6 SimGrid was conceived as a tool to study distributed algorithms. Its
7 modern :ref:`S4U interface <S4U_doc>` makes it easy to assess Cloud,
8 P2P, HPC, IoT and similar settings.
10 A typical SimGrid simulation is composed of several |Actors|_, that
11 execute user-provided functions. The actors have to explicitly use the
12 S4U interface to express their computation, communication, disk usage
13 and other |Activities|_, so that they get reflected within the
14 simulator. These activities take place on **Resources** (|Hosts|_,
15 |Links|_, |Storages|_). SimGrid predicts the time taken by each
16 activity and orchestrates accordingly the actors waiting for the
17 completion of these activities.
19 Each actor executes a user-provided function on a simulated |Host|_
20 with which it can interact. Communications are not directly sent to
21 actors, but posted onto a |Mailbox|_ that serve as rendez-vous point
22 between communicating actors.
24 .. |Actors| replace:: **Actors**
25 .. _Actors: api/classsimgrid_1_1s4u_1_1Actor.html
27 .. |Activities| replace:: **Activities**
28 .. _Activities: api/classsimgrid_1_1s4u_1_1Activity.html
30 .. |Hosts| replace:: **Hosts**
31 .. _Hosts: api/classsimgrid_1_1s4u_1_1Host.html
33 .. |Links| replace:: **Links**
34 .. _Links: api/classsimgrid_1_1s4u_1_1Link.html
36 .. |Storages| replace:: **Storages**
37 .. _Storages: api/classsimgrid_1_1s4u_1_1Storage.html
39 .. |VirtualMachines| replace:: **VirtualMachines**
40 .. _VirtualMachines: api/classsimgrid_1_1s4u_1_1VirtualMachine.html
42 .. |Host| replace:: **Host**
43 .. _Host: api/classsimgrid_1_1s4u_1_1Host.html
45 .. |Link| replace:: **Link**
46 .. _Link: api/classsimgrid_1_1s4u_1_1Link.html
48 .. |Mailbox| replace:: **Mailbox**
49 .. _Mailbox: api/classsimgrid_1_1s4u_1_1Mailbox.html
51 .. |Barrier| replace:: **Barrier**
52 .. _Barrier: api/classsimgrid_1_1s4u_1_1Barrier.html
54 .. |ConditionVariable| replace:: **ConditionVariable**
55 .. _ConditionVariable: api/classsimgrid_1_1s4u_1_1ConditionVariable.html
57 .. |Mutex| replace:: **Mutex**
58 .. _Mutex: api/classsimgrid_1_1s4u_1_1Mutex.html
61 **In the remainder of this tutorial**, you will discover a simple yet
62 fully functioning example of SimGrid simulation: the Master/Workers
63 application. We will detail each part of the code and necessary
64 configuration to make it working. After this tour, several exercises
65 are proposed to let you discover some of the SimGrid features, hands
66 on the keyboard. This practical session will be given in C++, that you
67 are supposed to know beforehand.
70 Discover the Master/Workers
71 ---------------------------
73 This section introduces a first example of SimGrid simulation. This
74 simple application is composed of two kind of actors: the **master**
75 is in charge of distributing some computational tasks to a set of
76 **workers** that execute them.
78 .. image:: /tuto_s4u/img/intro.svg
81 We first present a round-robin version of this application, where the
82 master dispatches the tasks to the workers, one after the other, until
83 all tasks are dispatched. Later in this tutorial, you will be given
84 the opportunity to improve this scheme.
89 Let's start with the code of the worker. It is represented by the
90 *master* function below. This simple function takes at least 3
91 parameters (the amount of tasks to dispatch, their computational size
92 in flops to compute and their communication size in bytes to
93 exchange). Every parameter after the third one must be the name of an
94 host on which a worker is waiting for something to compute.
96 Then, the tasks are sent one after the other, each on a mailbox named
97 after the worker's hosts. On the other side, a given worker (which
98 code is given below) wait for incoming tasks on its own
103 At the end, once all tasks are dispatched, the master dispatches
104 another task per worker, but this time with a negative amount of flops
105 to compute. Indeed, this application decided by convention, that the
106 workers should stop when encountering such a negative compute_size.
