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: app_s4u.html#s4u-actor
27 .. |Activities| replace:: **Activities**
28 .. _Activities: app_s4u.html#s4u-activity
30 .. |Hosts| replace:: **Hosts**
31 .. _Hosts: app_s4u.html#s4u-host
33 .. |Links| replace:: **Links**
34 .. _Links: app_s4u.html#s4u-link
36 .. |Storages| replace:: **Storages**
37 .. _Storages: app_s4u.html#s4u-storage
39 .. |VirtualMachines| replace:: **VirtualMachines**
40 .. _VirtualMachines: app_s4u.html#s4u-virtualmachine
42 .. |Host| replace:: **Host**
43 .. _Host: app_s4u.html#s4u-host
45 .. |Link| replace:: **Link**
46 .. _Link: app_s4u.html#s4u-link
48 .. |Mailbox| replace:: **Mailbox**
49 .. _Mailbox: app_s4u.html#s4u-mailbox
51 .. |Barrier| replace:: **Barrier**
52 .. _Barrier: app_s4u.html#s4u-barrier
54 .. |ConditionVariable| replace:: **ConditionVariable**
55 .. _ConditionVariable: app_s4u.html#s4u-conditionvariable
57 .. |Mutex| replace:: **Mutex**
58 .. _Mutex: app_s4u.html#s4u-mutex
60 **In the remainder of this tutorial**, you will discover a simple yet
61 fully functioning example of SimGrid simulation: the Master/Workers
62 application. We will detail each part of the code and necessary
63 configuration to make it working. After this tour, several exercises
64 are proposed to let you discover some of the SimGrid features, hands
65 on the keyboard. This practical session will be given in C++, that you
66 are supposed to know beforehand.
69 Discover the Master/Workers
70 ---------------------------
72 This section introduces a first example of SimGrid simulation. This
73 simple application is composed of two kind of actors: the **master**
74 is in charge of distributing some computational tasks to a set of
75 **workers** that execute them.
77 .. image:: /tuto_s4u/img/intro.svg
80 We first present a round-robin version of this application, where the
81 master dispatches the tasks to the workers, one after the other, until
82 all tasks are dispatched. Later in this tutorial, you will be given
83 the opportunity to improve this scheme.
88 Let's start with the code of the worker. It is represented by the
89 *master* function below. This simple function takes at least 3
90 parameters (the amount of tasks to dispatch, their computational size
91 in flops to compute and their communication size in bytes to
92 exchange). Every parameter after the third one must be the name of an
93 host on which a worker is waiting for something to compute.
95 Then, the tasks are sent one after the other, each on a mailbox named
96 after the worker's hosts. On the other side, a given worker (which
97 code is given below) wait for incoming tasks on its own
102 At the end, once all tasks are dispatched, the master dispatches
103 another task per worker, but this time with a negative amount of flops
104 to compute. Indeed, this application decided by convention, that the
105 workers should stop when encountering such a negative compute_size.
107 At the end of the day, the only SimGrid specific functions used in
108 this example are :cpp:func:`simgrid::s4u::Mailbox::by_name` and
109 :cpp:func:`simgrid::s4u::Mailbox::put`. Also, :c:macro:`XBT_INFO` is used
110 as a replacement to printf() or to cout to ensure that the messages
111 are nicely logged along with the simulated time and actor name.
114 .. literalinclude:: ../../examples/s4u/app-masterworkers/s4u-app-masterworkers-fun.cpp
116 :start-after: master-begin
117 :end-before: master-end
119 Here comes the code of the worker actors. This function expects no
120 parameter from its vector of strings. Its code is very simple: it
121 expects messages on the mailbox that is named after its own host. As long as it gets valid
122 computation requests (whose compute_amount is positive), it compute
123 this task and waits for the next one.
125 The worker retrieves its own host with
126 :cpp:func:`simgrid::s4u::this_actor::get_host`. The
127 :ref:`simgrid::s4u::this_actor <API_s4u_this_actor>`
128 namespace contains many such helping functions.
