1 \documentclass[conference]{IEEEtran}
3 \usepackage[T1]{fontenc}
4 \usepackage[utf8]{inputenc}
5 \usepackage{amsfonts,amssymb}
7 %\usepackage{algorithm}
8 \usepackage{algpseudocode}
11 \usepackage[american]{babel}
12 % Extension pour les liens intra-documents (tagged PDF)
13 % et l'affichage correct des URL (commande \url{http://example.com})
14 %\usepackage{hyperref}
17 \DeclareUrlCommand\email{\urlstyle{same}}
19 \usepackage[autolanguage,np]{numprint}
21 \renewcommand*\npunitcommand[1]{\text{#1}}
22 \npthousandthpartsep{}}
25 \usepackage[textsize=footnotesize]{todonotes}
26 \newcommand{\AG}[2][inline]{%
27 \todo[color=green!50,#1]{\sffamily\textbf{AG:} #2}\xspace}
28 \newcommand{\DL}[2][inline]{%
29 \todo[color=yellow!50,#1]{\sffamily\textbf{DL:} #2}\xspace}
30 \newcommand{\LZK}[2][inline]{%
31 \todo[color=blue!10,#1]{\sffamily\textbf{LZK:} #2}\xspace}
32 \newcommand{\RC}[2][inline]{%
33 \todo[color=red!10,#1]{\sffamily\textbf{RC:} #2}\xspace}
35 \algnewcommand\algorithmicinput{\textbf{Input:}}
36 \algnewcommand\Input{\item[\algorithmicinput]}
38 \algnewcommand\algorithmicoutput{\textbf{Output:}}
39 \algnewcommand\Output{\item[\algorithmicoutput]}
41 \newcommand{\MI}{\mathit{MaxIter}}
45 \title{Simulation of Asynchronous Iterative Numerical Algorithms Using SimGrid}
49 Charles Emile Ramamonjisoa\IEEEauthorrefmark{1},
50 David Laiymani\IEEEauthorrefmark{1},
51 Arnaud Giersch\IEEEauthorrefmark{1},
52 Lilia Ziane Khodja\IEEEauthorrefmark{2} and
53 Raphaël Couturier\IEEEauthorrefmark{1}
55 \IEEEauthorblockA{\IEEEauthorrefmark{1}%
56 Femto-ST Institute -- DISC Department\\
57 Université de Franche-Comté,
58 IUT de Belfort-Montbéliard\\
59 19 avenue du Maréchal Juin, BP 527, 90016 Belfort cedex, France\\
60 Email: \email{{charles.ramamonjisoa,david.laiymani,arnaud.giersch,raphael.couturier}@univ-fcomte.fr}
62 \IEEEauthorblockA{\IEEEauthorrefmark{2}%
63 Inria Bordeaux Sud-Ouest\\
64 200 avenue de la Vieille Tour, 33405 Talence cedex, France \\
65 Email: \email{lilia.ziane@inria.fr}
71 \RC{Ordre des autheurs pas définitif.}
73 In recent years, the scalability of large-scale implementation in a
74 distributed environment of algorithms becoming more and more complex has
75 always been hampered by the limits of physical computing resources
76 capacity. One solution is to run the program in a virtual environment
77 simulating a real interconnected computers architecture. The results are
78 convincing and useful solutions are obtained with far fewer resources
79 than in a real platform. However, challenges remain for the convergence
80 and efficiency of a class of algorithms that concern us here, namely
81 numerical parallel iterative algorithms executed in asynchronous mode,
82 especially in a large scale level. Actually, such algorithm requires a
83 balance and a compromise between computation and communication time
84 during the execution. Two important factors determine the success of the
85 experimentation: the convergence of the iterative algorithm on a large
86 scale and the execution time reduction in asynchronous mode. Once again,
87 from the current work, a simulated environment like SimGrid provides
88 accurate results which are difficult or even impossible to obtain in a
89 physical platform by exploiting the flexibility of the simulator on the
90 computing units clusters and the network structure design. Our
91 experimental outputs showed a saving of up to \np[\%]{40} for the algorithm
92 execution time in asynchronous mode compared to the synchronous one with
93 a residual precision up to \np{E-11}. Such successful results open
94 perspectives on experimentations for running the algorithm on a
95 simulated large scale growing environment and with larger problem size.