108 At the end of the day, the only SimGrid specific functions used in
109 this example are :cpp:func:`simgrid::s4u::Mailbox::by_name` and
110 :cpp:func:`simgrid::s4u::Mailbox::put`. Also, :c:macro:`XBT_INFO` is used
111 as a replacement to printf() or to cout to ensure that the messages
112 are nicely logged along with the simulated time and actor name.
115 .. literalinclude:: ../../examples/s4u/app-masterworkers/s4u-app-masterworkers-fun.cpp
117 :start-after: master-begin
118 :end-before: master-end
120 Here comes the code of the worker actors. This function expects no
121 parameter from its vector of strings. Its code is very simple: it
122 expects messages on the mailbox that is named after its own host. As long as it gets valid
123 computation requests (whose compute_amount is positive), it compute
124 this task and waits for the next one.
126 The worker retrieves its own host with
127 :cpp:func:`simgrid::s4u::this_actor::get_host`. The
128 :ref:`simgrid::s4u::this_actor <namespace_simgrid__s4u__this_actor>`
129 namespace contains many such helping functions.
131 .. literalinclude:: ../../examples/s4u/app-masterworkers/s4u-app-masterworkers-fun.cpp
133 :start-after: worker-begin
134 :end-before: worker-end
136 Starting the Simulation
137 .......................
139 And this is it. In only a few lines, we defined the algorithm of our
140 master/workers examples.
142 That being said, an algorithm alone is not enough to define a
143 simulation: SimGrid is a library, not a program. So you need to define
144 your own ``main()`` function as follows. This function is in charge of
145 creating a SimGrid simulation engine (on line 3), register the actor
146 functions to the engine (on lines 7 and 8), load the virtual platform
147 from its description file (on line 11), map actors onto that platform
148 (on line 12) and run the simulation until its completion on line 15.
150 .. literalinclude:: ../../examples/s4u/app-masterworkers/s4u-app-masterworkers-fun.cpp
152 :start-after: main-begin
153 :end-before: main-end
156 As you can see, this also requires a platform file and a deployment
162 Platform files define the virtual platform on which the provided
163 application will take place. In contains one or several **Network
164 Zone** |api_s4u_NetZone|_ that contain both |Host|_ and |Link|_
165 Resources, as well as routing information.
167 Such files can get rather long and boring, so the example below is
168 only an excerpts of the full ``examples/platforms/small_platform.xml``
169 file. For example, most routing information are missing, and only the
170 route between the hosts Tremblay and Fafard is given. This path
171 traverses 6 links (named 4, 3, 2, 0, 1 and 8). There are several
172 examples of platforms in the archive under ``examples/platforms``.
174 .. |api_s4u_NetZone| image:: /img/extlink.png
177 .. _api_s4u_NetZone: api/classsimgrid_1_1s4u_1_1NetZone.html#class-documentation
179 .. |api_s4u_Link| image:: /img/extlink.png
182 .. _api_s4u_Link: api/classsimgrid_1_1s4u_1_1Link.html#class-documentation
184 .. literalinclude:: ../../examples/platforms/small_platform.xml
186 :lines: 1-10,12-20,56-62,192-
187 :caption: (excerpts of the small_platform.xml file)
192 Deployment files specify the execution scenario: it lists the actors
193 that should be started, along with their parameter. In the following
194 example, we start 6 actors: one master and 5 workers.
196 .. literalinclude:: ../../examples/s4u/app-masterworkers/s4u-app-masterworkers_d.xml
202 This time, we have all parts: once the program is compiled, we can
203 execute it as follows. Note how the XBT_INFO() requests turned into
204 informative messages.
206 .. literalinclude:: ../../examples/s4u/app-masterworkers/s4u-app-masterworkers.tesh
208 :start-after: s4u-app-masterworkers-fun
209 :prepend: $$$ ./masterworkers platform.xml deploy.xml
217 In this section, you will modify the example presented earlier to
218 explore the quality of the proposed algorithm. For now, it works and
219 the simulation prints things, but the truth is that we have no idea of
220 whether this is a good algorithm to dispatch tasks to the workers.
