6 SimGrid was conceived as a tool to study distributed algorithms. Its
7 :ref:`S4U interface <S4U_doc>` makes it easy to assess Cloud,
8 P2P, HPC, IoT, and other similar settings (:ref:`more info <index>`).
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|_, |Disks|_). 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 serves as a 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 .. |Disks| replace:: **Disks**
37 .. _Disks: app_s4u.html#s4u-disk
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 the necessary
63 configuration to make it work. 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++ or Python,
66 which you are supposed to know beforehand.
69 Discover the Master/Workers
70 ---------------------------
72 This section introduces an example of SimGrid simulation. This
73 simple application is composed of two kinds 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 The provided code dispatches these tasks in `round-robin scheduling <https://en.wikipedia.org/wiki/Round-robin_scheduling>`_,
81 i.e. in circular order: tasks are dispatched to each worker one after the other, until all tasks are dispatched.
82 You will improve this scheme later in this tutorial.
87 Let's start with the code of the master. It is represented by the
88 *master* function below. This simple function takes at least 3
89 parameters (the number of tasks to dispatch, their computational size
90 in flops to compute, and their communication size in bytes to
91 exchange). Every parameter after the third one must be the name of a
92 host on which a worker is waiting for something to compute.
94 Then, the tasks are sent one after the other, each on a mailbox named
95 after the worker's hosts. On the other side, a given worker (which
96 code is given below) waits for incoming tasks on its mailbox.
99 In the end, once all tasks are dispatched, the master dispatches
100 another task per worker, but this time with a negative amount of flops
101 to compute. Indeed, this application decided by convention, that the
102 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` (to retrieve or create a mailbox) and
110 :cpp:func:`simgrid::s4u::Mailbox::put` (so send something over a mailbox). Also, :c:macro:`XBT_INFO` is used
111 as a replacement to ``printf()`` or ``std::cout`` to ensure that the messages
112 are nicely logged along with the simulated time and actor name.
114 .. literalinclude:: ../../examples/cpp/app-masterworkers/s4u-app-masterworkers-fun.cpp
116 :start-after: master-begin
117 :end-before: master-end
119 .. group-tab:: Python
121 At the end of the day, the only SimGrid specific functions used in
122 this example are :py:func:`simgrid.Mailbox.by_name` (to retrieve or create a mailbox) and
123 :py:func:`simgrid.Mailbox.put` (so send something over a mailbox). Also, :py:func:`simgrid.this_actor.info` is used
124 as a replacement to `print` to ensure that the messages
125 are nicely logged along with the simulated time and actor name.
127 .. literalinclude:: ../../examples/python/app-masterworkers/app-masterworkers.py
129 :start-after: master-begin
130 :end-before: master-end
132 Then comes the code of the worker actors. This function expects no
133 parameter from its vector of strings. Its code is very simple: it
134 expects messages on the mailbox that is named after its host. As long as it gets valid
135 computation requests (whose compute_amount is positive), it computes
136 this task and waits for the next one.
142 The worker retrieves its own host with
143 :cpp:func:`simgrid::s4u::this_actor::get_host`. The
144 :ref:`simgrid::s4u::this_actor <API_s4u_this_actor>`
145 namespace contains many such helping functions.
147 .. literalinclude:: ../../examples/cpp/app-masterworkers/s4u-app-masterworkers-fun.cpp
149 :start-after: worker-begin
150 :end-before: worker-end
152 .. group-tab:: Python
154 The worker retrieves its own host with :py:func:`simgrid.this_actor.get_host`. The
155 :ref:`this_actor <API_s4u_this_actor>` object contains many such helping functions.
157 .. literalinclude:: ../../examples/python/app-masterworkers/app-masterworkers.py
159 :start-after: worker-begin
160 :end-before: worker-end
162 Starting the Simulation
163 .......................
165 And this is it. In only a few lines, we defined the algorithm of our master/workers example.
171 That being said, an algorithm alone is not enough to define a
172 simulation: SimGrid is a library, not a program. So you need to define
173 your own ``main()`` function as follows. This function is in charge of
174 creating a SimGrid simulation engine (on line 3), register the actor
175 functions to the engine (on lines 7 and 8), load the simulated platform
176 from its description file (on line 11), map actors onto that platform
177 (on line 12) and run the simulation until its completion on line 15.