130 .. literalinclude:: ../../examples/s4u/app-masterworkers/s4u-app-masterworkers-fun.cpp
132 :start-after: worker-begin
133 :end-before: worker-end
135 Starting the Simulation
136 .......................
138 And this is it. In only a few lines, we defined the algorithm of our
139 master/workers examples.
141 That being said, an algorithm alone is not enough to define a
142 simulation: SimGrid is a library, not a program. So you need to define
143 your own ``main()`` function as follows. This function is in charge of
144 creating a SimGrid simulation engine (on line 3), register the actor
145 functions to the engine (on lines 7 and 8), load the simulated platform
146 from its description file (on line 11), map actors onto that platform
147 (on line 12) and run the simulation until its completion on line 15.
149 .. literalinclude:: ../../examples/s4u/app-masterworkers/s4u-app-masterworkers-fun.cpp
151 :start-after: main-begin
152 :end-before: main-end
155 As you can see, this also requires a platform file and a deployment
161 Platform files define the simulated platform on which the provided
162 application will take place. In contains one or several **Network
163 Zone** |api_s4u_NetZone|_ that contain both |Host|_ and |Link|_
164 Resources, as well as routing information.
166 Such files can get rather long and boring, so the example below is
167 only an excerpts of the full ``examples/platforms/small_platform.xml``
168 file. For example, most routing information are missing, and only the
169 route between the hosts Tremblay and Fafard is given. This path
170 traverses 6 links (named 4, 3, 2, 0, 1 and 8). There are several
171 examples of platforms in the archive under ``examples/platforms``.
173 .. |api_s4u_NetZone| image:: /img/extlink.png
176 .. _api_s4u_NetZone: api/classsimgrid_1_1s4u_1_1NetZone.html#class-documentation
178 .. |api_s4u_Link| image:: /img/extlink.png
181 .. _api_s4u_Link: api/classsimgrid_1_1s4u_1_1Link.html#class-documentation
183 .. literalinclude:: ../../examples/platforms/small_platform.xml
185 :lines: 1-10,12-20,56-62,192-
186 :caption: (excerpts of the small_platform.xml file)
191 Deployment files specify the execution scenario: it lists the actors
192 that should be started, along with their parameter. In the following
193 example, we start 6 actors: one master and 5 workers.
195 .. literalinclude:: ../../examples/s4u/app-masterworkers/s4u-app-masterworkers_d.xml
201 This time, we have all parts: once the program is compiled, we can
202 execute it as follows. Note how the XBT_INFO() requests turned into
203 informative messages.
205 .. literalinclude:: ../../examples/s4u/app-masterworkers/s4u-app-masterworkers.tesh
207 :start-after: s4u-app-masterworkers-fun
208 :prepend: $$$ ./masterworkers platform.xml deploy.xml
216 In this section, you will modify the example presented earlier to
217 explore the quality of the proposed algorithm. For now, it works and
218 the simulation prints things, but the truth is that we have no idea of
219 whether this is a good algorithm to dispatch tasks to the workers.
220 This very simple setting raises many interesting questions:
222 .. image:: /tuto_s4u/img/question.svg
225 - Which algorithm should the master use? Or should the worker decide
228 Round Robin is not an efficient algorithm when all tasks are not
229 processed at the same speed. It would probably be more efficient
230 if the workers were asking for tasks when ready.
232 - Should tasks be grouped in batches or sent separately?
234 The workers will starve if they don't get the tasks fast
235 enough. One possibility to reduce latency would be to send tasks
236 in pools instead of one by one. But if the pools are too big, the
237 load balancing will likely get uneven, in particular when
238 distributing the last tasks.
240 - How does the quality of such algorithm dependent on the platform
241 characteristics and on the task characteristics?
243 Whenever the input communication time is very small compared to
244 processing time and workers are homogeneous, it is likely that the
245 round-robin algorithm performs very well. Would it still hold true
246 when transfer time is not negligible? What if some tasks are
247 performed faster on some specific nodes?