221 This very simple setting raises many interesting questions:
223 .. image:: /tuto_s4u/img/question.svg
226 - Which algorithm should the master use? Or should the worker decide
229 Round Robin is not an efficient algorithm when all tasks are not
230 processed at the same speed. It would probably be more efficient
231 if the workers were asking for tasks when ready.
233 - Should tasks be grouped in batches or sent separately?
235 The workers will starve if they don't get the tasks fast
236 enough. One possibility to reduce latency would be to send tasks
237 in pools instead of one by one. But if the pools are too big, the
238 load balancing will likely get uneven, in particular when
239 distributing the last tasks.
241 - How does the quality of such algorithm dependent on the platform
242 characteristics and on the task characteristics?
244 Whenever the input communication time is very small compared to
245 processing time and workers are homogeneous, it is likely that the
246 round-robin algorithm performs very well. Would it still hold true
247 when transfer time is not negligible? What if some tasks are
248 performed faster on some specific nodes?
250 - The network topology interconnecting the master and the workers
251 may be quite complicated. How does such a topology impact the
254 When data transfers are the bottleneck, it is likely that a good
255 modeling of the platform becomes essential. The SimGrid platform
256 models are particularly handy to account for complex platform
259 - What is the best applicative topology?
261 Is a flat master worker deployment sufficient? Should we go for a
262 hierarchical algorithm, with some forwarders taking large pools of
263 tasks from the master, each of them distributing their tasks to a
264 sub-pool of workers? Or should we introduce super-peers,
265 dupplicating the master's role in a peer-to-peer manner? Do the
266 algorithms require a perfect knowledge of the network?
268 - How is such an algorithm sensitive to external workload variation?
270 What if bandwidth, latency and computing speed can vary with no
271 warning? Shouldn't you study whether your algorithm is sensitive
272 to such load variations?
274 - Although an algorithm may be more efficient than another, how does
275 it interfere with unrelated applications executing on the same
278 **SimGrid was invented to answer such questions.** Do not believe the
279 fools saying that all you need to study such settings is a simple
280 discrete event simulator. Do you really want to reinvent the wheel,
281 debug and optimize your own tool, and validate its models against real
282 settings for ages, or do you prefer to sit on the shoulders of a
283 giant? With SimGrid, you can focus on your algorithm. The whole
284 simulation mechanism is already working.
286 Here is the visualization of a SimGrid simulation of two master worker
287 applications (one in light gray and the other in dark gray) running in
288 concurrence and showing resource usage over a long period of time. It
289 was obtained with the Triva software.
291 .. image:: /tuto_s4u/img/result.png
297 The easiest way to take the tutorial is to use the dedicated Docker
298 image. Once you `installed Docker itself
299 <https://docs.docker.com/install/>`_, simply do the following:
301 .. code-block:: shell
303 docker pull simgrid/tuto-s4u
304 docker run -it --rm --name simgrid --volume ~/simgrid-tutorial:/source/tutorial simgrid/tuto-s4u bash
306 This will start a new container with all you need to take this
307 tutorial, and create a ``simgrid-tutorial`` directory in your home on
308 your host machine that will be visible as ``/source/tutorial`` within the
309 container. You can then edit the files you want with your favorite
310 editor in ``~/simgrid-tutorial``, and compile them within the
311 container to enjoy the provided dependencies.
315 Any change to the container out of ``/source/tutorial`` will be lost
316 when you log out of the container, so don't edit the other files!
318 All needed dependencies are already installed in this container
319 (SimGrid, a C++ compiler, cmake, pajeng and R). Vite being only
320 optional in this tutorial, it is not installed to reduce the image
323 The code template is available under ``/source/simgrid-template-s4u`` in
324 the image. You should copy it to your working directory when you first
327 .. code-block:: shell
329 cp -r /source/simgrid-template-s4u/* /source/tutorial
332 Using your Computer Natively
333 ............................