179 .. literalinclude:: ../../examples/cpp/app-masterworkers/s4u-app-masterworkers-fun.cpp
181 :start-after: main-begin
182 :end-before: main-end
185 .. group-tab:: Python
187 That being said, an algorithm alone is not enough to define a simulation:
188 you need a main block to setup the simulation and its components as follows.
189 This code creates a SimGrid simulation engine (on line 4), registers the actor
190 functions to the engine (on lines 7 and 8), loads the simulated platform
191 from its description file (on line 11), map actors onto that platform
192 (on line 12) and run the simulation until its completion on line 15.
194 .. literalinclude:: ../../examples/python/app-masterworkers/app-masterworkers.py
196 :start-after: main-begin
197 :end-before: main-end
200 Finally, this example requires a platform file and a deployment file.
205 Platform files define the simulated platform on which the provided
206 application will take place. It contains one or several **Network
207 Zone** |api_s4u_NetZone|_ that contains both |Host|_ and |Link|_
208 Resources, as well as routing information.
210 Such files can get rather long and boring, so the example below is
211 only an excerpt of the full ``examples/platforms/small_platform.xml``
212 file. For example, most routing information is missing, and only the
213 route between the hosts Tremblay and Fafard is given. This path
214 traverses 6 links (named 4, 3, 2, 0, 1, and 8). There are several
215 examples of platforms in the archive under ``examples/platforms``.
217 .. |api_s4u_NetZone| image:: /img/extlink.png
220 .. _api_s4u_NetZone: app_s4u.html#s4u-netzone
222 .. |api_s4u_Link| image:: /img/extlink.png
225 .. _api_s4u_Link: app_s4u.html#s4u-link
227 .. literalinclude:: ../../examples/platforms/small_platform.xml
229 :lines: 1-10,12-20,56-62,192-
230 :caption: (excerpts of the small_platform.xml file)
235 Deployment files specify the execution scenario: it lists the actors
236 that should be started, along with their parameters. In the following
237 example, we start 6 actors: one master and 5 workers.
239 .. literalinclude:: ../../examples/cpp/app-masterworkers/s4u-app-masterworkers_d.xml
245 This time, we have all parts: once the program is compiled, we can execute it as follows.
251 Note how the :c:macro:`XBT_INFO` requests turned into informative messages.
253 .. "WARNING: non-whitespace stripped by dedent" is expected here as we remove the $ marker this way
255 .. literalinclude:: ../../examples/cpp/app-masterworkers/s4u-app-masterworkers.tesh
257 :start-after: s4u-app-masterworkers-fun
258 :prepend: $$$ ./masterworkers platform.xml deploy.xml
262 .. group-tab:: Python
264 Note how the :py:func:`simgrid.this_actor.info` calls turned into informative messages.
266 .. literalinclude:: ../../examples/python/app-masterworkers/app-masterworkers.tesh
268 :start-after: app-masterworkers_d.xml
269 :prepend: $$$ python ./app-masterworkers.py platform.xml deploy.xml
273 Each example included in the SimGrid distribution comes with a `tesh`
274 file that presents how to start the example once compiled, along with
275 the expected output. These files are used for the automatic testing of
276 the framework but can be used to see the examples' output without
277 compiling them. See e.g. the file
278 `examples/cpp/app-masterworkers/s4u-app-masterworkers.tesh <https://framagit.org/simgrid/simgrid/-/blob/master/examples/cpp/app-masterworkers/s4u-app-masterworkers.tesh>`_.
279 Lines starting with `$` are the commands to execute;
280 lines starting with `>` are the expected output of each command, while
281 lines starting with `!` are configuration items for the test runner.
287 In this section, you will modify the example presented earlier to
288 explore the quality of the proposed algorithm. It already works, and
289 the simulation prints things, but the truth is that we have no idea of
290 whether this is a good algorithm to dispatch tasks to the workers.
291 This very simple setting raises many interesting questions:
293 .. image:: /tuto_s4u/img/question.svg
296 - Which algorithm should the master use? Or should the worker decide
299 Round Robin is not an efficient algorithm when all tasks are not
300 processed at the same speed. It would probably be more efficient
301 if the workers were asking for tasks when ready.