249 - The network topology interconnecting the master and the workers
250 may be quite complicated. How does such a topology impact the
253 When data transfers are the bottleneck, it is likely that a good
254 modeling of the platform becomes essential. The SimGrid platform
255 models are particularly handy to account for complex platform
258 - What is the best applicative topology?
260 Is a flat master worker deployment sufficient? Should we go for a
261 hierarchical algorithm, with some forwarders taking large pools of
262 tasks from the master, each of them distributing their tasks to a
263 sub-pool of workers? Or should we introduce super-peers,
264 dupplicating the master's role in a peer-to-peer manner? Do the
265 algorithms require a perfect knowledge of the network?
267 - How is such an algorithm sensitive to external workload variation?
269 What if bandwidth, latency and computing speed can vary with no
270 warning? Shouldn't you study whether your algorithm is sensitive
271 to such load variations?
273 - Although an algorithm may be more efficient than another, how does
274 it interfere with unrelated applications executing on the same
277 **SimGrid was invented to answer such questions.** Do not believe the
278 fools saying that all you need to study such settings is a simple
279 discrete event simulator. Do you really want to reinvent the wheel,
280 debug and optimize your own tool, and validate its models against real
281 settings for ages, or do you prefer to sit on the shoulders of a
282 giant? With SimGrid, you can focus on your algorithm. The whole
283 simulation mechanism is already working.
285 Here is the visualization of a SimGrid simulation of two master worker
286 applications (one in light gray and the other in dark gray) running in
287 concurrence and showing resource usage over a long period of time. It
288 was obtained with the Triva software.
290 .. image:: /tuto_s4u/img/result.png
296 The easiest way to take the tutorial is to use the dedicated Docker
297 image. Once you `installed Docker itself
298 <https://docs.docker.com/install/>`_, simply do the following:
300 .. code-block:: shell
302 docker pull simgrid/tuto-s4u
303 docker run -it --rm --name simgrid --volume ~/simgrid-tutorial:/source/tutorial simgrid/tuto-s4u bash
305 This will start a new container with all you need to take this
306 tutorial, and create a ``simgrid-tutorial`` directory in your home on
307 your host machine that will be visible as ``/source/tutorial`` within the
308 container. You can then edit the files you want with your favorite
309 editor in ``~/simgrid-tutorial``, and compile them within the
310 container to enjoy the provided dependencies.
314 Any change to the container out of ``/source/tutorial`` will be lost
315 when you log out of the container, so don't edit the other files!
317 All needed dependencies are already installed in this container
318 (SimGrid, a C++ compiler, cmake, pajeng and R). Vite being only
319 optional in this tutorial, it is not installed to reduce the image
322 The code template is available under ``/source/simgrid-template-s4u.git``
323 in the image. You should copy it to your working directory and
324 recompile it when you first log in:
326 .. code-block:: shell
328 cp -r /source/simgrid-template-s4u.git/* /source/tutorial
333 Using your Computer Natively
334 ............................
336 To take the tutorial on your machine, you first need to :ref:`install
337 a recent version of SimGrid <install>`, a C++ compiler and also
338 ``pajeng`` to visualize the traces. You may want to install `Vite
339 <http://vite.gforge.inria.fr/>`_ to get a first glance at the traces.
340 The provided code template requires cmake to compile. On Debian and
341 Ubuntu for example, you can get them as follows:
343 .. code-block:: shell
345 sudo apt install simgrid pajeng cmake g++ vite
347 For R analysis of the produced traces, you may want to install R,
348 and the `pajengr <https://github.com/schnorr/pajengr#installation/>`_ package.
350 .. code-block:: shell
352 sudo apt install r-base r-cran-devtools cmake flex bison
353 Rscript -e "library(devtools); install_github('schnorr/pajengr');"
355 An initial version of the source code is provided on framagit. This
356 template compiles with cmake. If SimGrid is correctly installed, you
357 should be able to clone the `repository
358 <https://framagit.org/simgrid/simgrid-template-s4u>`_ and recompile
359 everything as follows:
361 .. code-block:: shell
363 # (exporting SimGrid_PATH is only needed if SimGrid is installed in a non-standard path)
364 export SimGrid_PATH=/where/to/simgrid
366 git clone https://framagit.org/simgrid/simgrid-template-s4u.git
367 cd simgrid-template-s4u/
371 If you struggle with the compilation, then you should double check
372 your :ref:`SimGrid installation <install>`. On need, please refer to
373 the :ref:`Troubleshooting your Project Setup
374 <install_yours_troubleshooting>` section.
