[rce2015.git] / paper.tex

\documentclass[times]{cpeauth}

\usepackage{moreverb}

%\usepackage[dvips,colorlinks,bookmarksopen,bookmarksnumbered,citecolor=red,urlcolor=red]{hyperref}

%\newcommand\BibTeX{{\rmfamily B\kern-.05em \textsc{i\kern-.025em b}\kern-.08em
%T\kern-.1667em\lower.7ex\hbox{E}\kern-.125emX}}

\def\volumeyear{2015}

\usepackage{graphicx}
\usepackage{wrapfig}
\usepackage{grffile}

\usepackage[T1]{fontenc}
\usepackage[utf8]{inputenc}
\usepackage{amsfonts,amssymb}
\usepackage{amsmath}
\usepackage{algorithm}
\usepackage{algpseudocode}
%\usepackage{amsthm}
\usepackage{graphicx}
\usepackage[american]{babel}
% Extension pour les liens intra-documents (tagged PDF)
% et l'affichage correct des URL (commande \url{http://example.com})
%\usepackage{hyperref}

\usepackage{url}
\DeclareUrlCommand\email{\urlstyle{same}}

\usepackage[autolanguage,np]{numprint}
\AtBeginDocument{%
  \renewcommand*\npunitcommand[1]{\text{#1}}
  \npthousandthpartsep{}}

\usepackage{xspace}
\usepackage[textsize=footnotesize]{todonotes}

\newcommand{\AG}[2][inline]{%
  \todo[color=green!50,#1]{\sffamily\textbf{AG:} #2}\xspace}
\newcommand{\RC}[2][inline]{%
  \todo[color=red!10,#1]{\sffamily\textbf{RC:} #2}\xspace}
\newcommand{\LZK}[2][inline]{%
  \todo[color=blue!10,#1]{\sffamily\textbf{LZK:} #2}\xspace}
\newcommand{\RCE}[2][inline]{%
  \todo[color=yellow!10,#1]{\sffamily\textbf{RCE:} #2}\xspace}

\algnewcommand\algorithmicinput{\textbf{Input:}}
\algnewcommand\Input{\item[\algorithmicinput]}

\algnewcommand\algorithmicoutput{\textbf{Output:}}
\algnewcommand\Output{\item[\algorithmicoutput]}

\newcommand{\TOLG}{\mathit{tol_{gmres}}}
\newcommand{\MIG}{\mathit{maxit_{gmres}}}
\newcommand{\TOLM}{\mathit{tol_{multi}}}
\newcommand{\MIM}{\mathit{maxit_{multi}}}
\newcommand{\TOLC}{\mathit{tol_{cgls}}}
\newcommand{\MIC}{\mathit{maxit_{cgls}}}

\usepackage{array}
\usepackage{color, colortbl}
\newcolumntype{M}[1]{>{\centering\arraybackslash}m{#1}}
\newcolumntype{Z}[1]{>{\raggedleft}m{#1}}

\newcolumntype{g}{>{\columncolor{Gray}}c}
\definecolor{Gray}{gray}{0.9}



\begin{document}
\RCE{Titre a confirmer.}
\title{Comparative performance analysis of simulated grid-enabled numerical iterative algorithms}
%\itshape{\journalnamelc}\footnotemark[2]}

\author{    Charles Emile Ramamonjisoa and
    David Laiymani and
    Arnaud Giersch and
    Lilia Ziane Khodja and
    Raphaël Couturier
}

\address{
        \centering    
    Femto-ST Institute - DISC Department\\
    Université de Franche-Comté\\
    Belfort\\
    Email: \email{{raphael.couturier,arnaud.giersch,david.laiymani,charles.ramamonjisoa}@univ-fcomte.fr}
}

%% Lilia Ziane Khodja: Department of Aerospace \& Mechanical Engineering\\ Non Linear Computational Mechanics\\ University of Liege\\ Liege, Belgium. Email: l.zianekhodja@ulg.ac.be

\begin{abstract}
  The behavior of multicore applications is always a challenge to predict, especially with a new architecture for which no experiment has been performed. With some applications, it is difficult, if not impossible, to build accurate performance models. That is why another solution is to use a simulation tool which allows us to change many parameters of the architecture (network bandwidth, latency, number of processors) and to simulate the execution of such applications. We have decided to use SimGrid as it enables to benchmark MPI applications.