335 To take the tutorial on your machine, you first need to :ref:`install
336 SimGrid <install>`, a C++ compiler and also ``pajeng`` to visualize
337 the traces. You may want to install `Vite
338 <http://vite.gforge.inria.fr/>`_ to get a first glance at the
339 traces. The provided code template requires cmake to compile. On
340 Debian and Ubuntu for example, you can get them as follows:
342 .. code-block:: shell
344 sudo apt install simgrid pajeng cmake g++ vite
346 An initial version of the source code is provided on framagit. This
347 template compiles with cmake. If SimGrid is correctly installed, you
348 should be able to clone the `repository
349 <https://framagit.org/simgrid/simgrid-template-s4u>`_ and recompile
350 everything as follows:
352 .. code-block:: shell
354 git clone git@framagit.org:simgrid/simgrid-template-s4u.git
355 cd simgrid-template-s4u/
359 If you struggle with the compilation, then you should double check
360 your :ref:`SimGrid installation <install>`. On need, please refer to
361 the :ref:`Troubleshooting your Project Setup
362 <install_yours_troubleshooting>` section.
364 Discovering the Provided Code
365 .............................
367 Please compile and execute the provided simulator as follows:
369 .. code-block:: shell
372 ./master-workers small_platform.xml master-workers_d.xml
374 For a more "fancy" output, you can use simgrid-colorizer.
376 .. code-block:: shell
378 ./master-workers small_platform.xml master-workers_d.xml 2>&1 | simgrid-colorizer
380 If you installed SimGrid to a non-standard path, you may have to
381 specify the full path to simgrid-colorizer on the above line, such as
382 ``/opt/simgrid/bin/simgrid-colorizer``. If you did not install it at all,
383 you can find it in <simgrid_root_directory>/bin/colorize.
385 For a classical Gantt-Chart vizualisation, you can use `Vite
386 <http://vite.gforge.inria.fr/>`_ if you have it installed, as
387 follows. But do not spend too much time installing Vite, because there
388 is a better way to visualize SimGrid traces (see below).
390 .. code-block:: shell
392 ./master-workers small_platform.xml master-workers_d.xml --cfg=tracing:yes --cfg=tracing/msg/process:yes
395 .. image:: /tuto_s4u/img/vite-screenshot.png
398 If you want the full power to visualize SimGrid traces, you need
399 to use R. As a start, you can download this `starter script
400 <https://framagit.org/simgrid/simgrid/raw/master/docs/source/tuto_s4u/draw_gantt.R>`_
401 and use it as follows:
403 .. code-block:: shell
405 ./master-workers small_platform.xml master-workers_d.xml --cfg=tracing:yes --cfg=tracing/msg/process:yes
406 pj_dump --ignore-incomplete-links simgrid.trace | grep STATE > gantt.csv
407 Rscript draw_gantt.R gantt.csv
409 It produces a ``Rplots.pdf`` with the following content:
411 .. image:: /tuto_s4u/img/Rscript-screenshot.png
415 Lab 1: Simpler Deployments
416 --------------------------
418 In the provided example, adding more workers quickly becomes a pain:
419 You need to start them (at the bottom of the file), and to inform the
420 master of its availability with an extra parameter. This is mandatory
421 if you want to inform the master of where the workers are running. But
422 actually, the master does not need to have this information.
424 We could leverage the mailbox mechanism flexibility, and use a sort of
425 yellow page system: Instead of sending data to the worker running on
426 Fafard, the master could send data to the third worker. Ie, instead of
427 using the worker location (which should be filled in two locations),
428 we could use their ID (which should be filled in one location
431 This could be done with the following deployment file. It's clearly
432 not shorter than the previous one, but it's still simpler because the
433 information is only written once. It thus follows the `DRY
434 <https://en.wikipedia.org/wiki/Don't_repeat_yourself>`_ `SPOT
435 <http://wiki.c2.com/?SinglePointOfTruth>`_ design principle.
437 .. literalinclude:: tuto_s4u/deployment1.xml
441 Copy your ``master-workers.cpp`` into ``master-workers-lab1.cpp`` and
442 add a new executable into ``CMakeLists.txt``. Then modify your worker
443 function so that it gets its mailbox name not from the name of its
444 host, but from the string passed as ``args[1]``. The master will send
445 messages to all workers based on their number, for example as follows:
449 for (int i = 0; i < tasks_count; i++) {
450 std::string worker_rank = std::to_string(i % workers_count);
451 std::string mailbox_name = std::string("worker-") + worker_rank;
452 simgrid::s4u::MailboxPtr mailbox = simgrid::s4u::Mailbox::by_name(mailbox_name);
463 The mailboxes are a very powerful mechanism in SimGrid, allowing many
464 interesting application settings. They may feel surprising if you are
465 used to BSD sockets or other classical systems, but you will soon
466 appreciate their power. They are only used to match the
467 communications, but have no impact on the communication
468 timing. ``put()`` and ``get()`` are matched regardless of their
469 initiators' location and then the real communication occures between
470 the involved parties.
