandwidth (i.e., values between 0 and 1 are valid), latency factors
807 increase the latency, i.e., values larger than or equal to 1 are valid here.
809 This is the default value:
812 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
816 The SimGrid-Team has developed a script to help you determine these
817 values. You can find more information and the download here:
818 1. http://simgrid.gforge.inria.fr/contrib/smpi-calibration-doc.html
819 2. http://simgrid.gforge.inria.fr/contrib/smpi-saturation-doc.html
821 \subsection options_smpi_global smpi/privatize_global_variables: Automatic privatization of global variables
823 MPI executables are meant to be executed in separated processes, but SMPI is
824 executed in only one process. Global variables from executables will be placed
825 in the same memory zone and shared between processes, causing hard to find bugs.
826 To avoid this, several options are possible :
827 - Manual edition of the code, for example to add __thread keyword before data
828 declaration, which allows the resulting code to work with SMPI, but only
829 if the thread factory (see \ref options_virt_factory) is used, as global
830 variables are then placed in the TLS (thread local storage) segment.
831 - Source-to-source transformation, to add a level of indirection
832 to the global variables. SMPI does this for F77 codes compiled with smpiff,
833 and used to provide coccinelle scripts for C codes, which are not functional anymore.
834 - Compilation pass, to have the compiler automatically put the data in
836 - Runtime automatic switching of the data segments. SMPI stores a copy of
837 each global data segment for each process, and at each context switch replaces
838 the actual data with its copy from the right process. This mechanism uses mmap,
839 and is for now limited to systems supporting this functionnality (all Linux
840 and some BSD should be compatible).
841 Another limitation is that SMPI only accounts for global variables defined in
842 the executable. If the processes use external global variables from dynamic
843 libraries, they won't be switched correctly. To avoid this, using static
844 linking is advised (but not with the simgrid library, to avoid replicating
845 its own global variables).
847 To use this runtime automatic switching, the variable \b smpi/privatize_global_variables
852 \subsection options_model_smpi_detached Simulating MPI detached send
854 This threshold specifies the size in bytes under which the send will return
855 immediately. This is different from the threshold detailed in \ref options_model_network_asyncsend
856 because the message is not effectively sent when the send is posted. SMPI still waits for the
857 correspondant receive to be posted to perform the communication operation. This threshold can be set
858 by changing the \b smpi/send_is_detached item. The default value is 65536.
860 \subsection options_model_smpi_collectives Simulating MPI collective algorithms
862 SMPI implements more than 100 different algorithms for MPI collective communication, to accurately
863 simulate the behavior of most of the existing MPI libraries. The \b smpi/coll_selector item can be used
864 to use the decision logic of either OpenMPI or MPICH libraries (values: ompi or mpich, by default SMPI
865 uses naive version of collective operations). Each collective operation can be manually selected with a
866 \b smpi/collective_name:algo_name. Available algorithms are listed in \ref SMPI_collective_algorithms .
868 \subsection options_model_smpi_iprobe smpi/iprobe: Inject constant times for calls to MPI_Iprobe
870 \b Default value: 0.0001
872 The behavior and motivation for this configuration option is identical with \a smpi/test, see
873 Section \ref options_model_smpi_test for details.
875 \subsection options_model_smpi_ois smpi/ois: Inject constant times for asynchronous send operations
877 This configuration option works exactly as \a smpi/os, see Section \ref options_model_smpi_os.
878 Of course, \a smpi/ois is used to account for MPI_Isend instead of MPI_Send.
880 \subsection options_model_smpi_os smpi/os: Inject constant times for send operations
882 In several network models such as LogP, send (MPI_Send, MPI_Isend) and receive (MPI_Recv)
883 operations incur costs (i.e., they consume CPU time). SMPI can factor these costs in as well, but the
884 user has to configure SMPI accordingly as these values may vary by machine.
885 This can be done by using smpi/os for MPI_Send operations; for MPI_Isend and
886 MPI_Recv, use \a smpi/ois and \a smpi/or, respectively. These work exactly as
889 \a smpi/os can consist of multiple sections; each section takes three values, for example:
895 Here, the sections are divided by ";" (that is, this example contains two sections).
896 Furthermore, each section consists of three values.
898 1. The first value denotes the minimum size for this section to take effect;
899 read it as "if message size is greater than this value (and other section has a larger
900 first value that is also smaller than the message size), use this".
901 In the first section above, this value is "1".
903 2. The second value is the startup time; this is a constant value that will always
904 be charged, no matter what the size of the message. In the first section above,
907 3. The third value is the \a per-byte cost. That is, it is charged for every
908 byte of the message (incurring cost messageSize*cost_per_byte)
909 and hence accounts also for larger messages. In the first
910 section of the example above, this value is "2".
912 Now, SMPI always checks which section it should take for a given message; that is,
913 if a message of size 11 is sent with the configuration of the example above, only
914 the second section will be used, not the first, as the first value of the second
915 section is closer to the message size. Hence, a message of size 11 incurs the
916 following cost inside MPI_Send:
922 %As 5 is the startup cost and 1 is the cost per byte.