In this paper, we focus our attention on two parallel iterative algorithms based
on the  Multisplitting algorithm  and we  compare them  to the  GMRES algorithm.
These algorithms  are used to  solve libear  systems. Two different  variantsof the Multisplitting are
studied: one  using synchronoous  iterations and  another one  with asynchronous
iterations. For each algorithm we have  tested different parameters to see their
influence.  We strongly  recommend people  interested  by investing  into a  new
expensive  hardware  architecture  to   benchmark  their  applications  using  a
simulation tool before.

  

  
\end{abstract}

\keywords{Algorithm; distributed; iterative; asynchronous; simulation; simgrid; performance}

\maketitle

\section{Introduction} 
The use of multi-core architectures for solving large scientific problems seems to  become imperative  in  a  lot  of  cases. 
Whatever the scale of these architectures (distributed clusters, computational grids, embedded multi-core \ldots) they  are generally 
well adapted to execute complexe parallel applications operating on a large amount of data.  Unfortunately,  users (industrials or scientists), 
who need such computational resources may not have an easy access to such efficient architectures. The cost of using the platform and/or the cost of 
testing and deploying an application are often very important.  So, in this context it is difficult to optimize a given application for a given 
architecture. In this way and in order to reduce the access cost to these computing resources it seems very interesting to use a simulation environment. 
The advantages are numerous: development life cycle, code debugging, ability to obtain results quickly \ldots

In this paper we focus on a class of highly efficient parallel algorithms called \emph{iterative algorithms}. The
parallel scheme of iterative methods is quite simple. It generally involves the division of the problem
into  several  \emph{blocks}  that  will  be  solved  in  parallel  on  multiple
processing units.  Then each processing unit has to
compute an iteration, to send/receive some data dependencies to/from
its neighbors and to iterate this process until the convergence of
the method. Several well-known methods demonstrate the convergence of these algorithms~\cite{BT89,Bahi07}.
In this processing mode a task cannot begin a new iteration while it
has not received data dependencies from its neighbors. We say that the iteration computation follows a synchronous scheme.
In the asynchronous scheme a task can compute a new iteration without having to
wait for the data dependencies coming from its neighbors. Both
communication and computations are asynchronous inducing that there is
no more idle times, due to synchronizations, between two
iterations~\cite{bcvc06:ij}. This model presents some advantages and drawbacks that we detail in section 2 but even if the number of iterations required to converge is 
generally  greater  than for the synchronous  case, it appears that the asynchronous  iterative scheme  can significantly  reduce  overall execution
times by  suppressing idle  times due to  synchronizations~\cite{Bahi07} for more details.

Nevertheless, in both cases (synchronous or asynchronous) it is very time consuming to find optimal configuration and deployment requirements 
for a given application on a given multi-core architecture. Finding good resource allocations policies under varying CPU power, network speeds and 
loads is very challenging and labor intensive.~\cite{Calheiros:2011:CTM:1951445.1951450}.   This problematic is even more difficult for the asynchronous scheme 
where variations of the parameters of the execution platform can lead to very different number of iterations required to converge and so to very different execution times.
In this challenging context we think that the use of a simulation tool can leverage the possibility of testing various platform scenarios.

The main contribution of this paper is to show that the use of a simulation tool (i.e. the SimGrid   toolkit~\cite{SimGrid}) in the context of real 
parallel applications (i.e. large linear system solver) can help developers to better tune their application for a given multi-core architecture.
To show the validity of this approach we first compare the simulated execution of the multisplitting algorithm  with  the  GMRES   (Generalized   Minimal  Residual) solver
\cite{ref1} both in synchronous mode. The obtained results on different simulated multi-core architectures confirm the results previously obtained on non simulated architecture. 
We also confirm  the efficiency  of the  asynchronous  multisplitting algorithm  comparing to the synchronous  GMRES. In this way and with a simple computing architecture (a laptop) 
SimGrid allows us (with small modifications of the MPI code) to run a test campaign  of  a  real parallel iterative  applications on  different simulated multi-core architectures. 
To our knowledge, there is no related work on the large-scale multi-core simulation of a real synchronous and asynchronous iterative application.  