97 % no keywords for IEEE conferences
98 % Keywords: Algorithm distributed iterative asynchronous simulation SimGrid
101 \section{Introduction}
103 Parallel computing and high performance computing (HPC) are becoming more and more imperative for solving various
104 problems raised by researchers on various scientific disciplines but also by industrial in the field. Indeed, the
105 increasing complexity of these requested applications combined with a continuous increase of their sizes lead to write
106 distributed and parallel algorithms requiring significant hardware resources (grid computing, clusters, broadband
107 network, etc.) but also a non-negligible CPU execution time. We consider in this paper a class of highly efficient
108 parallel algorithms called \texttt{numerical iterative algorithms} executed in a distributed environment. As their name
109 suggests, these algorithm solves a given problem by successive iterations ($X_{n +1} = f(X_{n})$) from an initial value
110 $X_{0}$ to find an approximate value $X^*$ of the solution with a very low residual error. Several well-known methods
111 demonstrate the convergence of these algorithms \cite{}.
113 Parallelization of such algorithms generally involved the division of the problem into several \emph{pieces} that will
114 be solved in parallel on multiple processing units. The latter will communicate each intermediate results before a new
115 iteration starts until the approximate solution is reached. These parallel computations can be performed either in
116 \emph{synchronous} communication mode where a new iteration begin only when all nodes communications are completed,
117 either \emph{asynchronous} mode where processors can continue independently without or few synchronization points. For
118 instance in the \textit{Asynchronous Iterations - Asynchronous Communications (AIAC)} model \cite{bcvc06:ij}, local
119 computations do not need to wait for required data. Processors can then perform their iterations with the data present
120 at that time. Even if the number of iterations required before the convergence is generally greater than for the
121 synchronous case, AIAC algorithms can significantly reduce overall execution times by suppressing idle times due to
122 synchronizations especially in a grid computing context (see \cite{bcvc06:ij} for more details).
124 Parallel numerical applications (synchronous or asynchronous) may have different configuration and deployment
125 requirements. Quantifying their performance of resource allocation policies and application scheduling algorithms in
126 grid computing environments under varying load, CPU power and network speeds is very costly, labor intensive and time
127 consuming \cite{BuRaCa}. The case of AIAC algorithms is even more problematic since they are very sensible to the
128 execution environment context. For instance, variations in the network bandwith (intra and inter- clusters), in the
129 number and the power of nodes, in the number of clusters... can lead to very different number of iterations and so to
130 very different execution times. In this context, it appears that the use of simulation tools to explore various platform
131 scenarios and to run enormous numbers of experiments quickly can be very promising. In this way, the use of a simulation
132 environment to execute parallel iterative algorithms found some interests in reducing the highly cost of access to
133 computing resources: (1) for the applications development life cycle and in code debugging (2) and in production to get
134 results in a reasonable execution time with a simulated infrastructure not accessible with physical resources. Indeed,
135 the launch of distributed iterative asynchronous algorithms to solve a given problem on a large-scale simulated
136 environment challenges to find optimal configurations giving the best results with a lowest residual error and in the
137 best of execution time.
139 To our knowledge, there is no existing work on the large-scale simulation of a real AIAC application. The aim of this
140 paper is to give a first approach of the simulation of AIAC algorithms using the SimGrid toolkit \cite{SimGrid}. We had
141 in the scope of this work implemented a program for solving large non-symmetric linear system of equations by numerical
142 method GMRES (Generalized Minimal Residual). SimGrid had allowed us to launch the application from a modest computing
143 infrastructure by simulating different distributed architectures composed by clusters nodes interconnected by variable
144 speed networks. The simulated results we obtained are in line with real results exposed in ?? In addition, it has been
145 permitted to show the effectiveness of asynchronous mode algorithm by comparing its performance with the synchronous
146 mode time. With selected parameters on the network platforms (bandwidth, latency of inter cluster network) and on the
147 clusters architecture (number, capacity calculation power) in the simulated environment, the experimental results have
148 demonstrated not only the algorithm convergence within a reasonable time compared with the physical environment
149 performance, but also a time saving of up to \np[\%]{40} in asynchronous mode.