472 Please refer to the full `API of Mailboxes
473 <api/classsimgrid_1_1s4u_1_1Mailbox.html#class-documentation>`_ for
477 Lab 2: Using the Whole Platform
478 -------------------------------
480 It is now easier to add a new worker, but you still has to do it
481 manually. It would be much easier if the master could start the
482 workers on its own, one per available host in the platform. The new
483 deployment file should be as simple as:
485 .. literalinclude:: tuto_s4u/deployment2.xml
489 Creating the workers from the master
490 ....................................
492 For that, the master needs to retrieve the list of hosts declared in
493 the platform with :cpp:func:`simgrid::s4u::Engine::get_all_hosts`.
494 Then, the master should start the worker actors with
495 :cpp:func:`simgrid::s4u::Actor::create`.
497 ``Actor::create(name, host, func, params...)`` is a very flexible
498 function. Its third parameter is the function that the actor should
499 execute. This function can take any kind of parameter, provided that
500 you pass similar parameters to ``Actor::create()``. For example, you
501 could have something like this:
505 void my_actor(int param1, double param2, std::string param3) {
508 int main(int argc, char argv**) {
510 simgrid::s4u::ActorPtr actor;
511 actor = simgrid::s4u::Actor::create("name", simgrid::s4u::Host::by_name("the_host"),
512 &my_actor, 42, 3.14, "thevalue");
517 Master-Workers Communication
518 ............................
520 Previously, the workers got from their parameter the name of the
521 mailbox they should use. We can still do so: the master should build
522 such a parameter before using it in the ``Actor::create()`` call. The
523 master could even pass directly the mailbox as a parameter to the
526 Since we want later to study concurrent applications, it is advised to
527 use a mailbox name that is unique over the simulation even if there is
528 more than one master.
530 One possibility for that is to use the actor ID (aid) of each worker
531 as a mailbox name. The master can retrieve the aid of the newly
532 created actor with ``actor->get_pid()`` while the actor itself can
533 retrieve its own aid with ``simgrid::s4u::this_actor::get_pid()``.
534 The retrieved value is an ``aid_t``, which is an alias for ``long``.
536 Instead of having one mailbox per worker, you could also reorganize
537 completely your application to have only one mailbox per master. All
538 the workers of a given master would pull their work from the same
539 mailbox, which should be passed as parameter to the workers. This
540 reduces the amount of mailboxes, but prevents the master from taking
541 any scheduling decision. It really depends on how you want to organize
542 your application and what you want to study with your simulator. In
543 this tutorial, that's probably not a good idea.
548 In this exercise, we reduced the amount of configuration that our
549 simulator requests. This is both a good idea, and a dangerous
550 trend. This simplification is another application of the good old DRY/SPOT
551 programming principle (`Don't Repeat Yourself / Single Point Of Truth
552 <https://en.wikipedia.org/wiki/Don%27t_repeat_yourself>`_), and you
553 really want your programming artefacts to follow these software
554 engineering principles.
556 But at the same time, you should be careful in separating your
557 scientific contribution (the master/workers algorithm) and the
558 artefacts used to test it (platform, deployment and workload). This is
559 why SimGrid forces you to express your platform and deployment files
560 in XML instead of using a programming interface: it forces a clear
561 separation of concerns between things of very different nature.
563 Lab 3: Fixed Experiment Duration
564 --------------------------------
566 In the current version, the number of tasks is defined through the
567 worker arguments. Hence, tasks are created at the very beginning of
568 the simulation. Instead, have the master dispatching tasks for a
569 predetermined amount of time. The tasks must now be created on demand
570 instead of beforehand.
572 Of course, usual time functions like ``gettimeofday`` will give you the
573 time on your real machine, which is prety useless in the
574 simulation. Instead, retrieve the time in the simulated world with
575 :cpp:func:`simgrid::s4u::Engine::get_clock`.