303 - Should tasks be grouped in batches or sent separately?
305 The workers will starve if they don't get the tasks fast
306 enough. One possibility to reduce latency would be to send tasks
307 in pools instead of one by one. But if the pools are too big, the
308 load balancing will likely get uneven, in particular when
309 distributing the last tasks.
311 - How does the quality of such an algorithm dependent on the platform
312 characteristics and on the task characteristics?
314 Whenever the input communication time is very small compared to
315 processing time and workers are homogeneous, it is likely that the
316 round-robin algorithm performs very well. Would it still hold true
317 when transfer time is not negligible? What if some tasks are
318 performed faster on some specific nodes?
320 - The network topology interconnecting the master and the workers
321 may be quite complicated. How does such a topology impact the
324 When data transfers are the bottleneck, it is likely that good
325 modeling of the platform becomes essential. The SimGrid platform
326 models are particularly handy to account for complex platform
329 - What is the best applicative topology?
331 Is a flat master-worker deployment sufficient? Should we go for a
332 hierarchical algorithm, with some forwarders taking large pools of
333 tasks from the master, each of them distributing their tasks to a
334 sub-pool of workers? Or should we introduce super-peers,
335 duplicating the master's role in a peer-to-peer manner? Do the
336 algorithms require a perfect knowledge of the network?
338 - How is such an algorithm sensitive to external workload variation?
340 What if bandwidth, latency, and computing speed can vary with no
341 warning? Shouldn't you study whether your algorithm is sensitive
342 to such load variations?
344 - Although an algorithm may be more efficient than another, how does
345 it interfere with unrelated applications executing on the same
348 **SimGrid was invented to answer such questions.** Do not believe the
349 fools saying that all you need to study such settings is a simple
350 discrete event simulator. Do you really want to reinvent the wheel,
351 debug and optimize your own tool, and validate its models against real
352 settings for ages, or do you prefer to sit on the shoulders of a
353 giant? With SimGrid, you can focus on your algorithm. The whole
354 simulation mechanism is already working.
356 Here is the visualization of a SimGrid simulation of two master-worker
357 applications (one in light gray and the other in dark gray) running in
358 concurrence and showing resource usage over a long period of time. It
359 was obtained with the Triva software.
361 .. image:: /tuto_s4u/img/result.png
367 The easiest way to take the tutorial is to use the dedicated Docker image.
368 Once you `installed Docker itself <https://docs.docker.com/install/>`_, simply do the following:
370 .. code-block:: console
372 $ docker pull simgrid/tuto-s4u
373 $ mkdir ~/simgrid-tutorial
374 $ docker run --user $UID:$GID -it --rm --name simgrid --volume ~/simgrid-tutorial:/source/tutorial simgrid/tuto-s4u bash
376 This will start a new container with all you need to take this
377 tutorial, and create a ``simgrid-tutorial`` directory in your home on
378 your host machine that will be visible as ``/source/tutorial`` within the
379 container. You can then edit the files you want with your favorite
380 editor in ``~/simgrid-tutorial``, and compile them within the
381 container to enjoy the provided dependencies.
385 Any change to the container out of ``/source/tutorial`` will be lost
386 when you log out of the container, so don't edit the other files!
388 All needed dependencies are already installed in this container
389 (SimGrid, a C++ compiler, CMake, pajeng, and R). Vite being only
390 optional in this tutorial, it is not installed to reduce the image
393 The docker does not run as root, so that the files can easily be exchanged between within the container and the outer world.
394 If you need to run a command as root within the container, simply type the following in another terminal to join the same container as root:
396 .. code-block:: console
398 $ docker container ls
399 # This lists all containers running on your machine. For example:
400 # CONTAINER ID IMAGE COMMAND CREATED STATUS PORTS NAMES
401 # 7e921b1b18a7 simgrid/stable "bash" 7 minutes ago Up 7 minutes adoring_shamir
403 $ docker exec --user 0:0 -it {container_name} bash
404 # In the previous example, container_name was "adoring_shamir"
406 The code template is available under ``/source/simgrid-template-s4u.git``
407 in the image. You should copy it to your working directory and
408 recompile it when you first log in:
410 .. code-block:: console
412 $ cp -r /source/simgrid-template-s4u.git/* /source/tutorial
413 $ cd /source/tutorial
417 Using your Computer Natively
418 ............................