151 This article is structured as follows: after this introduction, the next section will give a brief description of
152 iterative asynchronous model. Then, the simulation framework SimGrid is presented with the settings to create various
153 distributed architectures. The algorithm of the multi-splitting method used by GMRES written with MPI primitives and
154 its adaptation to SimGrid with SMPI (Simulated MPI) is detailed in the next section. At last, the experiments results
155 carried out will be presented before some concluding remarks and future works.
157 \section{Motivations and scientific context}
159 As exposed in the introduction, parallel iterative methods are now widely used in many scientific domains. They can be
160 classified in three main classes depending on how iterations and communications are managed (for more details readers
161 can refer to \cite{bcvc02:ip}). In the \textit{Synchronous Iterations - Synchronous Communications (SISC)} model data
162 are exchanged at the end of each iteration. All the processors must begin the same iteration at the same time and
163 important idle times on processors are generated. The \textit{Synchronous Iterations - Asynchronous Communications
164 (SIAC)} model can be compared to the previous one except that data required on another processor are sent asynchronously
165 i.e. without stopping current computations. This technique allows to partially overlap communications by computations
166 but unfortunately, the overlapping is only partial and important idle times remain. It is clear that, in a grid
167 computing context, where the number of computational nodes is large, heterogeneous and widely distributed, the idle
168 times generated by synchronizations are very penalizing. One way to overcome this problem is to use the
169 \textit{Asynchronous Iterations - Asynchronous Communications (AIAC)} model. Here, local computations do not need to
170 wait for required data. Processors can then perform their iterations with the data present at that time. Figure
171 \ref{fig:aiac} illustrates this model where the grey blocks represent the computation phases, the white spaces the idle
172 times and the arrows the communications. With this algorithmic model, the number of iterations required before the
173 convergence is generally greater than for the two former classes. But, and as detailed in \cite{bcvc06:ij}, AIAC
174 algorithms can significantly reduce overall execution times by suppressing idle times due to synchronizations especially
175 in a grid computing context.
179 \includegraphics[width=8cm]{AIAC.pdf}
180 \caption{The Asynchronous Iterations - Asynchronous Communications model }
185 It is very challenging to develop efficient applications for large scale, heterogeneous and distributed platforms such
186 as computing grids. Researchers and engineers have to develop techniques for maximizing application performance of these
187 multi-cluster platforms, by redesigning the applications and/or by using novel algorithms that can account for the
188 composite and heterogeneous nature of the platform. Unfortunately, the deployment of such applications on these very
189 large scale systems is very costly, labor intensive and time consuming. In this context, it appears that the use of
190 simulation tools to explore various platform scenarios at will and to run enormous numbers of experiments quickly can be
191 very promising. Several works...
193 In the context of AIAC algorithms, the use of simulation tools is even more relevant. Indeed, this class of applications
194 is very sensible to the execution environment context. For instance, variations in the network bandwith (intra and
195 inter-clusters), in the number and the power of nodes, in the number of clusters... can lead to very different number of
196 iterations and so to very different execution times.
203 SimGrid~\cite{casanova+legrand+quinson.2008.simgrid,SimGrid} is a simulation
204 framework to sudy the behavior of large-scale distributed systems. As its name
205 says, it emanates from the grid computing community, but is nowadays used to
206 study grids, clouds, HPC or peer-to-peer systems.