376 Discovering the Provided Code
377 .............................
379 Please compile and execute the provided simulator as follows:
381 .. code-block:: shell
384 ./master-workers small_platform.xml master-workers_d.xml
386 For a more "fancy" output, you can use simgrid-colorizer.
388 .. code-block:: shell
390 ./master-workers small_platform.xml master-workers_d.xml 2>&1 | simgrid-colorizer
392 If you installed SimGrid to a non-standard path, you may have to
393 specify the full path to simgrid-colorizer on the above line, such as
394 ``/opt/simgrid/bin/simgrid-colorizer``. If you did not install it at all,
395 you can find it in <simgrid_root_directory>/bin/colorize.
397 For a classical Gantt-Chart vizualisation, you can use `Vite
398 <http://vite.gforge.inria.fr/>`_ if you have it installed, as
399 follows. But do not spend too much time installing Vite, because there
400 is a better way to visualize SimGrid traces (see below).
402 .. code-block:: shell
404 ./master-workers small_platform.xml master-workers_d.xml --cfg=tracing:yes --cfg=tracing/msg/process:yes
407 .. image:: /tuto_s4u/img/vite-screenshot.png
410 If you want the full power to visualize SimGrid traces, you need
411 to use R. As a start, you can download this `starter script
412 <https://framagit.org/simgrid/simgrid/raw/master/docs/source/tuto_s4u/draw_gantt.R>`_
413 and use it as follows:
415 .. code-block:: shell
417 ./master-workers small_platform.xml master-workers_d.xml --cfg=tracing:yes --cfg=tracing/msg/process:yes
418 Rscript draw_gantt.R simgrid.trace
420 It produces a ``Rplots.pdf`` with the following content:
422 .. image:: /tuto_s4u/img/Rscript-screenshot.png
426 Lab 1: Simpler Deployments
427 --------------------------
429 In the provided example, adding more workers quickly becomes a pain:
430 You need to start them (at the bottom of the file), and to inform the
431 master of its availability with an extra parameter. This is mandatory
432 if you want to inform the master of where the workers are running. But
433 actually, the master does not need to have this information.
435 We could leverage the mailbox mechanism flexibility, and use a sort of
436 yellow page system: Instead of sending data to the worker running on
437 Fafard, the master could send data to the third worker. Ie, instead of
438 using the worker location (which should be filled in two locations),
439 we could use their ID (which should be filled in one location
442 This could be done with the following deployment file. It's clearly
443 not shorter than the previous one, but it's still simpler because the
444 information is only written once. It thus follows the `DRY
445 <https://en.wikipedia.org/wiki/Don't_repeat_yourself>`_ `SPOT
446 <http://wiki.c2.com/?SinglePointOfTruth>`_ design principle.
448 .. literalinclude:: tuto_s4u/deployment1.xml
452 Copy your ``master-workers.cpp`` into ``master-workers-lab1.cpp`` and
453 add a new executable into ``CMakeLists.txt``. Then modify your worker
454 function so that it gets its mailbox name not from the name of its
455 host, but from the string passed as ``args[1]``. The master will send
456 messages to all workers based on their number, for example as follows:
460 for (int i = 0; i < tasks_count; i++) {
461 std::string worker_rank = std::to_string(i % workers_count);
462 std::string mailbox_name = std::string("worker-") + worker_rank;
463 simgrid::s4u::Mailbox* mailbox = simgrid::s4u::Mailbox::by_name(mailbox_name);
474 The mailboxes are a very powerful mechanism in SimGrid, allowing many
475 interesting application settings. They may feel surprising if you are
476 used to BSD sockets or other classical systems, but you will soon
477 appreciate their power. They are only used to match the
478 communications, but have no impact on the communication
479 timing. ``put()`` and ``get()`` are matched regardless of their
480 initiators' location and then the real communication occures between
481 the involved parties.
