1 /*! @page introduction Introduction to SimGrid
3 [SimGrid](http://simgrid.gforge.inria.fr/) is a toolkit
4 that provides core functionalities for the simulation of distributed
5 applications in heterogeneous distributed environments.
7 The specific goal of the project is to facilitate research in the area of
8 distributed and parallel application scheduling on distributed computing
9 platforms ranging from simple network of workstations to Computational
13 The goal of this practical session is to illustrate various usage of
14 the MSG interface. To this end we will use the following simple setting:
16 > Assume we have a (possibly large) bunch of (possibly large) data to
17 > process and which originally reside on a server (a.k.a. master). For
18 > sake of simplicity, we assume all input file require the same amount
19 > of computation. We assume the server can be helped by a (possibly
20 > large) set of worker machines. What is the best way to organize the
23 Although this looks like a very simple setting it raises several
24 interesting questions:
26 - Which algorithm should the master use to send workload?
28 The most obvious algorithm would be to send tasks to workers in a
29 round-robin fashion. This is the initial code we provide you.
31 A less obvious but probably more efficient approach would be to set up
32 a request mechanism where a client first ask for tasks, which allows
33 the server to decide which request to answer and possibly to send
34 the tasks to the fastest machines. Maybe you can think of a
37 - How many tasks should the client ask for?
39 Indeed, if we set up a request mechanism so that workers only
40 send request whenever they have no more task to process, they are
41 likely to be poorly exploited since they will have to wait for the
42 master to consider their request and for the input data to be
43 transferred. A client should thus probably request a pool of tasks
44 but if it requests too many tasks, it is likely to lead to a poor
47 - How is the quality of such algorithm dependent on the platform
48 characteristics and on the task characteristics?
50 Whenever the input communication time is very small compared to
51 processing time and workers are homogeneous, it is likely that the
52 round-robin algorithm performs very well. Would it still hold true
53 when transfer time is not negligible and the platform is, say,
54 a volunteer computing system ?
56 - The network topology interconnecting the master and the workers
57 may be quite complicated. How does such a topology impact the
60 When data transfers are the bottleneck, it is likely that a good
61 modeling of the platform becomes essential. In this case, you may
62 want to be able to account for complex platform topologies.
64 - Do the algorithms depend on a perfect knowledge of this
67 Should we still use a flat master worker deployment or should we
70 - How is such an algorithm sensitive to external workload variation?
72 What if bandwidth, latency and power can vary with no warning?
73 Shouldn't you study whether your algorithm is sensitive to such
76 - Although an algorithm may be more efficient than another, how
77 does it interfere with other applications?
79 %As you can see, this very simple setting may need to evolve way
80 beyond what you initially imagined.
82 <blockquote> Premature optimization is the root of all evil. -- D.E.Knuth</blockquote>
84 Furthermore, writing your own simulator is much harder than you
85 may imagine. This is why you should rely on an established and flexible
88 The following figure is a screenshot of [triva][fn:1] visualizing a [SimGrid
89 simulation][fn:2] of two master worker applications (one in light gray and
90 the other in dark gray) running in concurrence and showing resource
91 usage over a long period of time.
93 ![Test](./sc3-description.png)
99 A lot of information on how to install and use Simgrid are
100 provided by the [online documentation][fn:4] and by several tutorials:
102 - http://simgrid.gforge.inria.fr/tutorials/simgrid-use-101.pdf
103 - http://simgrid.gforge.inria.fr/tutorials/simgrid-tracing-101.pdf
104 - http://simgrid.gforge.inria.fr/tutorials/simgrid-platf-101.pdf
106 ## Installing SimGrid
108 Please see our Section on \ref install "installing SimGrid".
111 Please note that your distribution may ship with an old version of
112 SimGrid; you may want to use [a newer release](https://gforge.inria.fr/frs/?group_id=12)
113 or even [clone our git repository](https://gforge.inria.fr/frs/?group_id=12)
114 (a [GitHub mirror](https://github.com/mquinson/simgrid) is also available).
120 This [software][fn:1] will be useful to make fancy graph or treemap
121 visualizations and get a better understanding of simulations. You
122 will first need to install pajeng:
125 sudo apt-get install git cmake build-essential libqt4-dev libboost-dev freeglut3-dev ;
126 git clone https://github.com/schnorr/pajeng.git
127 cd pajeng && mkdir -p build && cd build && cmake ../ -DCMAKE_INSTALL_PREFIX=$HOME && make -j install
131 Then you can install viva.
134 sudo apt-get install libboost-dev libconfig++-dev libconfig8-dev libgtk2.0-dev freeglut3-dev
135 git clone https://github.com/schnorr/viva.git
136 cd viva && mkdir -p build_graph && cd build_graph && cmake ../ -DTUPI_LIBRARY=ON -DVIVA=ON -DCMAKE_INSTALL_PREFIX=$HOME && make -j install
142 This [software][fn:5] provides a Gantt-chart visualization.
145 sudo apt-get install paje.app
150 This software provides a [Gantt-chart visualization][fn:6].
153 sudo apt-get install vite
159 ## Setting up and Compiling
161 The corresponding archive with all source files and platform files
162 can be obtained [here](http://simgrid.gforge.inria.fr/tutorials/msg-tuto/msg-tuto.tgz).
170 %As you can see, there is already a nice Makefile that compiles
171 everything for you. Now the tiny example has been compiled and it
172 can be easily run as follows:
175 ./masterworker0 platforms/platform.xml deployment0.xml 2>&1
178 If you create a single self-content C-file named foo.c, the
179 corresponding program will be simply compiled and linked with
186 For a more "fancy" output, you can try:
189 ./masterworker0 platforms/platform.xml deployment0.xml 2>&1 | simgrid-colorizer
192 For a really fancy output, you should use [viva/triva][fn:1]:
195 ./masterworker0 platforms/platform.xml deployment0.xml --cfg=tracing:yes \
196 --cfg=tracing/uncategorized:yes --cfg=viva/uncategorized:uncat.plist
197 LANG=C ; viva simgrid.trace uncat.plist
200 For a more classical Gantt-Chart visualization, you can produce a
204 ./masterworker0 platforms/platform.xml deployment0.xml --cfg=tracing:yes \
205 --cfg=tracing/msg/process:yes
206 LANG=C ; Paje simgrid.trace
209 Alternatively, you can use [vite][fn:6].
212 ./masterworker0 platforms/platform.xml deployment0.xml --cfg=tracing:yes \
213 --cfg=tracing/msg/process:yes --cfg=tracing/basic:yes
217 ## Getting Rid of Workers in the Deployment File
219 In the previous example, the deployment file `deployment0.xml`
220 is tightly connected to the platform file `platform.xml` and a
221 worker process is launched on each host:
224 <?xml version='1.0'?>
225 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
226 <platform version="3">
227 <!-- The master process (with some arguments) -->
228 <process host="Tremblay" function="master">
229 <argument value="20"/> <!-- Number of tasks -->
230 <argument value="50000000"/> <!-- Computation size of tasks -->
231 <argument value="1000000"/> <!-- Communication size of tasks -->
232 <argument value="Jupiter"/> <!-- First worker -->
233 <argument value="Fafard"/> <!-- Second worker -->
234 <argument value="Ginette"/> <!-- Third worker -->
235 <argument value="Bourassa"/> <!-- Last worker -->
236 <argument value="Tremblay"/> <!-- Me! I can work too! -->
238 <!-- The worker process (with no argument) -->
239 <process host="Tremblay" function="worker" on_failure="RESTART"/>
240 <process host="Jupiter" function="worker" on_failure="RESTART"/>
241 <process host="Fafard" function="worker" on_failure="RESTART"/>
242 <process host="Ginette" function="worker" on_failure="RESTART"/>
243 <process host="Bourassa" function="worker" on_failure="RESTART"/>
247 This is ok as the platform is rather small but will be painful when
248 using larger platforms. Instead, modify the simulator
249 `masterworker0.c` into `masterworker1.c` so that the master
250 launches a worker process on all the other machines at startup. The
251 new deployment file `deployment1.xml` should thus now simply be:
254 <?xml version='1.0'?>
255 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
256 <platform version="3">
257 <!-- The master process (with some arguments) -->
258 <process host="Tremblay" function="master">
259 <argument value="20"/> <!-- Number of tasks -->
260 <argument value="50000000"/> <!-- Computation size of tasks -->
261 <argument value="1000000"/> <!-- Communication size of tasks -->
266 To this end you may need the following MSG functions (click on the links
267 to see their descriptions):
270 int MSG_get_host_number(void);
271 xbt_dynar_t MSG_hosts_as_dynar(void);
272 void * xbt_dynar_to_array (xbt_dynar_t dynar);
273 msg_process_t MSG_process_create(const char *name, xbt_main_func_t code,
274 void *data, msg_host_t host);
278 It may avoid bugs later to avoid launching a worker on
279 the master host so you probably want to remove it from the host
282 The `data` field of the @ref MSG_process_create can be used to pass
283 a channel name that will be private between master
284 and workers (e.g., `master_name:worker_name`). Adding the
285 `master_name` in the channel name will allow to easily have several
286 masters and a worker per master on each machine. To this end, you
287 may need to use the following functions:
290 msg_host_t MSG_host_self(void);
291 const char * MSG_host_get_name(msg_host_t host);
292 msg_process_t MSG_process_self(void);
293 void * MSG_process_get_data(msg_process_t process);
296 If you are not too familiar with string
297 manipulation in C, you may want to use the following functions
298 (see the C reference for details):
301 char *strcpy(char *dest, const char *src);
302 char *strcat(char *dest, const char *src);
305 ## Setting up a Time Limit Mechanism
307 In the current version, the number of tasks is defined through the
308 worker arguments. Hence, tasks are created at the very beginning of
309 the simulation. Instead, create tasks as needed and provide a time
310 limit indicating when it stops sending tasks. To this end, you will
311 obviously need to know what time it is:
314 double MSG_get_clock(void);
317 Otherwise, a quite effective way of terminating the simulation
318 would be to use some of the following functions:
321 void MSG_process_kill(msg_process_t process);
322 int MSG_process_killall(int reset_PIDs);
325 Anyway, the new deployment `deployment2.xml` file should thus look
329 <?xml version='1.0'?>
330 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
331 <platform version="3">
332 <process host="Tremblay" function="master">
333 <argument value="3600"/> <!-- Simulation timeout -->
334 <argument value="50000000"/> <!-- Computation size of tasks -->
335 <argument value="1000000"/> <!-- Communication size of tasks -->
340 It may also be a good idea to transform most of the `XBT_INFO` into
341 `XBT_DEBUG` (e.g., keep the information on the total number of
342 tasks processed). These debug messages can be activated as follows:
345 ./masterworker2 platforms/platform.xml deployment2.xml --log=msg_test.thres:debug
348 ## Using the Tracing Mechanism
350 SimGrid can trace all resource consumption and the outcome can be
351 displayed with viva as illustrated in the section \ref intro_setup. However, when several
352 masters are deployed, it is hard to understand what happens.
355 <?xml version='1.0'?>
356 <!DOCTYPE platform SYSTEM "http://simgrid.gforge.inria.fr/simgrid.dtd">
357 <platform version="3">
358 <process host="Tremblay" function="master">
359 <argument value="3600"/> <!-- Simulation timeout -->
360 <argument value="50000000"/> <!-- Computation size of tasks -->
361 <argument value="10"/> <!-- Communication size of tasks -->
363 <process host="Fafard" function="master">
364 <argument value="3600"/> <!-- Simulation timeout -->
365 <argument value="50000000"/> <!-- Computation size of tasks -->
366 <argument value="10"/> <!-- Communication size of tasks -->
368 <process host="Jupiter" function="master">
369 <argument value="3600"/> <!-- Simulation timeout -->
370 <argument value="50000000"/> <!-- Computation size of tasks -->
371 <argument value="10"/> <!-- Communication size of tasks -->
376 So let's use categories to track more precisely who does what and when:
379 void TRACE_category(const char *category);
380 void MSG_task_set_category (msg_task_t task, const char *category);
383 The outcome can then be visualized as follows:
386 ./masterworker3 platforms/platform.xml deployment3.xml --cfg=tracing:yes\
387 --cfg=tracing/categorized:yes --cfg=viva/categorized:viva_cat.plist
388 LANG=C; viva simgrid.trace viva_cat.plist
391 Right now, you should realize that nothing is behaving like you
392 expect. Most workers are idle even though input data are ridiculous
393 and there are several masters deployed on the platform. Using a
394 Gantt-chart visualization may help:
397 ./masterworker3 platforms/platform.xml deployment3.xml --cfg=tracing:yes \
398 --cfg=tracing/msg/process:yes
399 LANG=C; Paje simgrid.trace
402 OK, so it should now be obvious that round robin is actually
405 ## Improving the Scheduling
407 Instead of a round-robin scheduling, let's implement a first-come
408 first-served mechanism. To this end, workers need to send a tiny
409 request first. A possible way to implement such a request with MSG
410 is to send on a specific channel (e.g., the name of the master
411 name) a task with payload 0 and whose attached data is the worker
412 name. This way, the master can keep track of which workers are idle
415 To know whether it has pending requests, the master can use the
416 following [function][fn:7]:
419 int MSG_task_listen(const char *alias);
422 If so, it should get the request and push the corresponding host
423 into a dynar so that they can later be retrieved when sending a
427 xbt_dynar_t xbt_dynar_new(const unsigned long elm_size,
428 void_f_pvoid_t const free_f);
429 void xbt_dynar_push(xbt_dynar_t const dynar, const void *src);
430 void xbt_dynar_shift(xbt_dynar_t const dynar, void *const dst);
431 unsigned long xbt_dynar_length(const xbt_dynar_t dynar);
434 %As you will soon realize, with such simple mechanisms, simple
435 deadlocks will soon appear. They can easily be removed with a
436 simple polling mechanism, hence the need for the following
440 msg_error_t MSG_process_sleep(double nb_sec);
443 %As you should quickly realize, on the simple previous example, it
444 will double the throughput of the platform but will be quite
445 ineffective when input size of the tasks is not negligible anymore.
447 From this, many things can easily be added. For example, you could:
448 - add a performance measurement mechanism;
449 - enable the master to make smart scheduling choices using
450 measurement information;
451 - allow workers to have several pending requests so as to overlap
452 communication and computations as much as possible;
455 ## Using More Elaborate Platforms
457 SimGrid offers a rather powerful platform modeling mechanism. The
458 `src/examples/platforms/` repository comprises a variety of platforms ranging
459 from simple to elaborate. Associated to a good
460 visualization tool to ensure your simulation is meaningful, they
461 can allow you to study to which extent your algorithm scales...
463 What is the largest number of tasks requiring 50e6 flops and 1e5
464 bytes that you manage to distribute and process in one hour on
465 `g5k.xml` (you should use `deployment_general.xml`)?
467 # TODO: Points to improve for the next time
469 - Propose equivalent exercises and skeleton in java.
470 - Propose a virtualbox image with everything (simgrid, paje, viva,
472 - Ease the installation on mac OS X (binary installer) and
474 - Explain that programming in C or java and having a working
475 development environment is a prerequisite.
477 [fn:1]: http://triva.gforge.inria.fr/index.html
478 [fn:2]: http://hal.inria.fr/inria-00529569
479 [fn:3]: http://hal.inria.fr/hal-00738321
480 [fn:4]: http://simgrid.gforge.inria.fr/documentation.html
481 [fn:5]: http://paje.sourceforge.net/
482 [fn:6]: http://vite.gforge.inria.fr/