925 The order of sections can be arbitrary; they will be ordered internally.
927 \subsection options_model_smpi_or smpi/or: Inject constant times for receive operations
929 This configuration option works exactly as \a smpi/os, see Section \ref options_model_smpi_os.
930 Of course, \a smpi/or is used to account for MPI_Recv instead of MPI_Send.
932 \subsection options_model_smpi_test smpi/test: Inject constant times for calls to MPI_Test
934 \b Default value: 0.0001
936 By setting this option, you can control the amount of time a process sleeps
937 when MPI_Test() is called; this is important, because SimGrid normally only
938 advances the time while communication is happening and thus,
939 MPI_Test will not add to the time, resulting in a deadlock if used as a
946 MPI_Test(request, flag, status);
952 Internally, in order to speed up execution, we use a counter to keep track
953 on how often we already checked if the handle is now valid or not. Hence, we
954 actually use counter*SLEEP_TIME, that is, the time MPI_Test() causes the process
955 to sleep increases linearly with the number of previously failed testk.
958 \subsection options_model_smpi_use_shared_malloc smpi/use_shared_malloc: Use shared memory
962 SMPI can use shared memory by calling shm_* functions; this might speed up the simulation.
963 This opens or creates a new POSIX shared memory object, kept in RAM, in /dev/shm.
965 If you want to disable this behavior, set the value to 0.
967 \subsection options_model_smpi_wtime smpi/wtime: Inject constant times for calls to MPI_Wtime
971 By setting this option, you can control the amount of time a process sleeps
972 when MPI_Wtime() is called; this is important, because SimGrid normally only
973 advances the time while communication is happening and thus,
974 MPI_Wtime will not add to the time, resulting in a deadlock if used as a
980 while(MPI_Wtime() < some_time_bound) {
985 If the time is never advanced, this loop will clearly never end as MPI_Wtime()
986 always returns the same value. Hence, pass a (small) value to the smpi/wtime
987 option to force a call to MPI_Wtime to advance the time as well.
990 \section options_generic Configuring other aspects of SimGrid
992 \subsection options_generic_clean_atexit Cleanup before termination
994 The C / C++ standard contains a function called \b [atexit](http://www.cplusplus.com/reference/cstdlib/atexit/).
995 atexit registers callbacks, which are called just before the program terminates.
997 By setting the configuration option clean_atexit to 1 (true), a callback
998 is registered and will clean up some variables and terminate/cleanup the tracing.
1000 TODO: Add when this should be used.
1002 \subsection options_generic_path XML file inclusion path
1004 It is possible to specify a list of directories to search into for the
1005 \<include\> tag in XML files by using the \b path configuration
1006 item. To add several directory to the path, set the configuration
1007 item several times, as in \verbatim
1008 --cfg=path:toto --cfg=path:tutu
1011 \subsection options_generic_exit Behavior on Ctrl-C
1013 By default, when Ctrl-C is pressed, the status of all existing
1014 simulated processes is displayed before exiting the simulation. This is very useful to debug your
1015 code, but it can reveal troublesome in some cases (such as when the
1016 amount of processes becomes really big). This behavior is disabled
1017 when \b verbose-exit is set to 0 (it is to 1 by default).
1019 \subsection options_exception_cutpath Truncate local path from exception backtrace
1021 <b>This configuration option is an internal option and should normally not be used
1022 by the user.</b> It is used to remove the path from the backtrace
1023 shown when an exception is thrown; if we didn't remove this part, the tests
1024 testing the exception parts of simgrid would fail on most machines, as we are
1025 currently comparing output. Clearly, the path used on different machines are almost
1026 guaranteed to be different and hence, the output would
1027 mismatch, causing the test to fail.
1029 \section options_log Logging Configuration
1031 It can be done by using XBT. Go to \ref XBT_log for more details.
1033 \section options_perf Performance optimizations
1035 \subsection options_perf_context Context factory
1037 In order to achieve higher performance, you might want to use the raw
1038 context factory which avoids any system call when switching between
1039 tasks. If it is not possible you might use ucontext instead.
1041 \subsection options_perf_guard_size Disabling stack guard pages
1043 A stack guard page is usually used which prevents the stack from
1044 overflowing on other parts of the memory. However this might have a
1045 performance impact if a huge number of processes is created. The
1046 option \b contexts:guard_size is the number of stack guard pages
1047 used. By setting it to 0, no guard pages will be used: in this case,
1048 you should avoid using small stacks (\b stack_size) as the stack will
1049 silently overflow on other parts of the memory.
1051 \section options_index Index of all existing configuration options
1054 Almost all options are defined in <i>src/simgrid/sg_config.c</i>. You may
1055 want to check this file, too, but this index should be somewhat complete
1056 for the moment (May 2015).
1059 \b Please \b note: You can also pass the command-line option "--help" and
1060 "--help-cfg" to an executable that uses simgrid.