424 To take the tutorial on your machine, you first need to :ref:`install
425 a recent version of SimGrid <install>`, a C++ compiler, and also
426 ``pajeng`` to visualize the traces. You may want to install `Vite
427 <http://vite.gforge.inria.fr/>`_ to get a first glance at the traces.
428 The provided code template requires CMake to compile. On Debian and
429 Ubuntu for example, you can get them as follows:
431 .. code-block:: console
433 $ sudo apt install simgrid pajeng cmake g++ vite
435 An initial version of the source code is provided on framagit. This
436 template compiles with CMake. If SimGrid is correctly installed, you
437 should be able to clone the `repository
438 <https://framagit.org/simgrid/simgrid-template-s4u>`_ and recompile
439 everything as follows:
441 .. code-block:: console
443 # (exporting SimGrid_PATH is only needed if SimGrid is installed in a non-standard path)
444 $ export SimGrid_PATH=/where/to/simgrid
446 $ git clone https://framagit.org/simgrid/simgrid-template-s4u.git
447 $ cd simgrid-template-s4u/
451 If you struggle with the compilation, then you should double-check
452 your :ref:`SimGrid installation <install>`. On need, please refer to
453 the :ref:`Troubleshooting your Project Setup
454 <install_yours_troubleshooting>` section.
456 .. group-tab:: Python
458 To take the tutorial on your machine, you first need to :ref:`install
459 a recent version of SimGrid <install>` and ``pajeng`` to visualize the
460 traces. You may want to install `Vite <http://vite.gforge.inria.fr/>`_ to get a first glance at the traces.
461 On Debian and Ubuntu for example, you can get them as follows:
463 .. code-block:: console
465 $ sudo apt install simgrid pajeng vite
467 An initial version of the source code is provided on framagit.
468 If SimGrid is correctly installed, you should be able to clone the `repository
469 <https://framagit.org/simgrid/simgrid-template-s4u>`_ and execute it as follows:
471 .. code-block:: console
473 $ git clone https://framagit.org/simgrid/simgrid-template-s4u.git
474 $ cd simgrid-template-s4u/
475 $ python master-workers.py small_platform.xml master-workers_d.xml
477 If you get some errors, then you should double-check
478 your :ref:`SimGrid installation <install>`. On need, please refer to
479 the :ref:`Troubleshooting your Project Setup <install_yours_troubleshooting>` section.
482 For R analysis of the produced traces, you may want to install R
483 and the `pajengr <https://github.com/schnorr/pajengr#installation/>`_ package.
485 .. code-block:: console
487 # install R and necessary packages
488 $ sudo apt install r-base r-cran-devtools r-cran-tidyverse
489 # install pajengr dependencies
490 $ sudo apt install git cmake flex bison
491 # install the pajengr R package
492 $ Rscript -e "library(devtools); install_github('schnorr/pajengr');"
495 Discovering the Provided Code
496 .............................
502 Please compile and execute the provided simulator as follows:
504 .. code-block:: console
506 $ make master-workers
507 $ ./master-workers small_platform.xml master-workers_d.xml
509 .. group-tab:: Python
511 Please execute the provided simulator as follows:
513 .. code-block:: console
515 $ python master-workers.py small_platform.xml master-workers_d.xml
517 For a more "fancy" output, you can use simgrid-colorizer.
519 .. code-block:: console
522 $ ./master-workers small_platform.xml master-workers_d.xml 2>&1 | simgrid-colorizer
525 $ python master-workers.py small_platform.xml master-workers_d.xml 2>&1 | simgrid-colorizer
527 If you installed SimGrid to a non-standard path, you may have to
528 specify the full path to simgrid-colorizer on the above line, such as
529 ``/opt/simgrid/bin/simgrid-colorizer``. If you did not install it at all,
530 you can find it in <simgrid_root_directory>/bin/colorize.