483 Please refer to the full `API of Mailboxes
484 <api/classsimgrid_1_1s4u_1_1Mailbox.html#class-documentation>`_ for
488 Lab 2: Using the Whole Platform
489 -------------------------------
491 It is now easier to add a new worker, but you still has to do it
492 manually. It would be much easier if the master could start the
493 workers on its own, one per available host in the platform. The new
494 deployment file should be as simple as:
496 .. literalinclude:: tuto_s4u/deployment2.xml
500 Creating the workers from the master
501 ....................................
503 For that, the master needs to retrieve the list of hosts declared in
504 the platform with :cpp:func:`simgrid::s4u::Engine::get_all_hosts`.
505 Then, the master should start the worker actors with
506 :cpp:func:`simgrid::s4u::Actor::create`.
508 ``Actor::create(name, host, func, params...)`` is a very flexible
509 function. Its third parameter is the function that the actor should
510 execute. This function can take any kind of parameter, provided that
511 you pass similar parameters to ``Actor::create()``. For example, you
512 could have something like this:
516 void my_actor(int param1, double param2, std::string param3) {
519 int main(int argc, char argv**) {
521 simgrid::s4u::ActorPtr actor;
522 actor = simgrid::s4u::Actor::create("name", simgrid::s4u::Host::by_name("the_host"),
523 &my_actor, 42, 3.14, "thevalue");
528 Master-Workers Communication
529 ............................
531 Previously, the workers got from their parameter the name of the
532 mailbox they should use. We can still do so: the master should build
533 such a parameter before using it in the ``Actor::create()`` call. The
534 master could even pass directly the mailbox as a parameter to the
537 Since we want later to study concurrent applications, it is advised to
538 use a mailbox name that is unique over the simulation even if there is
539 more than one master.
541 One possibility for that is to use the actor ID (aid) of each worker
542 as a mailbox name. The master can retrieve the aid of the newly
543 created actor with ``actor->get_pid()`` while the actor itself can
544 retrieve its own aid with ``simgrid::s4u::this_actor::get_pid()``.
545 The retrieved value is an ``aid_t``, which is an alias for ``long``.
547 Instead of having one mailbox per worker, you could also reorganize
548 completely your application to have only one mailbox per master. All
549 the workers of a given master would pull their work from the same
550 mailbox, which should be passed as parameter to the workers. This
551 reduces the amount of mailboxes, but prevents the master from taking
552 any scheduling decision. It really depends on how you want to organize
553 your application and what you want to study with your simulator. In
554 this tutorial, that's probably not a good idea.
559 In this exercise, we reduced the amount of configuration that our
560 simulator requests. This is both a good idea, and a dangerous
561 trend. This simplification is another application of the good old DRY/SPOT
562 programming principle (`Don't Repeat Yourself / Single Point Of Truth
563 <https://en.wikipedia.org/wiki/Don%27t_repeat_yourself>`_), and you
564 really want your programming artefacts to follow these software
565 engineering principles.
567 But at the same time, you should be careful in separating your
568 scientific contribution (the master/workers algorithm) and the
569 artefacts used to test it (platform, deployment and workload). This is
570 why SimGrid forces you to express your platform and deployment files
571 in XML instead of using a programming interface: it forces a clear
572 separation of concerns between things of very different nature.
574 Lab 3: Fixed Experiment Duration
575 --------------------------------
577 In the current version, the number of tasks is defined through the
578 worker arguments. Hence, tasks are created at the very beginning of
579 the simulation. Instead, have the master dispatching tasks for a
580 predetermined amount of time. The tasks must now be created on demand
581 instead of beforehand.
583 Of course, usual time functions like ``gettimeofday`` will give you the
584 time on your real machine, which is prety useless in the
585 simulation. Instead, retrieve the time in the simulated world with
586 :cpp:func:`simgrid::s4u::Engine::get_clock`.