577 You can still stop your workers with a specific task as previously,
578 or you may kill them forcefully with
579 :cpp:func:`simgrid::s4u::Actor::kill` (if you already have a reference
580 to the actor you want to kill) or
581 :cpp:func:`void simgrid::s4u::Actor::kill(aid_t)` (if you only have its ID).
584 Anyway, the new deployment `deployment3.xml` file should thus look
587 .. literalinclude:: tuto_s4u/deployment3.xml
590 Controlling the message verbosity
591 .................................
593 Not all messages are equally informative, so you probably want to
594 change some of the ``XBT_INFO`` into ``XBT_DEBUG`` so that they are
595 hidden by default. For example, you may want to use ``XBT_INFO`` once
596 every 100 tasks and ``XBT_DEBUG`` when sending all the other tasks. Or
597 you could show only the total number of tasks processed by
598 default. You can still see the debug messages as follows:
600 .. code-block:: shell
602 ./master-workers-lab3 small_platform.xml deployment3.xml --log=msg_test.thres:debug
605 Lab 4: Competing Applications
606 -----------------------------
608 It is now time to start several applications at once, with the following ``deployment4.xml`` file.
610 .. literalinclude:: tuto_s4u/deployment4.xml
613 Things happen when you do so, but it remains utterly difficult to
614 understand what's happening exactely. Even Gantt visualizations
615 contain too much information to be useful: it is impossible to
616 understand which task belong to which application. To fix this, we
617 will categorize the tasks.
619 Instead of starting the execution in one function call only with
620 ``this_actor::execute(cost)``, you need to
621 create the execution activity, set its tracing category, and then start
622 it and wait for its completion, as follows:
626 simgrid::s4u::ExecPtr exec = simgrid::s4u::this_actor::exec_init(compute_cost);
627 exec->set_tracing_category(category);
628 // exec->start() is optional here as wait() starts the activity on need
631 You can make the same code shorter as follows:
635 simgrid::s4u::this_actor::exec_init(compute_cost)->set_tracing_category(category)->wait();
637 Visualizing the result
638 .......................
640 vite is not enough to understand the situation, because it does not
641 deal with categorization. This time, you absolutely must switch to R,
642 as explained on `this page
643 <http://simgrid.gforge.inria.fr/contrib/R_visualization.php>`_.
647 Include here the minimal setting to view something in R.
650 Lab 5: Better Scheduling
651 ------------------------
653 You don't need a very advanced visualization solution to notice that
654 round-robin is completely suboptimal: most of the workers keep waiting
655 for more work. We will move to a First-Come First-Served mechanism
658 For that, your workers should explicitely request for work with a
659 message sent to a channel that is specific to their master. The name
660 of that private channel can be the one used to categorize the
661 executions, as it is already specific to each master.
663 The master should serve in a round-robin manner the requests it
664 receives, until the time is up. Changing the communication schema can
665 be a bit hairy, but once it works, you will see that such as simple
666 FCFS schema allows to double the amount of tasks handled over time
667 here. Things may be different with another platform file.
672 From this, many things can easily be added. For example, you could:
674 - Allow workers to have several pending requests so as to overlap
675 communication and computations as much as possible. Non-blocking
676 communication will probably become handy here.
677 - Add a performance measurement mechanism, enabling the master to make smart scheduling choices.
678 - Test your code on other platforms, from the ``examples/platforms``
679 directory in your archive.
681 What is the largest number of tasks requiring 50e6 flops and 1e5
682 bytes that you manage to distribute and process in one hour on
684 - Optimize not only for the amount of tasks handled, but also for the
685 total energy dissipated.
686 - And so on. If you come up with a really nice extension, please share
687 it with us so that we can extend this tutorial.
692 This tutorial is now terminated. You could keep reading the [online documentation][fn:4] or
693 [tutorials][fn:7], or you could head up to the example section to read some code.
697 TODO: Points to improve for the next time
699 - Propose equivalent exercises and skeleton in java.
700 - Propose a virtualbox image with everything (simgrid, pajeng, ...) already set up.
701 - Ease the installation on mac OS X (binary installer) and windows.
703 .. LocalWords: SimGrid