This paper is organized as follows:


\section{The asynchronous iteration model}

\section{SimGrid}

%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%

\section{Two-stage multisplitting methods}
\label{sec:04}
\subsection{Synchronous and asynchronous two-stage methods for sparse linear systems}
\label{sec:04.01}
In this paper we focus on two-stage multisplitting methods in their both versions synchronous and asynchronous~\cite{Frommer92,Szyld92,Bru95}. These iterative methods are based on multisplitting methods~\cite{O'leary85,White86,Alefeld97} and use two nested iterations: the outer iteration and the inner iteration. Let us consider the following sparse linear system of $n$ equations in $\mathbb{R}$
\begin{equation}
Ax=b,
\label{eq:01}
\end{equation}
where $A$ is a sparse square and nonsingular matrix, $b$ is the right-hand side and $x$ is the solution of the system. Our work in this paper is restricted to the block Jacobi splitting method. This approach of multisplitting consists in partitioning the matrix $A$ into $L$ horizontal band matrices of order $\frac{n}{L}\times n$ without overlapping (i.e. sub-vectors $\{x_\ell\}_{1\leq\ell\leq L}$ are disjoint). The two-stage multisplitting methods solve the linear system~(\ref{eq:01}) iteratively as follows
\begin{equation}
x_\ell^{k+1} = A_{\ell\ell}^{-1}(b_\ell - \displaystyle\sum^{L}_{\substack{m=1\\m\neq\ell}}{A_{\ell m}x^k_m}),\mbox{~for~}\ell=1,\ldots,L\mbox{~and~}k=1,2,3,\ldots
\label{eq:02}
\end{equation}
where $x_\ell$ are sub-vectors of the solution $x$, $b_\ell$ are the sub-vectors of the right-hand side $b$, and $A_{\ell\ell}$ and $A_{\ell m}$ are diagonal and off-diagonal blocks of matrix $A$ respectively. The iterations of these methods can naturally be computed in parallel such that each processor or cluster of processors is responsible for solving one splitting as a linear sub-system 
\begin{equation}
A_{\ell\ell} x_\ell = c_\ell,\mbox{~for~}\ell=1,\ldots,L,
\label{eq:03}
\end{equation}
where right-hand sides $c_\ell=b_\ell-\sum_{m\neq\ell}A_{\ell m}x_m$ are computed using the shared vectors $x_m$. In this paper, we use the well-known iterative method GMRES ({\it Generalized Minimal RESidual})~\cite{saad86} as an inner iteration to approximate the solutions of the different splittings arising from the block Jacobi multisplitting of matrix $A$. The algorithm in Figure~\ref{alg:01} shows the main key points of our block Jacobi two-stage method executed by a cluster of processors. In line~\ref{solve}, the linear sub-system~(\ref{eq:03}) is solved in parallel using GMRES method where $\MIG$ and $\TOLG$ are the maximum number of inner iterations and the tolerance threshold for GMRES respectively. The convergence of the two-stage multisplitting methods, based on synchronous or asynchronous iterations, is studied by many authors for example~\cite{Bru95,bahi07}. 

\begin{figure}[t]
%\begin{algorithm}[t]
%\caption{Block Jacobi two-stage multisplitting method}
\begin{algorithmic}[1]
  \Input $A_\ell$ (sparse matrix), $b_\ell$ (right-hand side)
  \Output $x_\ell$ (solution vector)\vspace{0.2cm}
  \State Set the initial guess $x^0$
  \For {$k=1,2,3,\ldots$ until convergence}
    \State $c_\ell=b_\ell-\sum_{m\neq\ell}A_{\ell m}x_m^{k-1}$
    \State $x^k_\ell=Solve_{gmres}(A_{\ell\ell},c_\ell,x^{k-1}_\ell,\MIG,\TOLG)$\label{solve}
    \State Send $x_\ell^k$ to neighboring clusters\label{send}
    \State Receive $\{x_m^k\}_{m\neq\ell}$ from neighboring clusters\label{recv}
  \EndFor
\end{algorithmic}
\caption{Block Jacobi two-stage multisplitting method}
\label{alg:01}
%\end{algorithm} 
\end{figure}

In this paper, we propose two algorithms of two-stage multisplitting methods. The first algorithm is based on asynchronous model which allows the communications to be overlapped by computations and reduces the idle times resulting from the synchronizations. So in the asynchronous mode, our two-stage algorithm uses asynchronous outer iterations and asynchronous communications between clusters. The communications (i.e. lines~\ref{send} and~\ref{recv} in Figure~\ref{alg:01}) are performed by message passing using MPI non-blocking communication routines. The convergence of the asynchronous iterations is detected when all clusters have locally converged
\begin{equation}
k\geq\MIM\mbox{~or~}\|x_\ell^{k+1}-x_\ell^k\|_{\infty }\leq\TOLM,
\label{eq:04}
\end{equation}    
where $\MIM$ is the maximum number of outer iterations and $\TOLM$ is the tolerance threshold for the two-stage algorithm. 