532 For a classical Gantt-Chart visualization, you can use `Vite
533 <http://vite.gforge.inria.fr/>`_ if you have it installed, as
534 follows. But do not spend too much time installing Vite, because there
535 is a better way to visualize SimGrid traces (see below).
537 .. code-block:: console
540 $ ./master-workers small_platform.xml master-workers_d.xml --cfg=tracing:yes --cfg=tracing/actor:yes
542 $ python master-workers.py small_platform.xml master-workers_d.xml --cfg=tracing:yes --cfg=tracing/actor:yes
544 # Visualize the produced trace
547 .. image:: /tuto_s4u/img/vite-screenshot.png
552 If you use an older version of SimGrid (before v3.26), you should use
553 ``--cfg=tracing/msg/process:yes`` instead of ``--cfg=tracing/actor:yes``.
555 If you want the full power to visualize SimGrid traces, you need
556 to use R. As a start, you can download this `starter script
557 <https://framagit.org/simgrid/simgrid/raw/master/docs/source/tuto_s4u/draw_gantt.R>`_
558 and use it as follows:
560 .. code-block:: console
563 $ ./master-workers small_platform.xml master-workers_d.xml --cfg=tracing:yes --cfg=tracing/actor:yes
565 $ python master-workers.py small_platform.xml master-workers_d.xml --cfg=tracing:yes --cfg=tracing/actor:yes
567 # Visualize the produced trace
568 $ Rscript draw_gantt.R simgrid.trace
570 It produces a ``Rplots.pdf`` with the following content:
572 .. image:: /tuto_s4u/img/Rscript-screenshot.png
576 Lab 1: Simpler deployments
577 --------------------------
579 .. rst-class:: compact-list
583 * Get your hands on the code and change the communication pattern
584 * Discover the Mailbox mechanism
586 In the provided example, adding more workers quickly becomes a pain:
587 You need to start them (at the bottom of the file) and inform the
588 master of its availability with an extra parameter. This is mandatory
589 if you want to inform the master of where the workers are running. But
590 actually, the master does not need to have this information.
592 We could leverage the mailbox mechanism flexibility, and use a sort of
593 yellow page system: Instead of sending data to the worker running on
594 Fafard, the master could send data to the third worker. Ie, instead of
595 using the worker location (which should be filled in two locations),
596 we could use their ID (which should be filled in one location
599 This could be done with the following deployment file. It's
600 not shorter than the previous one, but it's still simpler because the
601 information is only written once. It thus follows the `DRY
602 <https://en.wikipedia.org/wiki/Don't_repeat_yourself>`_ `SPOT
603 <http://wiki.c2.com/?SinglePointOfTruth>`_ design principle.
605 .. literalinclude:: tuto_s4u/deployment1.xml
612 Copy your ``master-workers.cpp`` into ``master-workers-lab1.cpp`` and
613 add a new executable into ``CMakeLists.txt``. Then modify your worker
614 function so that it gets its mailbox name not from the name of its
615 host, but from the string passed as ``args[1]``. The master will send
616 messages to all workers based on their number, for example as follows:
620 for (int i = 0; i < tasks_count; i++) {
621 std::string worker_rank = std::to_string(i % workers_count);
622 std::string mailbox_name = std::string("worker-") + worker_rank;
623 simgrid::s4u::Mailbox* mailbox = simgrid::s4u::Mailbox::by_name(mailbox_name);
630 .. group-tab:: Python
632 Copy your ``master-workers.py`` into ``master-workers-lab1.py`` then
634 function so that it gets its mailbox name not from the name of its
635 host, but from the string passed as ``args[0]``. The master will send
636 messages to all workers based on their number, for example as follows:
640 for i in range(tasks_count):
641 mailbox = Mailbox.by_name(str(i % worker_count))
647 The mailboxes are a very powerful mechanism in SimGrid, allowing many
648 interesting application settings. They may feel unusual if you are
649 used to BSD sockets or other classical systems, but you will soon
650 appreciate their power. They are only used to match
651 communications but have no impact on the communication
652 timing. ``put()`` and ``get()`` are matched regardless of their
653 initiators' location and then the real communication occurs between
654 the involved parties.