207 %- open source, developped since 1999, one of the major solution in the field
209 SimGrid provides several programming interfaces: MSG to simulate Concurrent
210 Sequential Processes, SimDAG to simulate DAGs of (parallel) tasks, and SMPI to
211 run real applications written in MPI~\cite{MPI}. Apart from the native C
212 interface, SimGrid provides bindings for the C++, Java, Lua and Ruby programming
213 languages. The SMPI interface supports applications written in C or Fortran,
214 with little or no modifications.
215 %- implements most of MPI-2 \cite{ref} standard [CHECK]
217 %%% explain simulation
218 %- simulated processes folded in one real process
219 %- simulates interactions on the network, fluid model
220 %- able to skip long-lasting computations
224 %- describe resources and their interconnection, with their properties
227 %%% validation + refs
229 \AG{Décrire SimGrid~\cite{casanova+legrand+quinson.2008.simgrid,SimGrid} (Arnaud)}
231 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
232 \section{Simulation of the multisplitting method}
233 %Décrire le problème (algo) traité ainsi que le processus d'adaptation à SimGrid.
234 Let $Ax=b$ be a large sparse system of $n$ linear equations in $\mathbb{R}$, where $A$ is a sparse square and nonsingular matrix, $x$ is the solution vector and $b$ is the right-hand side vector. We use a multisplitting method based on the block Jacobi splitting to solve this linear system on a large scale platform composed of $L$ clusters of processors. In this case, we apply a row-by-row splitting without overlapping
236 \left(\begin{array}{ccc}
237 A_{11} & \cdots & A_{1L} \\
238 \vdots & \ddots & \vdots\\
239 A_{L1} & \cdots & A_{LL}
242 \left(\begin{array}{c}
248 \left(\begin{array}{c}
252 \end{array} \right)\]
253 in such a way that successive rows of matrix $A$ and both vectors $x$ and $b$ are assigned to one cluster, where for all $l,m\in\{1,\ldots,L\}$ $A_{lm}$ is a rectangular block of $A$ of size $n_l\times n_m$, $X_l$ and $B_l$ are sub-vectors of $x$ and $b$, respectively, of size $n_l$ each and $\sum_{l} n_l=\sum_{m} n_m=n$.
255 The multisplitting method proceeds by iteration to solve in parallel the linear system on $L$ clusters of processors, in such a way each sub-system
259 A_{ll}X_l = Y_l \mbox{,~such that}\\
260 Y_l = B_l - \displaystyle\sum_{\substack{m=1\\ m\neq l}}^{L}A_{lm}X_m
265 is solved independently by a cluster and communications are required to update the right-hand side sub-vector $Y_l$, such that the sub-vectors $X_m$ represent the data dependencies between the clusters. As each sub-system (\ref{eq:4.1}) is solved in parallel by a cluster of processors, our multisplitting method uses an iterative method as an inner solver which is easier to parallelize and more scalable than a direct method. In this work, we use the parallel algorithm of GMRES method~\cite{ref1} which is one of the most used iterative method by many researchers.