The second two-stage algorithm is based on synchronous outer iterations. We propose to use the Krylov iteration based on residual minimization to improve the slow convergence of the multisplitting methods. In this case, a $n\times s$ matrix $S$ is set using solutions issued from the inner iteration 
\begin{equation}
S=[x^1,x^2,\ldots,x^s],~s\ll n.
\label{eq:05}
\end{equation}   
At each $s$ outer iterations, the algorithm computes a new approximation $\tilde{x}=S\alpha$ which minimizes the residual
\begin{equation}
\min_{\alpha\in\mathbb{R}^s}{\|b-AS\alpha\|_2}.
\label{eq:06}
\end{equation} 
The algorithm in Figure~\ref{alg:02} includes the procedure of the residual minimization and the outer iteration is restarted with a new approximation $\tilde{x}$ at every $s$ iterations. The least-squares problem~(\ref{eq:06}) is solved in parallel by all clusters using CGLS method~\cite{Hestenes52} such that $\MIC$ is the maximum number of iterations and $\TOLC$ is the tolerance threshold for this method (line~\ref{cgls} in Figure~\ref{alg:02}).
 
\begin{figure}[t]
%\begin{algorithm}[t]
%\caption{Krylov two-stage method using block Jacobi multisplitting}
\begin{algorithmic}[1]
  \Input $A_\ell$ (sparse matrix), $b_\ell$ (right-hand side)
  \Output $x_\ell$ (solution vector)\vspace{0.2cm}
  \State Set the initial guess $x^0$
  \For {$k=1,2,3,\ldots$ until convergence}
    \State $c_\ell=b_\ell-\sum_{m\neq\ell}A_{\ell m}x_m^{k-1}$
    \State $x^k_\ell=Solve_{gmres}(A_{\ell\ell},c_\ell,x^{k-1}_\ell,\MIG,\TOLG)$
    \State $S_{\ell,k\mod s}=x_\ell^k$
    \If{$k\mod s = 0$}
       \State $\alpha = Solve_{cgls}(AS,b,\MIC,\TOLC)$\label{cgls}
       \State $\tilde{x_\ell}=S_\ell\alpha$
       \State Send $\tilde{x_\ell}$ to neighboring clusters
       \Else 
         \State Send $x_\ell^k$ to neighboring clusters
    \EndIf
    \State Receive $\{x_m^k\}_{m\neq\ell}$ from neighboring clusters
  \EndFor
\end{algorithmic}
\caption{Krylov two-stage method using block Jacobi multisplitting}
\label{alg:02}
%\end{algorithm} 
\end{figure}

\subsection{Simulation of two-stage methods using SimGrid framework}
\label{sec:04.02}

One of our objectives when simulating the application in SIMGRID is, as in real life, to get accurate results (solutions of the problem) but also ensure the test reproducibility under the same conditions. According our experience, very few modifications are required to adapt a MPI program to run in SIMGRID simulator using SMPI (Simulator MPI).The first modification is to include SMPI libraries and related header files (smpi.h). The second and important modification is to eliminate all global variables in moving them to local subroutine or using a Simgrid selector called "runtime automatic switching" (smpi/privatize\_global\_variables). Indeed, global variables can generate side effects on runtime between the threads running in the same process, generated by the Simgrid to simulate the grid environment.The last modification on the MPI program pointed out for some cases, the review of the sequence of the MPI\_Isend, MPI\_Irecv and MPI\_Waitall instructions which might cause an infinite loop.     


\paragraph{SIMGRID Simulator parameters}

\begin{itemize}
        \item hostfile: Hosts description file.
        \item plarform: File describing the platform architecture : clusters (CPU power,
\dots{}), intra cluster network description, inter cluster network (bandwidth bw,
latency lat, \dots{}).
        \item archi   : Grid computational description (Number of clusters, Number of 
nodes/processors for each cluster). 
\end{itemize}


In addition, the following arguments are given to the programs at runtime:

\begin{itemize}
        \item Maximum number of inner and outer iterations;
        \item Inner and outer precisions;
        \item Matrix size (N$_{x}$, N$_{y}$ and N$_{z}$);
        \item Matrix diagonal value = 6.0;
        \item Execution Mode: synchronous or asynchronous.
\end{itemize}

At last, note that the two solver algorithms have been executed with the Simgrid selector -cfg=smpi/running\_power which determine the computational power (here 19GFlops) of the simulator host machine.  