656 Please refer to the full `Mailboxes' documentation <app_s4u.html#s4u-mailbox>`_
660 Lab 2: Using the Whole Platform
661 -------------------------------
663 .. rst-class:: compact-list
667 * Interact with the platform (get the list of all hosts)
668 * Create actors directly from your program instead of the deployment file
670 It is now easier to add a new worker, but you still have to do it
671 manually. It would be much easier if the master could start the
672 workers on its own, one per available host in the platform. The new
673 deployment file should be as simple as:
675 .. literalinclude:: tuto_s4u/deployment2.xml
679 Creating the workers from the master
680 ....................................
686 For that, the master needs to retrieve the list of hosts declared in
687 the platform with :cpp:func:`simgrid::s4u::Engine::get_all_hosts`.
688 Then, the master should start the worker actors with
689 :cpp:func:`simgrid::s4u::Actor::create`.
691 ``Actor::create(name, host, func, params...)`` is a very flexible
692 function. Its third parameter is the function that the actor should
693 execute. This function can take any kind of parameter, provided that
694 you pass similar parameters to ``Actor::create()``. For example, you
695 could have something like this:
699 void my_actor(int param1, double param2, std::string param3) {
702 int main(int argc, char argv**) {
704 simgrid::s4u::ActorPtr actor;
705 actor = simgrid::s4u::Actor::create("name", simgrid::s4u::Host::by_name("the_host"),
706 &my_actor, 42, 3.14, "thevalue");
710 .. group-tab:: Python
712 For that, the master needs to retrieve the list of hosts declared in
713 the platform with :py:func:`simgrid.Engine.get_all_hosts`. Since this method is not static,
714 you may want to call it on the Engine instance, as in ``Engine.instance().get_all_hosts()``.
715 Then, the master should start the worker actors with :py:func:`simgrid.Actor.create`.
717 ``Actor.create(name, host, func, params...)`` is a very flexible
718 function. Its third parameter is the function that the actor should
719 execute. This function can take any kind of parameter, provided that
720 you pass similar parameters to ``Actor.create()``. For example, you
721 could have something like this:
723 .. code-block:: python
725 def my_actor(param1, param2, param3):
726 # your code comes here
727 actor = simgrid.Actor.create("name", the_host, my_actor, 42, 3.14, "thevalue")
730 Master-Workers Communication
731 ............................
733 Previously, the workers got from their parameter the name of the
734 mailbox they should use. We can still do so: the master should build
735 such a parameter before using it in the ``Actor::create()`` call. The
736 master could even pass directly the mailbox as a parameter to the
739 Since we want later to study concurrent applications, it is advised to
740 use a mailbox name that is unique over the simulation even if there is
741 more than one master.
747 One possibility for that is to use the actor ID (aid) of each worker
748 as a mailbox name. The master can retrieve the aid of the newly
749 created actor with :cpp:func:`simgrid::s4u::Actor::get_pid()` while the actor itself can
750 retrieve its own aid with :cpp:func:`simgrid::s4u::this_actor::get_pid()`.
751 The retrieved value is an :cpp:type:`aid_t`, which is an alias for ``long``.
753 .. group-tab:: Python
755 One possibility for that is to use the actor ID of each worker
756 as a mailbox name. The master can retrieve the aid of the newly
757 created actor with :py:func:`simgrid.Actor.pid` while the actor itself can
758 retrieve its own aid with :py:func:`simgrid.this_actor.get_pid()`.
763 In this exercise, we reduced the amount of configuration that our
764 simulator requests. This is both a good idea and a dangerous
765 trend. This simplification is another application of the good old DRY/SPOT
766 programming principle (`Don't Repeat Yourself / Single Point Of Truth
767 <https://en.wikipedia.org/wiki/Don%27t_repeat_yourself>`_), and you
768 really want your programming artifacts to follow these software
769 engineering principles.
771 But at the same time, you should be careful in separating your
772 scientific contribution (the master/workers algorithm) and the
773 artifacts used to test it (platform, deployment, and workload). This is
774 why SimGrid forces you to express your platform and deployment files
775 in XML instead of using a programming interface: it forces a clear
776 separation of concerns between things of different nature.