268 %%% IEEE instructions forbid to use an algorithm environment here, use figure
270 \begin{algorithmic}[1]
271 \Input $A_l$ (sparse sub-matrix), $B_l$ (right-hand side sub-vector)
272 \Output $X_l$ (solution sub-vector)\vspace{0.2cm}
273 \State Load $A_l$, $B_l$
274 \State Set the initial guess $x^0$
275 \For {$k=0,1,2,\ldots$ until the global convergence}
276 \State Restart outer iteration with $x^0=x^k$
277 \State Inner iteration: \Call{InnerSolver}{$x^0$, $k+1$}
278 \State Send shared elements of $X_l^{k+1}$ to neighboring clusters
279 \State Receive shared elements in $\{X_m^{k+1}\}_{m\neq l}$
284 \Function {InnerSolver}{$x^0$, $k$}
285 \State Compute local right-hand side $Y_l$: \[Y_l = B_l - \sum\nolimits^L_{\substack{m=1 \\m\neq l}}A_{lm}X_m^0\]
286 \State Solving sub-system $A_{ll}X_l^k=Y_l$ with the parallel GMRES method
287 \State \Return $X_l^k$
290 \caption{A multisplitting solver with GMRES method}
294 Algorithm on Figure~\ref{algo:01} shows the main key points of the
295 multisplitting method to solve a large sparse linear system. This algorithm is
296 based on an outer-inner iteration method where the parallel synchronous GMRES
297 method is used to solve the inner iteration. It is executed in parallel by each
298 cluster of processors. For all $l,m\in\{1,\ldots,L\}$, the matrices and vectors
299 with the subscript $l$ represent the local data for cluster $l$, while
300 $\{A_{lm}\}_{m\neq l}$ are off-diagonal matrices of sparse matrix $A$ and
301 $\{X_m\}_{m\neq l}$ contain vector elements of solution $x$ shared with
302 neighboring clusters. At every outer iteration $k$, asynchronous communications
303 are performed between processors of the local cluster and those of distant
304 clusters (lines $6$ and $7$ in Figure~\ref{algo:01}). The shared vector
305 elements of the solution $x$ are exchanged by message passing using MPI
306 non-blocking communication routines.
310 \includegraphics[width=60mm,keepaspectratio]{clustering}
311 \caption{Example of three clusters of processors interconnected by a virtual unidirectional ring network.}
315 The global convergence of the asynchronous multisplitting solver is detected when the clusters of processors have all converged locally. We implemented the global convergence detection process as follows. On each cluster a master processor is designated (for example the processor with rank $1$) and masters of all clusters are interconnected by a virtual unidirectional ring network (see Figure~\ref{fig:4.1}). During the resolution, a Boolean token circulates around the virtual ring from a master processor to another until the global convergence is achieved. So starting from the cluster with rank $1$, each master processor $i$ sets the token to {\it True} if the local convergence is achieved or to {\it False} otherwise, and sends it to master processor $i+1$. Finally, the global convergence is detected when the master of cluster $1$ receives from the master of cluster $L$ a token set to {\it True}. In this case, the master of cluster $1$ broadcasts a stop message to masters of other clusters. In this work, the local convergence on each cluster $l$ is detected when the following condition is satisfied
316 \[(k\leq \MI) \mbox{~or~} (\|X_l^k - X_l^{k+1}\|_{\infty}\leq\epsilon)\]
317 where $\MI$ is the maximum number of outer iterations and $\epsilon$ is the tolerance threshold of the error computed between two successive local solution $X_l^k$ and $X_l^{k+1}$.
319 \LZK{Description du processus d'adaptation de l'algo multisplitting à SimGrid}
320 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
329 \section{Experimental results}
331 When the \emph{real} application runs in the simulation environment and produces
332 the expected results, varying the input parameters and the program arguments
333 allows us to compare outputs from the code execution. We have noticed from this
334 study that the results depend on the following parameters: (1) at the network
335 level, we found that the most critical values are the bandwidth (bw) and the
336 network latency (lat). (2) Hosts power (GFlops) can also influence on the
337 results. And finally, (3) when submitting job batches for execution, the
338 arguments values passed to the program like the maximum number of iterations or
339 the \emph{external} precision are critical to ensure not only the convergence of the
340 algorithm but also to get the main objective of the experimentation of the
341 simulation in having an execution time in asynchronous less than in synchronous
342 mode, in others words, in having a \emph{speedup} less than 1
343 ({speedup}${}={}${execution time in synchronous mode}${}/{}${execution time in
346 A priori, obtaining a speedup less than 1 would be difficult in a local area
347 network configuration where the synchronous mode will take advantage on the rapid
348 exchange of information on such high-speed links. Thus, the methodology adopted
349 was to launch the application on clustered network. In this last configuration,
350 degrading the inter-cluster network performance will \emph{penalize} the synchronous
351 mode allowing to get a speedup lower than 1. This action simulates the case of
352 clusters linked with long distance network like Internet.
354 As a first step, the algorithm was run on a network consisting of two clusters
355 containing fifty hosts each, totaling one hundred hosts. Various combinations of
356 the above factors have providing the results shown in
357 Table~\ref{tab.cluster.2x50} with a matrix size ranging from $N_x = N_y = N_z =
358 62 \text{ to } 171$ elements or from $62^{3} = \np{238328}$ to $171^{3} =
359 \np{5211000}$ entries.
363 \caption{2 clusters, each with 50 nodes}
364 \label{tab.cluster.2x50}
365 \renewcommand{\arraystretch}{1.3}
367 \begin{tabular}{|>{\bfseries}r|*{12}{c|}}
370 & 5 & 5 & 5 & 5 & 5 & 50 \\
373 & 0.02 & 0.02 & 0.02 & 0.02 & 0.02 & 0.02 \\
376 & 1 & 1 & 1 & 1.5 & 1.5 & 1.5 \\
379 & 62 & 62 & 62 & 100 & 100 & 110 \\
382 & \np{E-5} & \np{E-8} & \np{E-9} & \np{E-11} & \np{E-11} & \np{E-11} \\
385 & 0.396 & 0.392 & 0.396 & 0.391 & 0.393 & 0.395 \\
391 \begin{tabular}{|>{\bfseries}r|*{12}{c|}}
394 & 50 & 50 & 50 & 50 & 10 & 10 \\
397 & 0.02 & 0.02 & 0.02 & 0.02 & 0.03 & 0.01 \\
400 & 1.5 & 1.5 & 1.5 & 1.5 & 1 & 1.5 \\
403 & 120 & 130 & 140 & 150 & 171 & 171 \\
406 & \np{E-11} & \np{E-11} & \np{E-11} & \np{E-11} & \np{E-5} & \np{E-5} \\
409 & 0.398 & 0.388 & 0.393 & 0.394 & 0.63 & 0.778 \\
414 Then we have changed the network configuration using three clusters containing
415 respectively 33, 33 and 34 hosts, or again by on hundred hosts for all the
416 clusters. In the same way as above, a judicious choice of key parameters has
417 permitted to get the results in Table~\ref{tab.cluster.3x33} which shows the
418 speedups less than 1 with a matrix size from 62 to 100 elements.
422 \caption{3 clusters, each with 33 nodes}
423 \label{tab.cluster.3x33}
424 \renewcommand{\arraystretch}{1.3}
426 \begin{tabular}{|>{\bfseries}r|*{6}{c|}}
429 & 10 & 5 & 4 & 3 & 2 & 6 \\
432 & 0.01 & 0.02 & 0.02 & 0.02 & 0.02 & 0.02 \\
435 & 1 & 1 & 1 & 1 & 1 & 1 \\
438 & 62 & 100 & 100 & 100 & 100 & 171 \\
441 & \np{E-5} & \np{E-5} & \np{E-5} & \np{E-5} & \np{E-5} & \np{E-5} \\
444 & 0.997 & 0.99 & 0.93 & 0.84 & 0.78 & 0.99 \\
450 In a final step, results of an execution attempt to scale up the three clustered
451 configuration but increasing by two hundreds hosts has been recorded in
452 Table~\ref{tab.cluster.3x67}.
456 \caption{3 clusters, each with 66 nodes}
457 \label{tab.cluster.3x67}
458 \renewcommand{\arraystretch}{1.3}
460 \begin{tabular}{|>{\bfseries}r|c|}
470 Prec/Eprec & \np{E-5} \\
477 Note that the program was run with the following parameters:
479 \paragraph*{SMPI parameters}
482 \item HOSTFILE: Hosts file description.
483 \item PLATFORM: file description of the platform architecture : clusters (CPU power,
484 \dots{}), intra cluster network description, inter cluster network (bandwidth bw,
485 lat latency, \dots{}).
489 \paragraph*{Arguments of the program}
492 \item Description of the cluster architecture;
493 \item Maximum number of internal and external iterations;
494 \item Internal and external precisions;
495 \item Matrix size $N_x$, $N_y$ and $N_z$;
496 \item Matrix diagonal value: \np{6.0};
497 \item Execution Mode: synchronous or asynchronous.
500 \paragraph*{Interpretations and comments}
502 After analyzing the outputs, generally, for the configuration with two or three
503 clusters including one hundred hosts (Tables~\ref{tab.cluster.2x50}
504 and~\ref{tab.cluster.3x33}), some combinations of the used parameters affecting
505 the results have given a speedup less than 1, showing the effectiveness of the
506 asynchronous performance compared to the synchronous mode.
508 In the case of a two clusters configuration, Table~\ref{tab.cluster.2x50} shows
509 that with a deterioration of inter cluster network set with \np[Mbits/s]{5} of
510 bandwidth, a latency in order of a hundredth of a millisecond and a system power
511 of one GFlops, an efficiency of about \np[\%]{40} in asynchronous mode is
512 obtained for a matrix size of 62 elements. It is noticed that the result remains
513 stable even if we vary the external precision from \np{E-5} to \np{E-9}. By
514 increasing the problem size up to 100 elements, it was necessary to increase the
515 CPU power of \np[\%]{50} to \np[GFlops]{1.5} for a convergence of the algorithm
516 with the same order of asynchronous mode efficiency. Maintaining such a system
517 power but this time, increasing network throughput inter cluster up to
518 \np[Mbits/s]{50}, the result of efficiency of about \np[\%]{40} is obtained with
519 high external precision of \np{E-11} for a matrix size from 110 to 150 side
522 For the 3 clusters architecture including a total of 100 hosts,
523 Table~\ref{tab.cluster.3x33} shows that it was difficult to have a combination
524 which gives an efficiency of asynchronous below \np[\%]{80}. Indeed, for a
525 matrix size of 62 elements, equality between the performance of the two modes
526 (synchronous and asynchronous) is achieved with an inter cluster of
527 \np[Mbits/s]{10} and a latency of \np[ms]{E-1}. To challenge an efficiency by
528 \np[\%]{78} with a matrix size of 100 points, it was necessary to degrade the
529 inter cluster network bandwidth from 5 to 2 Mbit/s.
531 A last attempt was made for a configuration of three clusters but more powerful
532 with 200 nodes in total. The convergence with a speedup of \np[\%]{90} was
533 obtained with a bandwidth of \np[Mbits/s]{1} as shown in
534 Table~\ref{tab.cluster.3x67}.
537 The experimental results on executing a parallel iterative algorithm in
538 asynchronous mode on an environment simulating a large scale of virtual
539 computers organized with interconnected clusters have been presented.
540 Our work has demonstrated that using such a simulation tool allow us to
541 reach the following three objectives:
543 \newcounter{numberedCntD}
545 \item To have a flexible configurable execution platform resolving the
546 hard exercise to access to very limited but so solicited physical
548 \item to ensure the algorithm convergence with a raisonnable time and
550 \item and finally and more importantly, to find the correct combination
551 of the cluster and network specifications permitting to save time in
552 executing the algorithm in asynchronous mode.
553 \setcounter{numberedCntD}{\theenumi}
555 Our results have shown that in certain conditions, asynchronous mode is
556 speeder up to \np[\%]{40} than executing the algorithm in synchronous mode
557 which is not negligible for solving complex practical problems with more
558 and more increasing size.
560 Several studies have already addressed the performance execution time of
561 this class of algorithm. The work presented in this paper has
562 demonstrated an original solution to optimize the use of a simulation
563 tool to run efficiently an iterative parallel algorithm in asynchronous
564 mode in a grid architecture.
566 \section*{Acknowledgment}
569 The authors would like to thank\dots{}
572 % trigger a \newpage just before the given reference
573 % number - used to balance the columns on the last page
574 % adjust value as needed - may need to be readjusted if
575 % the document is modified later
576 \bibliographystyle{IEEEtran}
577 \bibliography{IEEEabrv,hpccBib}
585 %%% ispell-local-dictionary: "american"