588 You can still stop your workers with a specific task as previously,
589 or you may kill them forcefully with
590 :cpp:func:`simgrid::s4u::Actor::kill` (if you already have a reference
591 to the actor you want to kill) or
592 :cpp:func:`void simgrid::s4u::Actor::kill(aid_t)` (if you only have its ID).
595 Anyway, the new deployment `deployment3.xml` file should thus look
598 .. literalinclude:: tuto_s4u/deployment3.xml
601 Controlling the message verbosity
602 .................................
604 Not all messages are equally informative, so you probably want to
605 change some of the ``XBT_INFO`` into ``XBT_DEBUG`` so that they are
606 hidden by default. For example, you may want to use ``XBT_INFO`` once
607 every 100 tasks and ``XBT_DEBUG`` when sending all the other tasks. Or
608 you could show only the total number of tasks processed by
609 default. You can still see the debug messages as follows:
611 .. code-block:: shell
613 ./master-workers-lab3 small_platform.xml deployment3.xml --log=msg_test.thres:debug
616 Lab 4: Competing Applications
617 -----------------------------
619 It is now time to start several applications at once, with the following ``deployment4.xml`` file.
621 .. literalinclude:: tuto_s4u/deployment4.xml
624 Things happen when you do so, but it remains utterly difficult to
625 understand what's happening exactely. Even Gantt visualizations
626 contain too much information to be useful: it is impossible to
627 understand which task belong to which application. To fix this, we
628 will categorize the tasks.
630 Instead of starting the execution in one function call only with
631 ``this_actor::execute(cost)``, you need to
632 create the execution activity, set its tracing category, and then start
633 it and wait for its completion, as follows:
637 simgrid::s4u::ExecPtr exec = simgrid::s4u::this_actor::exec_init(compute_cost);
638 exec->set_tracing_category(category);
639 // exec->start() is optional here as wait() starts the activity on need
642 You can make the same code shorter as follows:
646 simgrid::s4u::this_actor::exec_init(compute_cost)->set_tracing_category(category)->wait();
648 Visualizing the result
649 .......................
651 vite is not enough to understand the situation, because it does not
652 deal with categorization. This time, you absolutely must switch to R,
653 as explained on `this page
654 <https://simgrid.org/contrib/R_visualization.html>`_.
658 Include here the minimal setting to view something in R.
661 Lab 5: Better Scheduling
662 ------------------------
664 You don't need a very advanced visualization solution to notice that
665 round-robin is completely suboptimal: most of the workers keep waiting
666 for more work. We will move to a First-Come First-Served mechanism
669 For that, your workers should explicitely request for work with a
670 message sent to a channel that is specific to their master. The name
671 of that private channel can be the one used to categorize the
672 executions, as it is already specific to each master.
674 The master should serve in a round-robin manner the requests it
675 receives, until the time is up. Changing the communication schema can
676 be a bit hairy, but once it works, you will see that such as simple
677 FCFS schema allows to double the amount of tasks handled over time
678 here. Things may be different with another platform file.
683 From this, many things can easily be added. For example, you could:
685 - Allow workers to have several pending requests so as to overlap
686 communication and computations as much as possible. Non-blocking
687 communication will probably become handy here.
688 - Add a performance measurement mechanism, enabling the master to make smart scheduling choices.
689 - Test your code on other platforms, from the ``examples/platforms``
690 directory in your archive.
692 What is the largest number of tasks requiring 50e6 flops and 1e5
693 bytes that you manage to distribute and process in one hour on
695 - Optimize not only for the amount of tasks handled, but also for the
696 total energy dissipated.
697 - And so on. If you come up with a really nice extension, please share
698 it with us so that we can extend this tutorial.
703 This tutorial is now terminated. You could keep reading the [online documentation][fn:4] or
704 [tutorials][fn:7], or you could head up to the example section to read some code.
708 Things to improve in the future:
710 - Propose equivalent exercises and skeleton in java (and Python once we have a python binding).
712 .. LocalWords: SimGrid