778 Lab 3: Fixed Experiment Duration
779 --------------------------------
781 .. rst-class:: compact-list
785 * Forcefully kill actors, and stop the simulation at a given point of time
786 * Control the logging verbosity
788 In the current version, the number of tasks is defined through the
789 worker arguments. Hence, tasks are created at the very beginning of
790 the simulation. Instead, have the master dispatching tasks for a
791 predetermined amount of time. The tasks must now be created on need
792 instead of beforehand.
794 Of course, usual time functions like ``gettimeofday`` will give you the
795 time on your real machine, which is pretty useless in the
796 simulation. Instead, retrieve the time in the simulated world with
797 :cpp:func:`simgrid::s4u::Engine::get_clock` (C++) or
798 :py:func:`simgrid.Engine.get_clock()`) (Python).
800 You can still stop your workers with a specific task as previously,
801 or you may kill them forcefully with :cpp:func:`simgrid::s4u::Actor::kill` (C++)
802 :py:func:`simgrid.Actor.kill` (Python).
804 Anyway, the new deployment `deployment3.xml` file should thus look
807 .. literalinclude:: tuto_s4u/deployment3.xml
810 Controlling the message verbosity
811 .................................
813 Not all messages are equally informative, so you probably want to
814 change some of the *info* messages (C: :c:macro:`XBT_INFO`; Python: :py:func:`simgrid.this_actor.info`)
815 into *debug* messages`(C: :c:macro:`XBT_DEBUG`; Python: :py:func:`simgrid.this_actor.debug`) so that they are
816 hidden by default. For example, you may want to use an *info* message once
817 every 100 tasks and *debug* when sending all the other tasks. Or
818 you could show only the total number of tasks processed by
819 default. You can still see the debug messages as follows:
821 .. code-block:: console
823 $ ./master-workers-lab3 small_platform.xml deployment3.xml --log=s4u_app_masterworker.thres:debug
825 Lab 4: What-if analysis
826 -----------------------
828 .. rst-class:: compact-list
832 * Change the platform characteristics during the simulation.
833 * Explore other communication patterns.
838 Attach a profile to your hosts, so that their computational speed automatically vary over time, modeling an external load on these machines.
839 This can be done with :cpp:func:`simgrid::s4u::Host::set_speed_profile` (C++) or :py:func:`simgrid.Host.set_speed_profile` (Python).
841 Make it so that one of the hosts get really really slow, and observe how your whole application performance decreases.
842 This is because one slow host slows down the whole process. Instead of a round-robin dispatch push,
843 you should completely reorganize your application in a First-Come First-Served manner (FCFS).
844 Actors should pull a task whenever they are ready, so that fast actors can overpass slow ones in the queue.
846 There is two ways to implement that: either the workers request a task to the master by sending their name to a specific mailbox,
847 or the master directly pushes the tasks to a centralized mailbox from which the workers pull their work. The first approach is closer
848 to what would happen with communications based on BSD sockets while the second is closer to message queues. You could also decide to
849 model your socket application in the second manner if you want to neglect these details and keep your simulator simple. It's your decision.
851 Changing the communication schema can be a bit hairy, but once it works, you will see that such as simple FCFS schema allows one to greatly
852 increase the amount of tasks handled over time here. Things may be different with another platform file.
857 Let's now modify the communication speed between hosts.
859 Retrieve a link from its name with :cpp:func:`simgrid::s4u::Link::by_name()` (C++) or :py:func:`simgrid.Link.by_name()` (python).
861 Retrieve all links in the platform with :cpp:func:`simgrid::s4u::Engine::get_all_links()` (C++) or :py:func:`simgrid.Engine.get_all_links()` (python).
863 Retrieve the list of links from one host to another with :cpp:func:`simgrid::s4u::Host::route_to` (C++) or :py:func:`simgrid.Host.route_to` (python).
865 Modify the bandwidth of a given link with :cpp:func:`simgrid::s4u::Link::set_bandwidth` (C++) or :py:func:`simgrid.Link.set_bandwidth` (python).
866 You can even have the bandwidth automatically vary over time with :cpp:func:`simgrid::s4u::Link::set_bandwidth_profile` (C++) or :py:func:`simgrid.Link.set_bandwidth_profile` (python).