%%%%%%%%%%%%%%%%%%%%%%%%%
%%%%%%%%%%%%%%%%%%%%%%%%%

\section{Experimental Results}


\subsection{Setup study and Methodology}

To conduct our study, we have put in place the following methodology 
which can be reused for any grid-enabled applications.

\textbf{Step 1} : Choose with the end users the class of algorithms or 
the application to be tested. Numerical parallel iterative algorithms 
have been chosen for the study in this paper. \\

\textbf{Step 2} : Collect the software materials needed for the 
experimentation. In our case, we have two variants algorithms for the 
resolution of three 3D-Poisson problem: (1) using the classical GMRES (Algo-1)(2) and the multisplitting method (Algo-2). In addition, SIMGRID simulator has been chosen to simulate the behaviors of the 
distributed applications. SIMGRID is running on the Mesocentre datacenter in Franche-Comte University but also in a virtual machine on a laptop. \\

\textbf{Step 3} : Fix the criteria which will be used for the future 
results comparison and analysis. In the scope of this study, we retain 
in one hand the algorithm execution mode (synchronous and asynchronous) 
and in the other hand the execution time and the number of iterations of 
the application before obtaining the convergence. \\

\textbf{Step 4 }: Setup up the different grid testbeds environment 
which will be simulated in the simulator tool to run the program. The 
following architecture has been configured in Simgrid : 2x16 - that is a 
grid containing 2 clusters with 16 hosts (processors/cores) each -, 4x8, 
4x16, 8x8 and 2x50. The network has been designed to operate with a 
bandwidth equals to 10Gbits (resp. 1Gbits/s) and a latency of 8E-6 
microseconds (resp. 5E-5) for the intra-clusters links (resp. 
inter-clusters backbone links). \\

\textbf{Step 5}: Conduct an extensive and comprehensive testings 
within these configurations in varying the key parameters, especially 
the CPU power capacity, the network parameters and also the size of the 
input matrix. Note that some parameters should be fixed to be invariant to allow the 
comparison like some program input arguments. \\

\textbf{Step 6} : Collect and analyze the output results.

\subsection{Factors impacting distributed applications performance in 
a grid environment}

From our previous experience on running distributed application in a 
computational grid, many factors are identified to have an impact on the 
program behavior and performance on this specific environment. Mainly, 
first of all, the architecture of the grid itself can obviously 
influence the performance results of the program. The performance gain 
might be important theoretically when the number of clusters and/or the 
number of nodes (processors/cores) in each individual cluster increase. 

Another important factor impacting the overall performance of the 
application is the network configuration. Two main network parameters 
can modify drastically the program output results : (i) the network 
bandwidth (bw=bits/s) also known as "the data-carrying capacity" 
of the network is defined as the maximum of data that can pass 
from one point to another in a unit of time. (ii) the network latency 
(lat : microsecond) defined as the delay from the start time to send the 
data from a source and the final time the destination have finished to 
receive it. Upon the network characteristics, another impacting factor 
is the application dependent volume of data exchanged between the nodes 
in the cluster and between distant clusters. Large volume of data can be 
transferred in transit between the clusters and nodes during the code 
execution. 

 In a grid environment, it is common to distinguish in one hand, the 
"\,intra-network" which refers to the links between nodes within a 
cluster and in the other hand, the "\,inter-network" which is the 
backbone link between clusters. By design, these two networks perform 
with different speed. The intra-network generally works like a high 
speed local network with a high bandwith and very low latency. In 
opposite, the inter-network connects clusters sometime via heterogeneous 
networks components thru internet with a lower speed. The network 
between distant clusters might be a bottleneck for the global 
performance of the application. 

