5 <img align=center src="simgrid_logo.png" alt="SimGrid"><br>
9 \section quick Quick start
11 SimGrid is a toolkit that provides core functionalities for the simulation
12 of distributed applications in heterogeneous distributed environments.
13 The specific goal of the project is to facilitate research in the area of
14 distributed and parallel application scheduling on distributed computing
15 platforms ranging from simple network of workstations to Computational
18 \subsection quick_dl Getting and installing the software
20 The SimGrid software package can be downloaded from
21 <a href="http://gcl.ucsd.edu/simgrid/dl/">here</a>.
23 To compile and install it, simply type the following. If you are not
24 familiar with compiling C files under UNIX and using libraries, please check
25 the \ref faq. SimGrid also works under Windows, but we do not distribute any
26 pre-compiled binaries [yet].
28 \verbatim $ ./configure
31 # make install\endverbatim
33 \subsection quick_more More information
35 The API is described <a href="API/html/modules.html">here</a> while
36 <a href="examples/html/modules.html">this page</a> presents some example of
39 For more information about the SimGrid toolkit, please simply keep reading
40 this page. It is organized as follow:
42 - \ref overview: Presentation of the toolkit, of each of its components
43 and of their interactions.
44 - \ref people: Who is behind this project.
45 - \ref publications: Some articles providing more details about the
46 SimGrid toolkit or using and validating it.
50 \section overview Overview of the toolkit components
52 As depicted by the following schema, the SimGrid toolkit is basically
59 \htmlinclude simgrid_modules.map
61 <br><b>Relationships between the SimGrid components</b>
65 \subsection overview_fondation Basement layer
67 The basement of the whole toolkit is constituted by the <b>\ref XBT_API
68 (eXtended Bundle of Tools)</b>.
70 It is a portable library providing some grounding features such as \ref
71 XBT_log, \ref XBT_error and \ref XBT_config. XBT also encompass the
72 following convenient datastructures: \ref XBT_dynar, \ref XBT_fifo, \ref
73 XBT_dict, \ref XBT_heap, \ref XBT_set and \ref XBT_swag.
75 See the \ref XBT_API section for more details.
77 \subsection overview_kernel Simulation kernel layer
79 The core functionnalities to simulate a virtual platform are provided by a
80 module called <b>\ref SURF_API</b> ("that's historical, my friend"). It is
81 very low-level and is not intended to be used as such by end-users. Instead,
82 it serve as a basis for the higher level layer.
84 SURF main features are a fast max-min linear solver and the ability to
85 change transparently the model used to describe the platform. This greatly
86 eases the comparison of the several models existing in the litterature.
88 See the \ref SURF_API section for more details.
90 \subsection overview_envs Programmation environments layer
92 This simulation kernel is used to build several programmation environments.
93 Each of them target a specific audiance and constitute a different paradigm.
94 To choose which of them you want to use, you have to think about what you
95 want to do and what would be the result of your work.
97 - If you want to study a theoritical problem and compare several
98 heuristics, you probably want to try <b>\ref MSG_API</b> (yet another
99 historical name). It was designed exactly to that extend and should allow
100 you to build easily rather realistic multi-agents simulation. Yet,
101 realism is not the main goal of this environment and the most annoying
102 technical issues of real platforms are masked here. Check the \ref
103 MSG_API section for more information.
105 - If you want to study the behaviour of a MPI application using emulation
106 technics, you should have a look at the <b>\ref SMPI_API</b> (Simulated
107 MPI) programming environment. Unfortunately, this work is still underway.
108 Check the \ref SMPI_API section for more information.
110 - If you want to develop a real distributed application, then you may find
111 <b>\ref GRAS_API</b> (Grid Reality And Simulation) useful. This is an API
112 for the realization of distributed applications.
114 Moreover, there is two implementations of this API: one on top of the
115 SURF (allowing to develop and test your application within the comfort of
116 the simulator) and another suited for deployment on real platforms
117 (allowing the resulting application to be highly portable and extremely
120 Even if you do not plan to run your code for real, you may want to switch
121 to GRAS if you intend to use MSG in a very intensive way (e.g. for
122 simulating a peer-to-peer environment).
124 See the \ref GRAS_API section for more details.
126 If your favorite programming environment/model is not there (BSP,
127 components, etc.) is not represented in the SimGrid toolkit yet, you may
128 consider adding it. You should contact us first, though.