868 Once implemented, you will notice that slow communications may still result in situations
869 where one worker only works at a given point of time. To overcome that, your master needs
870 to send data to several workers in parallel, using
871 :cpp:func:`simgrid::s4u::Mailbox::put_async` (C++) or :py:func:`simgrid.Mailbox.put_async` (Python)
872 to start several communications in parallel, and
873 :cpp:func:`simgrid::s4u::Comm::wait_any` (C++) or and :py:func:`simgrid.Comm.wait_any` (Python)
874 to react to the completion of one of these communications. Actually, since this code somewhat tricky
875 to write, it's provided as :ref:`an example <s4u_ex_communication>` in the distribution (search for
876 ``wait_any`` in that page).
878 Dealing with failures
879 .....................
881 Turn a given link off with :cpp:func:`simgrid::s4u::Link::turn_off` (C++) or :py:func:`simgrid.Link.turn_off` (python).
882 You can even implement churn where a link automatically turn off and on again over time with :cpp:func:`simgrid::s4u::Link::set_state_profile` (C++) or :py:func:`simgrid.Link.set_state_profile` (python).
884 If a link fails while you try to use it, ``wait()`` will raise a ``NetworkFailureException`` that you need to catch.
885 Again, there is a nice example demoing this feature, :ref:`under platform-failures <s4u_ex_communication>`.
887 Lab 5: Competing Applications
888 -----------------------------
890 .. rst-class:: compact-list
894 * Advanced vizualization through tracing categories
897 It is now time to start several applications at once, with the following ``deployment5.xml`` file.
899 .. literalinclude:: tuto_s4u/deployment5.xml
902 Things happen when you do so, but it remains utterly difficult to
903 understand what's happening exactly. Even Gantt visualizations
904 contain too much information to be useful: it is impossible to
905 understand which task belongs to which application. To fix this, we
906 will categorize the tasks.
908 Instead of starting the execution in one function call only with
909 ``this_actor::execute(cost)``, you need to
910 create the execution activity, set its tracing category, start it
911 and wait for its completion, as follows.
917 Use :cpp:func:`simgrid::s4u::Exec::set_tracing_category` to change the category of an execution.
921 simgrid::s4u::ExecPtr exec = simgrid::s4u::this_actor::exec_init(compute_cost);
922 exec->set_tracing_category(category);
923 // exec->start() is optional here as wait() starts the activity on need
926 You can shorten this code as follows:
930 simgrid::s4u::this_actor::exec_init(compute_cost)->set_tracing_category(category)->wait();
932 .. group-tab:: Python
934 Use :py:func:`simgrid.Exec.set_tracing_category` to change the category of an execution.
936 .. code-block:: python
938 exec = simgrid:.this_actor.exec_init(compute_cost)
939 exec.set_tracing_category(category)
940 # exec.start() is optional here as wait() starts the activity on need
943 You can shorten this code as follows:
945 .. code-block:: python
947 simgrid.this_actor.exec_init(compute_cost).set_tracing_category(category).wait()
950 Visualizing the result
951 .......................
953 vite is not enough to understand the situation, because it does not
954 deal with categorization. This time, you absolutely must switch to R,
955 as explained on `this page
956 <https://simgrid.org/contrib/R_visualization.html>`_.
960 Include here the minimal setting to view something in R.
965 From this, many things can easily be added. For example, you could:
967 - Allow workers to have several pending requests to overlap
968 communication and computations as much as possible. Non-blocking
969 communication will probably become handy here.
970 - Add a performance measurement mechanism, enabling the master to make smart scheduling choices.
971 - Test your code on other platforms, from the ``examples/platforms``
972 directory in your archive.
974 What is the largest number of tasks requiring 50e6 flops and 1e5
975 bytes that you manage to distribute and process in one hour on
977 - Optimize not only for the number of tasks handled but also for the total energy dissipated.
978 - And so on. If you come up with a nice extension, please share
979 it with us so that we can extend this tutorial.
984 This tutorial is now terminated. You could keep reading the online documentation and
985 tutorials, or you could head up to the :ref:`example section <s4u_examples>` to read some code.
989 Things to improve in the future:
991 - Propose equivalent exercises and skeleton in Java once we fix the Java binding.
997 .. LocalWords: SimGrid