\subsection{Comparing GMRES and Multisplitting algorithms in 
synchronous mode}

In the scope of this paper, our first objective is to demonstrate the 
Algo-2 (Multisplitting method) shows a better performance in grid 
architecture compared with Algo-1 (Classical GMRES) both running in 
\textbf{\textit{synchronous mode}}. Better algorithm performance 
should means a less number of iterations output and a less execution time 
before reaching the convergence. For a systematic study, the experiments 
should figure out that, for various grid parameters values, the 
simulator will confirm the targeted outcomes, particularly for poor and 
slow networks, focusing on the impact on the communication performance 
on the chosen class of algorithm.

The following paragraphs present the test conditions, the output results 
and our comments.\\


\textit{3.a Executing the algorithms on various computational grid 
architecture scaling up the input matrix size}
\\

% environment
\begin{footnotesize}
\begin{tabular}{r c }
 \hline  
 Grid & 2x16, 4x8, 4x16 and 8x8\\ %\hline
 Network & N2 : bw=1Gbits/s - lat=5.10$^{-5}$ \\ %\hline
 Input matrix size & N$_{x}$ x N$_{y}$ x N$_{z}$ =150 x 150 x 150\\ %\hline
 - &  N$_{x}$ x N$_{y}$ x N$_{z}$  =170 x 170 x 170    \\ \hline
 \end{tabular}
Table 1 : Clusters x Nodes with N$_{x}$=150 or N$_{x}$=170 \\

\end{footnotesize}


 
%\RCE{J'ai voulu mettre les tableaux des données mais je pense que c'est inutile et ça va surcharger}


The results in figure 3 show the non-variation of the number of 
iterations of classical GMRES for a given input matrix size; it is not 
the case for the multisplitting method. 

%\begin{wrapfigure}{l}{100mm}
\begin{figure} [ht!]
\centering
\includegraphics[width=100mm]{cluster_x_nodes_nx_150_and_nx_170.pdf}
\caption{Cluster x Nodes N$_{x}$=150 and N$_{x}$=170} 
%\label{overflow}}
\end{figure}
%\end{wrapfigure}

Unless the 8x8 cluster, the time 
execution difference between the two algorithms is important when 
comparing between different grid architectures, even with the same number of 
processors (like 2x16 and 4x8 = 32 processors for example). The 
experiment concludes the low sensitivity of the multisplitting method 
(compared with the classical GMRES) when scaling up to higher input 
matrix size. 

\textit{\\3.b Running on various computational grid architecture\\}

% environment
\begin{footnotesize}
\begin{tabular}{r c }
 \hline  
 Grid & 2x16, 4x8\\ %\hline
 Network & N1 : bw=10Gbs-lat=8.10$^{-6}$ \\ %\hline
 - & N2 : bw=1Gbs-lat=5.10$^{-5}$ \\
 Input matrix size & N$_{x}$ x N$_{y}$ x N$_{z}$ =150 x 150 x 150\\ \hline \\
 \end{tabular}
Table 2 : Clusters x Nodes - Networks N1 x N2 \\

 \end{footnotesize}



%\begin{wrapfigure}{l}{100mm}
\begin{figure} [ht!]
\centering
\includegraphics[width=100mm]{cluster_x_nodes_n1_x_n2.pdf}
\caption{Cluster x Nodes N1 x N2}
%\label{overflow}}
\end{figure}
%\end{wrapfigure}

The experiments compare the behavior of the algorithms running first on 
a speed inter- cluster network (N1) and a less performant network (N2). 
Figure 4 shows that end users will gain to reduce the execution time 
for both algorithms in using a grid architecture like 4x16 or 8x8: the 
performance was increased in a factor of 2. The results depict also that 
when the network speed drops down, the difference between the execution 
times can reach more than 25\%. 

\textit{\\3.c Network latency impacts on performance\\}

% environment
\begin{footnotesize}
\begin{tabular}{r c }
 \hline  
 Grid & 2x16\\ %\hline
 Network & N1 : bw=1Gbs \\ %\hline
 Input matrix size & N$_{x}$ x N$_{y}$ x N$_{z}$ =150 x 150 x 150\\ \hline\\
 \end{tabular}
Table 3 : Network latency impact \\

\end{footnotesize}



\begin{figure} [ht!]
\centering
\includegraphics[width=100mm]{network_latency_impact_on_execution_time.pdf}
\caption{Network latency impact on execution time}
%\label{overflow}}
\end{figure}


According the results in figure 5, degradation of the network 
latency from 8.10$^{-6}$ to 6.10$^{-5}$ implies an absolute time 
increase more than 75\% (resp. 82\%) of the execution for the classical 
GMRES (resp. multisplitting) algorithm. In addition, it appears that the 
multisplitting method tolerates more the network latency variation with 
a less rate increase of the execution time. Consequently, in the worst case (lat=6.10$^{-5
}$), the execution time for GMRES is almost the double of the time for 
the multisplitting, even though, the performance was on the same order 
of magnitude with a latency of 8.10$^{-6}$. 