130 Any question, remark or suggestion are welcome on the
131 <a href=https://listes.ens-lyon.fr/wws/info/simgrid2-users>SimGrid users
136 \section people People
138 The authors of SimGrid are:
140 - Henri Casanova <casanova#cs.ucsd.edu>
141 - Arnaud Legrand <arnaud.legrand#imag.fr>
142 - Martin Quinson <mquinson#debian.org>
144 \subsection contributers Contributers and alumni project members
146 - Loris Marchal: wrote the new algorithm for simulation TCP
148 - Julien Lerouge : wrote a XML parser for ENV descriptions and helped for
149 the general design during a 4 month period (march-june 2002)
151 - Clément Menier and Marc Perache : wrote a first prototype of the MSG
152 interface during a project at ENS-Lyon (jan 2002).
153 - Dmitrii Zagorodnov : wrote some parts of the first version of SimGrid
158 \section publications Selected publications
160 When citing SimGrid, the prefered reference paper is <i>Scheduling
161 Distributed Applications: the SimGrid Simulation Framework</i>, even if it's
162 a bit old now. We are actively working on improving this.
164 \subsection simulation About simulation
166 \li <b>Scheduling Distributed Applications: the
167 SimGrid Simulation Framework</b>\n
168 by <em>Henri Casanova and Arnaud Legrand and Loris Marchal</em>\n
169 Proceedings of the third IEEE International Symposium
170 on Cluster Computing and the Grid (CCGrid'03)\n
171 Since the advent of distributed computer systems an active field
172 of research has been the investigation of scheduling strategies
173 for parallel applications. The common approach is to employ
174 scheduling heuristics that approximate an optimal
175 schedule. Unfortunately, it is often impossible to obtain
176 analytical results to compare the efficacy of these heuristics.
177 One possibility is to conducts large numbers of back-to-back
178 experiments on real platforms. While this is possible on
179 tightly-coupled platforms, it is infeasible on modern distributed
180 platforms (i.e. Grids) as it is labor-intensive and does not
181 enable repeatable results. The solution is to resort to
182 simulations. Simulations not only enables repeatable results but
183 also make it possible to explore wide ranges of platform and
184 application scenarios.\n
185 In this paper we present the SimGrid framework which enables the
186 simulation of distributed applications in distributed computing
187 environments for the specific purpose of developing and evaluating
188 scheduling algorithms. This paper focuses on SimGrid v2, which
189 greatly improves on the first version of the software with more
190 realistic network models and topologies. SimGrid v2 also enables
191 the simulation of distributed scheduling agents, which has become
192 critical for current scheduling research in large-scale platforms.
193 After describing and validating these features, we present a case
194 study by which we demonstrate the usefulness of SimGrid for
195 conducting scheduling research.
198 \li <b>A Network Model for Simulation of Grid Application</b>\n
199 by <em>Henri Casanova and Loris Marchal</em>\n
201 In this work we investigate network models that can be
202 potentially employed in the simulation of scheduling algorithms for
203 distributed computing applications. We seek to develop a model of TCP
204 communication which is both high-level and realistic. Previous research
205 works show that accurate and global modeling of wide-area networks, such
206 as the Internet, faces a number of challenging issues. However, some
207 global models of fairness and bandwidth-sharing exist, and can be link
208 withthe behavior of TCP. Using both previous results and simulation (with
209 NS), we attempt to understand the macroscopic behavior of
210 TCP communications. We then propose a global model of the network for the
211 Grid platform. We perform partial validation of this model in
212 simulation. The model leads to an algorithm for computing
213 bandwidth-sharing. This algorithm can then be implemented as part of Grid
214 application simulations. We provide such an implementation for the
215 SimGrid simulation toolkit.\n
216 ftp://ftp.ens-lyon.fr/pub/LIP/Rapports/RR/RR2002/RR2002-40.ps.gz
219 \li <b>MetaSimGrid : Towards realistic scheduling simulation of
220 distributed applications</b>\n
221 by <em>Arnaud Legrand and Julien Lerouge</em>\n
222 Most scheduling problems are already hard on homogeneous
223 platforms, they become quite intractable in an heterogeneous
224 framework such as a metacomputing grid. In the best cases, a
225 guaranteed heuristic can be found, but most of the time, it is
226 not possible. Real experiments or simulations are often
227 involved to test or to compare heuristics. However, on a
228 distributed heterogeneous platform, such experiments are
229 technically difficult to drive, because of the genuine
230 instability of the platform. It is almost impossible to
231 guarantee that a platform which is not dedicated to the
232 experiment, will remain exactly the same between two tests,
233 thereby forbidding any meaningful comparison. Simulations are
234 then used to replace real experiments, so as to ensure the
235 reproducibility of measured data. A key issue is the
236 possibility to run the simulations against a realistic
237 environment. The main idea of trace-based simulation is to
238 record the platform parameters today, and to simulate the
239 algorithms tomorrow, against the recorded data: even though it
240 is not the current load of the platform, it is realistic,
241 because it represents a fair summary of what happened
242 previously. A good example of a trace-based simulation tool is
243 SimGrid, a toolkit providing a set of core abstractions and
244 functionalities that can be used to easily build simulators for
245 specific application domains and/or computing environment
246 topologies. Nevertheless, SimGrid lacks a number of convenient
247 features to craft simulations of a distributed application
248 where scheduling decisions are not taken by a single
249 process. Furthermore, modeling a complex platform by hand is
250 fastidious for a few hosts and is almost impossible for a real
251 grid. This report is a survey on simulation for scheduling
252 evaluation purposes and present MetaSimGrid, a simulator built
254 ftp://ftp.ens-lyon.fr/pub/LIP/Rapports/RR/RR2002/RR2002-28.ps.gz
256 \li <b>SimGrid: A Toolkit for the Simulation of Application
258 by <em>Henri Casanova</em>\n
259 Advances in hardware and software technologies have made it
260 possible to deploy parallel applications over increasingly large
261 sets of distributed resources. Consequently, the study of
262 scheduling algorithms for such applications has been an active area
263 of research. Given the nature of most scheduling problems one must
264 resort to simulation to effectively evaluate and compare their
265 efficacy over a wide range of scenarios. It has thus become
266 necessary to simulate those algorithms for increasingly complex
267 distributed, dynamic, heterogeneous environments. In this paper we
268 present SimGrid, a simulation toolkit for the study of scheduling
269 algorithms for distributed application. This paper gives the main
270 concepts and models behind SimGrid, describes its API and
271 highlights current implementation issues. We also give some
272 experimental results and describe work that builds on SimGrid's
274 http://grail.sdsc.edu/papers/simgrid_ccgrid01.ps.gz
276 \subsection research Papers using SimGrid results
278 \li <b>Optimal algorithms for scheduling divisible workloads on
279 heterogeneous systems</b>\n
280 by <em>Olivier Beaumont and Arnaud Legrand and Yves Robert</em>\n
281 In this paper, we discuss several algorithms for scheduling
282 divisible loads on heterogeneous systems. Our main contributions
283 are (i) new optimality results for single-round algorithms and (ii)
284 the design of an asymptotically optimal multi-round algorithm. This
285 multi-round algorithm automatically performs resource selection, a
286 difficult task that was previously left to the user. Because it is
287 periodic, it is simpler to implement, and more robust to changes in
288 the speeds of processors or communication links. On the theoretical
289 side, to the best of our knowledge, this is the first published
290 result assessing the absolute performance of a multi-round
291 algorithm. On the practical side, extensive simulations reveal
292 that our multi-round algorithm outperforms existing solutions on a
293 large variety of platforms, especially when the
294 communication-to-computation ratio is not very high (the difficult
296 ftp://ftp.ens-lyon.fr/pub/LIP/Rapports/RR/RR2002/RR2002-36.ps.gz
297 \li <b>On-line Parallel Tomography</b>\n
298 by <em>Shava Smallen</em>\n
299 Masters Thesis, UCSD, May 2001
300 \li <b>Applying Scheduling and Tuning to On-line Parallel Tomography </b>\n
301 by <em>Shava Smallen, Henri Casanova, Francine Berman</em>\n
302 in Proceedings of Supercomputing 2001
303 \li <b>Heuristics for Scheduling Parameter Sweep applications in
304 Grid environments</b>\n
305 by <em>Henri Casanova, Arnaud Legrand, Dmitrii Zagorodnov and
306 Francine Berman</em>\n
307 in Proceedings of the 9th Heterogeneous Computing workshop
308 (HCW'2000), pp349-363.