1062 - \c clean_atexit: \ref options_generic_clean_atexit
1064 - \c contexts/factory: \ref options_virt_factory
1065 - \c contexts/guard_size: \ref options_virt_parallel
1066 - \c contexts/nthreads: \ref options_virt_parallel
1067 - \c contexts/parallel_threshold: \ref options_virt_parallel
1068 - \c contexts/stack_size: \ref options_virt_stacksize
1069 - \c contexts/synchro: \ref options_virt_parallel
1071 - \c cpu/maxmin_selective_update: \ref options_model_optim
1072 - \c cpu/model: \ref options_model_select
1073 - \c cpu/optim: \ref options_model_optim
1075 - \c exception/cutpath: \ref options_exception_cutpath
1077 - \c host/model: \ref options_model_select
1079 - \c maxmin/precision: \ref options_model_precision
1081 - \c msg/debug_multiple_use: \ref options_msg_debug_multiple_use
1083 - \c model-check: \ref options_modelchecking
1084 - \c model-check/checkpoint: \ref options_modelchecking_steps
1085 - \c model-check/communications_determinism: \ref options_modelchecking_comm_determinism
1086 - \c model-check/send_determinism: \ref options_modelchecking_comm_determinism
1087 - \c model-check/dot_output: \ref options_modelchecking_dot_output
1088 - \c model-check/hash: \ref options_modelchecking_hash
1089 - \c model-check/property: \ref options_modelchecking_liveness
1090 - \c model-check/max_depth: \ref options_modelchecking_max_depth
1091 - \c model-check/record: \ref options_modelchecking_recordreplay
1092 - \c model-check/reduction: \ref options_modelchecking_reduction
1093 - \c model-check/replay: \ref options_modelchecking_recordreplay
1094 - \c model-check/send_determinism: \ref options_modelchecking_sparse_checkpoint
1095 - \c model-check/sparse_checkpoint: \ref options_modelchecking_sparse_checkpoint
1096 - \c model-check/termination: \ref options_modelchecking_termination
1097 - \c model-check/timeout: \ref options_modelchecking_timeout
1098 - \c model-check/visited: \ref options_modelchecking_visited
1100 - \c network/bandwidth_factor: \ref options_model_network_coefs
1101 - \c network/coordinates: \ref options_model_network_coord
1102 - \c network/crosstraffic: \ref options_model_network_crosstraffic
1103 - \c network/latency_factor: \ref options_model_network_coefs
1104 - \c network/maxmin_selective_update: \ref options_model_optim
1105 - \c network/model: \ref options_model_select
1106 - \c network/optim: \ref options_model_optim
1107 - \c network/sender_gap: \ref options_model_network_sendergap
1108 - \c network/TCP_gamma: \ref options_model_network_gamma
1109 - \c network/weight_S: \ref options_model_network_coefs
1111 - \c ns3/TcpModel: \ref options_pls
1112 - \c path: \ref options_generic_path
1113 - \c plugin: \ref options_generic_plugin
1115 - \c surf/nthreads: \ref options_model_nthreads
1116 - \c surf/precision: \ref options_model_precision
1118 - \c <b>For collective operations of SMPI, please refer to Section \ref options_index_smpi_coll</b>
1119 - \c smpi/async_small_thresh: \ref options_model_network_asyncsend
1120 - \c smpi/bw_factor: \ref options_model_smpi_bw_factor
1121 - \c smpi/coll_selector: \ref options_model_smpi_collectives
1122 - \c smpi/cpu_threshold: \ref options_smpi_bench
1123 - \c smpi/display_timing: \ref options_smpi_timing
1124 - \c smpi/lat_factor: \ref options_model_smpi_lat_factor
1125 - \c smpi/IB_penalty_factors: \ref options_model_network_coefs
1126 - \c smpi/iprobe: \ref options_model_smpi_iprobe
1127 - \c smpi/ois: \ref options_model_smpi_ois
1128 - \c smpi/or: \ref options_model_smpi_or
1129 - \c smpi/os: \ref options_model_smpi_os
1130 - \c smpi/privatize_global_variables: \ref options_smpi_global
1131 - \c smpi/running_power: \ref options_smpi_bench
1132 - \c smpi/send_is_detached_thresh: \ref options_model_smpi_detached
1133 - \c smpi/simulate_computation: \ref options_smpi_bench
1134 - \c smpi/test: \ref options_model_smpi_test
1135 - \c smpi/use_shared_malloc: \ref options_model_smpi_use_shared_malloc
1136 - \c smpi/wtime: \ref options_model_smpi_wtime
1138 - \c <b>Tracing configuration options can be found in Section \ref tracing_tracing_options</b>.
1140 - \c storage/model: \ref options_storage_model
1141 - \c verbose-exit: \ref options_generic_exit
1143 - \c vm/model: \ref options_vm_model
1145 \subsection options_index_smpi_coll Index of SMPI collective algorithms options
1147 TODO: All available collective algorithms will be made available via the ``smpirun --help-coll`` command.