\textit{\\3.d Network bandwidth impacts on performance\\}

% environment
\begin{footnotesize}
\begin{tabular}{r c }
 \hline  
 Grid & 2x16\\ %\hline
 Network & N1 : bw=1Gbs - lat=5.10$^{-5}$ \\ %\hline
 Input matrix size & N$_{x}$ x N$_{y}$ x N$_{z}$ =150 x 150 x 150\\ \hline \\
 \end{tabular}
Table 4 : Network bandwidth impact \\

\end{footnotesize}


\begin{figure} [ht!]
\centering
\includegraphics[width=100mm]{network_bandwith_impact_on_execution_time.pdf}
\caption{Network bandwith impact on execution time}
%\label{overflow}
\end{figure}



The results of increasing the network bandwidth depict the improvement 
of the performance by reducing the execution time for both of the two 
algorithms (Figure 6). However, and again in this case, the multisplitting method 
presents a better performance in the considered bandwidth interval with 
a gain of 40\% which is only around 24\% for classical GMRES.

\textit{\\3.e Input matrix size impacts on performance\\}

% environment
\begin{footnotesize}
\begin{tabular}{r c }
 \hline  
 Grid & 4x8\\ %\hline
 Network & N2 : bw=1Gbs - lat=5.10$^{-5}$ \\ %\hline
 Input matrix size & N$_{x}$ = From 40 to 200\\ \hline \\
 \end{tabular}
Table 5 : Input matrix size impact\\

\end{footnotesize}


\begin{figure} [ht!]
\centering
\includegraphics[width=100mm]{pb_size_impact_on_execution_time.pdf}
\caption{Pb size impact on execution time}
%\label{overflow}}
\end{figure}

In this experimentation, the input matrix size has been set from 
N$_{x}$ = N$_{y}$ = N$_{z}$ = 40 to 200 side elements that is from 40$^{3}$ = 64.000 to 
200$^{3}$ = 8.000.000 points. Obviously, as shown in the figure 7, 
the execution time for the two algorithms convergence increases with the 
input matrix size. But the interesting results here direct on (i) the 
drastic increase (300 times) of the number of iterations needed before 
the convergence for the classical GMRES algorithm when the matrix size 
go beyond N$_{x}$=150; (ii) the classical GMRES execution time also almost 
the double from N$_{x}$=140 compared with the convergence time of the 
multisplitting method. These findings may help a lot end users to setup 
the best and the optimal targeted environment for the application 
deployment when focusing on the problem size scale up. Note that the 
same test has been done with the grid 2x16 getting the same conclusion.

\textit{\\3.f CPU Power impact on performance\\}

% environment
\begin{footnotesize}
\begin{tabular}{r c }
 \hline  
 Grid & 2x16\\ %\hline
 Network & N2 : bw=1Gbs - lat=5E-05 \\ %\hline
 Input matrix size & N$_{x}$ = 150 x 150 x 150\\ \hline
 \end{tabular}
Table 6 : CPU Power impact \\

\end{footnotesize}


\begin{figure} [ht!]
\centering
\includegraphics[width=100mm]{cpu_power_impact_on_execution_time.pdf}
\caption{CPU Power impact on execution time}
%\label{overflow}}
\end{figure}

Using the SIMGRID simulator flexibility, we have tried to determine the 
impact on the algorithms performance in varying the CPU power of the 
clusters nodes from 1 to 19 GFlops. The outputs depicted in the figure 6 
confirm the performance gain, around 95\% for both of the two methods, 
after adding more powerful CPU. Note that the execution time axis in the 
figure is in logarithmic scale.

\subsection{Comparing GMRES in native synchronous mode and 
Multisplitting algorithms in asynchronous mode}

The previous paragraphs put in evidence the interests to simulate the 
behavior of the application before any deployment in a real environment. 
We have focused the study on analyzing the performance in varying the 
key factors impacting the results. In the same line, the study compares 
the performance of the two proposed methods in \textbf{synchronous mode
}. In this section, with the same previous methodology, the goal is to 
demonstrate the efficiency of the multisplitting method in \textbf{
asynchronous mode} compare with the classical GMRES staying in the 
synchronous mode.

Note that the interest of using the asynchronous mode for data exchange 
is mainly, in opposite of the synchronous mode, the non-wait aspects of 
the current computation after a communication operation like sending 
some data between nodes. Each processor can continue their local 
calculation without waiting for the end of the communication. Thus, the 
asynchronous may theoretically reduce the overall execution time and can 
improve the algorithm performance.

As stated supra, SIMGRID simulator tool has been used to prove the 
efficiency of the multisplitting in asynchronous mode and to find the 
best combination of the grid resources (CPU, Network, input matrix size, 
\ldots ) to get the highest "\,relative gain" in comparison with the 
classical GMRES time. 


The test conditions are summarized in the table below : \\

% environment
\begin{footnotesize}
\begin{tabular}{r c }
 \hline  
 Grid & 2x50 totaling 100 processors\\ %\hline
 Processors & 1 GFlops to 1.5 GFlops\\
   Intra-Network & bw=1.25 Gbits - lat=5E-05 \\ %\hline
   Inter-Network & bw=5 Mbits - lat=2E-02\\
 Input matrix size & N$_{x}$ = From 62 to 150\\ %\hline
 Residual error precision: 10$^{-5}$ to 10$^{-9}$\\ \hline \\
 \end{tabular}
\end{footnotesize}

Again, comprehensive and extensive tests have been conducted varying the 
CPU power and the network parameters (bandwidth and latency) in the 
simulator tool with different problem size. The relative gains greater 
than 1 between the two algorithms have been captured after each step of 
the test. Table I below has recorded the best grid configurations 
allowing a multiplitting method time more than 2.5 times lower than 
classical GMRES execution and convergence time. The finding thru this 
experimentation is the tolerance of the multisplitting method under a 
low speed network that we encounter usually with distant clusters thru the 
internet.

% use the same column width for the following three tables
\newlength{\mytablew}\settowidth{\mytablew}{\footnotesize\np{E-11}}
\newenvironment{mytable}[1]{% #1: number of columns for data
  \renewcommand{\arraystretch}{1.3}%
  \begin{tabular}{|>{\bfseries}r%
                  |*{#1}{>{\centering\arraybackslash}p{\mytablew}|}}}{%
    \end{tabular}}

\begin{table}[!t]
  \centering
  \caption{Relative gain of the multisplitting algorithm compared with 
the classical GMRES}
  \label{"Table 7"}

  \begin{mytable}{6}
    \hline
    bandwidth (Mbit/s)
    & 5         & 5         & 5         & 5         & 5 \\
    \hline
    latency (ms)
    & 20      & 20      & 20      & 20      & 20 \\
    \hline
    power (GFlops)
    & 1         & 1         & 1         & 1.5       & 1.5 \\
    \hline
    size (N)
    & 62        & 62        & 62        & 100       & 100 \\
    \hline
    Precision
    & \np{E-5}  & \np{E-8}  & \np{E-9}  & \np{E-11} & \np{E-11} \\
    \hline
    Relative gain
    & 2.52     & 2.55     & 2.52     & 2.57     & 2.54 \\
    \hline
  \end{mytable}

  \smallskip

  \begin{mytable}{6}
    \hline
    bandwidth (Mbit/s)
    & 50        & 50        & 50        & 50        & 50 \\
    \hline
    latency (ms)
    & 20      & 20      & 20      & 20      & 20 \\
    \hline
    power (GFlops)
    & 1.5         & 1.5         & 1         & 1.5       & 1.5 \\
    \hline
    size (N)
    & 110       & 120       & 130       & 140       & 150 \\
    \hline
    Precision
    & \np{E-11} & \np{E-11} & \np{E-11} & \np{E-11} & \np{E-11}\\
    \hline
    Relative gain
    & 2.53     & 2.51     & 2.58     & 2.55     & 2.54 \\
    \hline
  \end{mytable}
\end{table}

\section{Conclusion}
CONCLUSION


\section*{Acknowledgment}


The authors would like to thank\dots{}


\bibliographystyle{wileyj}
\bibliography{biblio}

\end{document}

%%% Local Variables:
%%% mode: latex
%%% TeX-master: t
%%% fill-column: 80
%%% ispell-local-dictionary: "american"
%%% End: