27-04-2014

[Krylov_multi.git] / krylov_multi.tex

diff --git a/krylov_multi.tex b/krylov_multi.tex
index 3482e92ac785183b99b2e6e535370dc342681fca..d296dcf35b8ffd43b590059731674e38adedecc7 100644 (file)
--- a/krylov_multi.tex
+++ b/krylov_multi.tex
@@ -5,6 +5,7 @@
 \usepackage{graphicx}
 \usepackage{algorithm}
 \usepackage{algpseudocode}
 \usepackage{graphicx}
 \usepackage{algorithm}
 \usepackage{algpseudocode}

+\usepackage{multirow}

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


@@ -12,6 +13,14 @@
 \algnewcommand\algorithmicoutput{\textbf{Output:}}
 \algnewcommand\Output{\item[\algorithmicoutput]}
 
 \algnewcommand\algorithmicoutput{\textbf{Output:}}
 \algnewcommand\Output{\item[\algorithmicoutput]}
 

+\newcommand{\Time}[1]{\mathit{Time}_\mathit{#1}}
+\newcommand{\Prec}{\mathit{prec}}
+\newcommand{\Ratio}{\mathit{Ratio}}
+
+\usepackage{xspace}
+\usepackage[textsize=footnotesize]{todonotes}
+\newcommand{\LZK}[2][inline]{%
+\todo[color=green!40,#1]{\sffamily\textbf{LZK:} #2}\xspace}

 
 \title{A scalable multisplitting algorithm for solving large sparse linear systems} 
 \date{}
 
 \title{A scalable multisplitting algorithm for solving large sparse linear systems} 
 \date{}


@@ -27,7 +36,6 @@
 %%%%%%%%%%%%%%%%%%%%%%%%
 %%%%%%%%%%%%%%%%%%%%%%%%
 
 %%%%%%%%%%%%%%%%%%%%%%%%
 %%%%%%%%%%%%%%%%%%%%%%%%
 

-

 \begin{abstract}
 In  this paper  we  revisit  the krylov  multisplitting  algorithm presented  in
 \cite{huang1993krylov}  which  uses  a  scalar  method to  minimize  the  krylov
 \begin{abstract}
 In  this paper  we  revisit  the krylov  multisplitting  algorithm presented  in
 \cite{huang1993krylov}  which  uses  a  scalar  method to  minimize  the  krylov


@@ -39,18 +47,15 @@ problem  are presented.   They  show  the obtained  improvements  compared to  a
 classical GMRES both in terms of number of iterations and execution times.
 \end{abstract}
 
 classical GMRES both in terms of number of iterations and execution times.
 \end{abstract}
 

-

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

-

 \section{Introduction}
 \section{Introduction}

-

 Iterative methods are used to solve  large sparse linear systems of equations of
 the form  $Ax=b$ because they are  easier to parallelize than  direct ones. Many
 iterative  methods have  been proposed  and  adapted by  many researchers.   For
 example, the GMRES method and the  Conjugate Gradient method are very well known
 Iterative methods are used to solve  large sparse linear systems of equations of
 the form  $Ax=b$ because they are  easier to parallelize than  direct ones. Many
 iterative  methods have  been proposed  and  adapted by  many researchers.   For
 example, the GMRES method and the  Conjugate Gradient method are very well known

-and  used by  many researchers  ~\cite{S96}. Both  the method  are based  on the
+and  used by  many researchers~\cite{S96}. Both  the method  are based  on the

 Krylov subspace which consists in forming  a basis of the sequence of successive
 matrix powers times the initial residual.
 
 Krylov subspace which consists in forming  a basis of the sequence of successive
 matrix powers times the initial residual.
 


@@ -61,8 +66,7 @@ Preconditionners can be  used in order to increase  the convergence of iterative
 solvers.   However, most  of the  good preconditionners  are not  sclalable when
 thousands of cores are used.
 
 solvers.   However, most  of the  good preconditionners  are not  sclalable when
 thousands of cores are used.
 

-
-Traditionnal iterative  solvers have  global synchronizations that  penalize the
+Traditional iterative  solvers have  global synchronizations that  penalize the

 scalability.   Two  possible solutions  consists  either  in using  asynchronous
 iterative  methods~\cite{ref18} or  to  use multisplitting  algorithms. In  this
 paper, we will  reconsider the use of a multisplitting  method. In opposition to
 scalability.   Two  possible solutions  consists  either  in using  asynchronous
 iterative  methods~\cite{ref18} or  to  use multisplitting  algorithms. In  this
 paper, we will  reconsider the use of a multisplitting  method. In opposition to


@@ -70,102 +74,74 @@ traditionnal  multisplitting  method  that  suffer  from  slow  convergence,  as
 proposed  in~\cite{huang1993krylov},  the  use  of a  minimization  process  can
 drastically improve the convergence.
 
 proposed  in~\cite{huang1993krylov},  the  use  of a  minimization  process  can
 drastically improve the convergence.
 

+\LZK[]{Suite\dots}

 
 

+%%%%%%%%%%%%%%%%%%%%%%%%
+%%%%%%%%%%%%%%%%%%%%%%%%

 
 

+\section{Related works}
+A general framework  for studying parallel multisplitting has  been presented in~\cite{o1985multi}
+by O'Leary and White. Convergence conditions are given for the
+most general case.  Many authors improved multisplitting algorithms by proposing
+for  example  an  asynchronous  version~\cite{bru1995parallel}  and  convergence
+conditions~\cite{bai1999block,bahi2000asynchronous}   in  this  case   or  other
+two-stage algorithms~\cite{frommer1992h,bru1995parallel}.

 
 

-%%%%%%%%%%%%%%%%%%%%%%%
-%% BEGIN
-%%%%%%%%%%%%%%%%%%%%%%%
-The key idea  of the multisplitting method for  solving a large system
-of linear equations $Ax=b$ consists  in partitioning the matrix $A$ in
-$L$ several ways
+In~\cite{huang1993krylov},  the  authors  proposed  a  parallel  multisplitting
+algorithm in which all the tasks except  one are devoted to solve a sub-block of
+the splitting  and to send their  local solution to  the first task which  is in
+charge to  combine the vectors at  each iteration.  These vectors  form a Krylov
+basis for  which the first task minimizes  the error function over  the basis to
+increase the convergence, then the other tasks receive the updated solution until
+convergence of the global system. 
+
+In~\cite{couturier2008gremlins}, the  authors proposed practical implementations
+of multisplitting algorithms that take benefit from multisplitting algorithms\LZK[]{répétition ???} to
+solve large scale linear systems. Inner  solvers could be based on scalar direct
+method with the LU method or scalar iterative one with GMRES.\LZK[]{lu et gmres par exemple}
+
+In~\cite{prace-multi},  the  authors have  proposed a  parallel  multisplitting
+algorithm in which large blocks are solved using a GMRES solver. The authors have
+performed large scale experiments up-to  32,768 cores and they conclude that
+asynchronous  multisplitting algorithm  could be more  efficient  than traditional
+solvers on exascale architecture with hundreds of thousands of cores.
+
+\LZK[]{Peut-être autres related works\ldots}\\
+
+The key idea of a multisplitting method to solve a large system of linear equations $Ax=b$ is defined as follows. The first step consists in partitioning the matrix $A$ in $L$ several ways 

 \begin{equation}
 \begin{equation}

-A = M_l - N_l,~l\in\{1,\ldots,L\},
+A = M_l - N_l,

 \label{eq01}
 \end{equation}
 \label{eq01}
 \end{equation}

-where $M_l$ are nonsingular matrices. Then the linear system is solved
-by iteration based on the multisplittings as follows
+where for all $l\in\{1,\ldots,L\}$ $M_l$ are non-singular matrices. Then the linear system is solved by iteration based on the obtained splittings as follows

 \begin{equation}
 x^{k+1}=\displaystyle\sum^L_{l=1} E_l M^{-1}_l (N_l x^k + b),~k=1,2,3,\ldots
 \label{eq02}
 \end{equation}
 \begin{equation}
 x^{k+1}=\displaystyle\sum^L_{l=1} E_l M^{-1}_l (N_l x^k + b),~k=1,2,3,\ldots
 \label{eq02}
 \end{equation}

-where $E_l$ are non-negative and diagonal weighting matrices such that
-$\sum^L_{l=1}E_l=I$ ($I$ is an identity matrix).  Thus the convergence
-of such a method is dependent on the condition
+where $E_l$ are non-negative and diagonal weighting matrices and their sum is an identity matrix $I$. The convergence of such a method is dependent on the condition

 \begin{equation}
 \rho(\displaystyle\sum^L_{l=1}E_l M^{-1}_l N_l)<1.
 \label{eq03}
 \end{equation}
 \begin{equation}
 \rho(\displaystyle\sum^L_{l=1}E_l M^{-1}_l N_l)<1.
 \label{eq03}
 \end{equation}

+where $\rho$ is the spectral radius of the square matrix.

 
 

-The advantage of  the multisplitting method is that  at each iteration
-$k$ there are $L$ different linear sub-systems
+The advantage of the multisplitting method is that at each iteration $k$ there are $L$ different linear sub-systems

 \begin{equation}
 v_l^k=M^{-1}_l N_l x_l^{k-1} + M^{-1}_l b,~l\in\{1,\ldots,L\},
 \label{eq04}
 \end{equation}
 \begin{equation}
 v_l^k=M^{-1}_l N_l x_l^{k-1} + M^{-1}_l b,~l\in\{1,\ldots,L\},
 \label{eq04}
 \end{equation}

-to be solved  independently by a direct or  an iterative method, where
-$v_l^k$  is   the  solution  of   the  local  sub-system.   Thus,  the
-calculations  of $v_l^k$  may be  performed in  parallel by  a  set of
-processors.   A multisplitting  method using  an iterative  method for
-solving the $L$ linear  sub-systems is called an inner-outer iterative
-method or a  two-stage method.  The results $v_l^k$  obtained from the
-different splittings~(\ref{eq04}) are combined to compute the solution
-$x^k$ of the linear system by using the diagonal weighting matrices
+to be solved independently by a direct or an iterative method, where $v_l^k$ is the solution of the local sub-system. Thus the computations of $\{v_l\}_{1\leq l\leq L}$ may be performed in parallel by a set of processors. A multisplitting method using an iterative method as an inner solver is called an inner-outer iterative method or a two-stage method. The results $v_l$ obtained from the different splittings~(\ref{eq04}) are combined to compute solution $x$ of the linear system by using the diagonal weighting matrices

 \begin{equation}
 x^k = \displaystyle\sum^L_{l=1} E_l v_l^k,
 \label{eq05}
 \end{equation}    
 \begin{equation}
 x^k = \displaystyle\sum^L_{l=1} E_l v_l^k,
 \label{eq05}
 \end{equation}    

-In the case where the diagonal weighting matrices $E_l$ have only zero
-and   one   factors  (i.e.   $v_l^k$   are   disjoint  vectors),   the
-multisplitting method is non-overlapping  and corresponds to the block
-Jacobi method.
-%%%%%%%%%%%%%%%%%%%%%%%
-%% END
-%%%%%%%%%%%%%%%%%%%%%%%
-
-\section{Related works}
-
-
-A general framework  for studying parallel multisplitting has  been presented in
-\cite{o1985multi} by O'Leary and White. Convergence conditions are given for the
-most general case.  Many authors improved multisplitting algorithms by proposing
-for  example  an  asynchronous  version  \cite{bru1995parallel}  and  convergence
-conditions  \cite{bai1999block,bahi2000asynchronous}   in  this  case   or  other
-two-stage algorithms~\cite{frommer1992h,bru1995parallel}.
-
-In  \cite{huang1993krylov},  the  authors  proposed  a  parallel  multisplitting
-algorithm in which all the tasks except  one are devoted to solve a sub-block of
-the splitting  and to send their  local solution to  the first task which  is in
-charge to  combine the vectors at  each iteration.  These vectors  form a Krylov
-basis for  which the first task minimizes  the error function over  the basis to
-increase the convergence, then the other tasks receive the update solution until
-convergence of the global system. 
-
-
-
-In \cite{couturier2008gremlins}, the  authors proposed practical implementations
-of multisplitting algorithms that take benefit from multisplitting algorithms to
-solve large scale linear systems. Inner  solvers could be based on scalar direct
-method with the LU method or scalar iterative one with GMRES.
-
-In~\cite{prace-multi},  the  authors  have  proposed a  parallel  multisplitting
-algorithm in which large block are solved using a GMRES solver. The authors have
-performed large scale experimentations upto  32.768 cores and they conclude that
-asynchronous  multisplitting algorithm  could more  efficient  than traditionnal
-solvers on exascale architecture with hunders of thousands of cores.
-
+In the case where the diagonal weighting matrices $E_l$ have only zero and one factors (i.e. $v_l$ are disjoint vectors), the multisplitting method is non-overlapping and corresponds to the block Jacobi method.

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

-

 \section{A two-stage method with a minimization}
 \section{A two-stage method with a minimization}

-Let $Ax=b$ be a given sparse  and large linear system of $n$ equations
-to  solve  in  parallel   on  $L$  clusters,  physically  adjacent  or
-geographically distant, where $A\in\mathbb{R}^{n\times n}$ is a square
-and  nonsingular matrix, $x\in\mathbb{R}^{n}$  is the  solution vector
-and   $b\in\mathbb{R}^{n}$  is   the  right-hand   side   vector.  The
-multisplitting of this linear system is defined as follows:
+Let $Ax=b$ be a given large and sparse linear system of $n$ equations to solve in parallel on $L$ clusters of processors, physically adjacent or geographically distant, where $A\in\mathbb{R}^{n\times n}$ is a square and  non-singular matrix, $x\in\mathbb{R}^{n}$ is the solution vector and $b\in\mathbb{R}^{n}$ is the right-hand side vector. The multisplitting of this linear system is defined as follows

 \begin{equation}
 \left\{
 \begin{array}{lll}
 \begin{equation}
 \left\{
 \begin{array}{lll}


@@ -176,81 +152,41 @@ b & = & [B_{1}, \ldots, B_{L}]
 \right.
 \label{sec03:eq01}
 \end{equation}  
 \right.
 \label{sec03:eq01}
 \end{equation}  

-where for  $l\in\{1,\ldots,L\}$, $A_l$ is a rectangular  block of size
-$n_l\times n$ and $X_l$ and  $B_l$ are sub-vectors of size $n_l$, such
-that  $\sum_ln_l=n$.  In this  case,  we  use  a row-by-row  splitting
-without overlapping in  such a way that successive  rows of the sparse
-matrix $A$ and  both vectors $x$ and $b$ are  assigned to one cluster.
-So,  the multisplitting  format of  the  linear system  is defined  as
-follows:
+where for $l\in\{1,\ldots,L\}$, $A_l$ is a rectangular block of size $n_l\times n$ and $X_l$ and $B_l$ are sub-vectors of size $n_l$ each, such that $\sum_ln_l=n$. In this work, we use a row-by-row splitting without overlapping in such a way that successive rows of sparse matrix $A$ and both vectors $x$ and $b$ are assigned to one cluster. So, the multisplitting format of the linear system is defined as follows

 \begin{equation}
 \begin{equation}

-\forall l\in\{1,\ldots,L\} \mbox{,~} \displaystyle\sum_{i=1}^{l-1}A_{li}X_i + A_{ll}X_l + \displaystyle\sum_{i=l+1}^{L}A_{li}X_i = B_l, 
+\forall l\in\{1,\ldots,L\} \mbox{,~} \displaystyle\sum_{m=1}^{l-1}A_{lm}X_m + A_{ll}X_l + \displaystyle\sum_{m=l+1}^{L}A_{lm}X_m = B_l, 

 \label{sec03:eq02}
 \end{equation} 
 \label{sec03:eq02}
 \end{equation} 

-where $A_{li}$ is  a block of size $n_l\times  n_i$ of the rectangular
-matrix  $A_l$, $X_i\neq  X_l$ is  a sub-vector  of size  $n_i$  of the
-solution vector  $x$ and $\sum_{i<l}n_i+\sum_{i>l}n_i+n_l=n$,  for all
-$i\in\{1,\ldots,l-1,l+1,\ldots,L\}$.
+where $A_{lm}$ is a sub-block of size $n_l\times  n_m$ of the rectangular matrix $A_l$, $X_m\neq  X_l$ is a sub-vector of size $n_m$ of the solution vector $x$ and $\sum_{m\neq l}n_m+n_l=n$, for all $m\in\{1,\ldots,L\}$.

 
 

-The multisplitting method proceeds by iteration for solving the linear
-system in such a way each sub-system
+Our multisplitting method proceeds by iteration for solving the linear system in such a way each sub-system

 \begin{equation}
 \left\{
 \begin{array}{l}
 A_{ll}X_l = Y_l \mbox{,~such that}\\
 \begin{equation}
 \left\{
 \begin{array}{l}
 A_{ll}X_l = Y_l \mbox{,~such that}\\

-Y_l = B_l - \displaystyle\sum_{i=1,i\neq l}^{L}A_{li}X_i,
+Y_l = B_l - \displaystyle\sum_{\substack{m=1\\m\neq l}}^{L}A_{lm}X_m,

 \end{array}
 \right.
 \label{sec03:eq03}
 \end{equation}
 \end{array}
 \right.
 \label{sec03:eq03}
 \end{equation}

-is solved  independently by a cluster of  processors and communication
-are required  to update the  right-hand side vectors $Y_l$,  such that
-the  vectors  $X_i$  represent   the  data  dependencies  between  the
-clusters. In this work,  we use the parallel GMRES method~\cite{ref34}
-as     an     inner      iteration     method     to     solve     the
-sub-systems~(\ref{sec03:eq03}).  It  is a well-known  iterative method
-which  gives good performances  to solve  sparse linear  systems in
-parallel on a cluster of processors.
-
-It should be noted that  the convergence of the inner iterative solver
-for  the  different  linear  sub-systems~(\ref{sec03:eq03})  does  not
-necessarily involve  the convergence of the  multisplitting method. It
-strongly depends on  the properties of the sparse  linear system to be
-solved                 and                the                computing
-environment~\cite{o1985multi,ref18}.  Furthermore,  the multisplitting
-of the  linear system among  several clusters of  processors increases
-the  spectral radius  of  the iteration  matrix,  thereby slowing  the
-convergence.  In   this  paper,  we   based  on  the   work  presented
-in~\cite{huang1993krylov} to increase  the convergence and improve the
-scalability of the multisplitting methods.
-
-In  order  to  accelerate  the  convergence, we  implement  the  outer
-iteration  of the multisplitting  solver as  a Krylov  subspace method
-which minimizes some error function over a Krylov subspace~\cite{S96}.
-The Krylov  space of  the method that  we used  is spanned by  a basis
-composed  of   successive  solutions  issued  from   solving  the  $L$
-splittings~(\ref{sec03:eq03})
+is solved independently by a {\it cluster of processors} and communication are required to update the right-hand side vectors $Y_l$, such that the vectors $X_m$ represent the data dependencies between the clusters. In this work, we use the parallel restarted GMRES method~\cite{ref34} as an inner iteration method to solve sub-systems~(\ref{sec03:eq03}). GMRES is one of the most used Krylov iterative methods to solve sparse linear systems in parallel on clusters of processors. %In practice, GMRES is used with a preconditioner to improve its convergence. In this work, we used a preconditioning matrix equivalent to the main diagonal of sparse sub-matrix $A_{ll}$. This preconditioner is straightforward to implement in parallel and gives good performances in many situations.  
+
+It should be noted that the convergence of the inner iterative solver for the different sub-systems~(\ref{sec03:eq03}) does not necessarily involve the convergence of the multisplitting method. It strongly depends on the properties of the global sparse linear system to be solved and the computing environment~\cite{o1985multi,ref18}. Furthermore, the multisplitting
+of the linear system among several clusters of processors increases the spectral radius of the iteration matrix, thereby slowing the convergence. In this work, we based on the work presented in~\cite{huang1993krylov} to increase the convergence and improve the scalability of the multisplitting methods.
+
+In order to accelerate the convergence, we implemented the outer iteration of the multisplitting solver as a Krylov subspace method which minimizes some error function over a Krylov subspace~\cite{S96}. The Krylov subspace that we used is spanned by a basis composed of successive solutions issued from solving the $L$ splittings~(\ref{sec03:eq03})

 \begin{equation}
 S=\{x^1,x^2,\ldots,x^s\},~s\leq n,
 \label{sec03:eq04}
 \end{equation}
 \begin{equation}
 S=\{x^1,x^2,\ldots,x^s\},~s\leq n,
 \label{sec03:eq04}
 \end{equation}

-where   for  $j\in\{1,\ldots,s\}$,  $x^j=[X_1^j,\ldots,X_L^j]$   is  a
-solution of the  global linear system. The advantage  of such a Krylov
-subspace   is  that   we  need   neither  an   orthogonal   basis  nor
-synchronizations  between  the  different  clusters to  generate  this
-basis.
-
-The  multisplitting   method  is  periodically   restarted  every  $s$
-iterations  with   a  new  initial   guess  $\tilde{x}=S\alpha$  which
-minimizes  the error  function $\|b-Ax\|_2$  over the  Krylov subspace
-spanned  by  the vectors  of  $S$.  So,  $\alpha$  is  defined as  the
-solution of the large overdetermined linear system
+where for $j\in\{1,\ldots,s\}$, $x^j=[X_1^j,\ldots,X_L^j]$ is a solution of the global linear system. The advantage of such a Krylov subspace is that we need neither an orthogonal basis nor synchronizations between clusters to generate this basis.
+
+The multisplitting method is periodically restarted every $s$ iterations with a new initial guess $\tilde{x}=S\alpha$ which minimizes the error function $\|b-Ax\|_2$ over the Krylov subspace spanned by vectors of  $S$. So $\alpha$ is defined as the solution of the large overdetermined linear system

 \begin{equation}
 R\alpha=b,
 \label{sec03:eq05}
 \end{equation}
 \begin{equation}
 R\alpha=b,
 \label{sec03:eq05}
 \end{equation}

-where $R=AS$  is a  dense rectangular matrix  of size $n\times  s$ and
-$s\ll n$. This leads us to solve the system of normal equations
+where $R=AS$ is a dense rectangular matrix of size $n\times s$ and $s\ll n$. This leads us to solve a system of normal equations

 \begin{equation}
 R^TR\alpha=R^Tb,
 \label{sec03:eq06}
 \begin{equation}
 R^TR\alpha=R^Tb,
 \label{sec03:eq06}


@@ -260,27 +196,24 @@ which is associated with the least squares problem
 \text{minimize}~\|b-R\alpha\|_2,
 \label{sec03:eq07}
 \end{equation}  
 \text{minimize}~\|b-R\alpha\|_2,
 \label{sec03:eq07}
 \end{equation}  

-where $R^T$ denotes the transpose  of the matrix $R$.  Since $R$ (i.e.
-$AS$) and  $b$ are  split among $L$  clusters, the  symmetric positive
-definite  system~(\ref{sec03:eq06}) is  solved in  parallel.  Thus, an
-iterative method would be more  appropriate than a direct one to solve
-this system.  We use  the parallel conjugate  gradient method  for the
-normal equations CGNR~\cite{S96,refCGNR}.
+where $R^T$ denotes the transpose of the matrix $R$. Since $R$ (i.e. $AS$) and $b$ are split among $L$ clusters, the symmetric positive definite system~(\ref{sec03:eq06}) is solved in  parallel. Thus an iterative method would be more  appropriate than a direct one to solve this system. We use the parallel conjugate gradient method for the normal equations CGNR~\cite{S96,refCGNR}.

 
 \begin{algorithm}[!t]
 \caption{A two-stage linear solver with inner iteration GMRES method}
 \begin{algorithmic}[1]
 
 \begin{algorithm}[!t]
 \caption{A two-stage linear solver with inner iteration GMRES method}
 \begin{algorithmic}[1]

-\Input $A_l$ (local sparse matrix), $B_l$ (local right-hand side), $x^0$ (initial guess)
-\Output $X_l$ (local solution vector)\vspace{0.2cm}
-\State Load $A_l$, $B_l$, $x^0$
-\State Initialize the minimizer $\tilde{x}^0=x^0$
+\Input $A_l$ (sparse sub-matrix), $B_l$ (right-hand side sub-vector)
+\Output $X_l$ (solution sub-vector)\vspace{0.2cm}
+\State Load $A_l$, $B_l$
+\State Initialize the initial guess $x^0$
+\State Set the minimizer $\tilde{x}^0=x^0$

 \For {$k=1,2,3,\ldots$ until the global convergence}
 \For {$k=1,2,3,\ldots$ until the global convergence}

-\State Restart with $x^0=\tilde{x}^{k-1}$: \textbf{for} $j=1,2,\ldots,s$ \textbf{do}
-\State\hspace{0.5cm} Inner iteration solver: \Call{InnerSolver}{$x^0$, $j$}
-\State\hspace{0.5cm} Construct the basis $S$: add the column vector $X_l^j$ to the matrix $S_l^k$
-\State\hspace{0.5cm} Exchange the local solution vector $X_l^j$ with the neighboring clusters
-\State\hspace{0.5cm} Compute the dense matrix $R$: $R_l^{k,j}=\sum^L_{i=1}A_{li}X_i^j$ 
-\State\textbf{end for} 
+\State Restart with $x^0=\tilde{x}^{k-1}$:
+\For {$j=1,2,\ldots,s$}
+\State Inner iteration solver: \Call{InnerSolver}{$x^0$, $j$}
+\State Construct basis $S$: add column vector $X_l^j$ to the matrix $S_l^k$
+\State Exchange local values of $X_l^j$ with the neighboring clusters
+\State Compute dense matrix $R$: $R_l^{k,j}=\sum^L_{i=1}A_{li}X_i^j$ 
+\EndFor 

 \State Minimization $\|b-R\alpha\|_2$: \Call{UpdateMinimizer}{$S_l$, $R$, $b$, $k$}
 \State Local solution of the linear system $Ax=b$: $X_l^k=\tilde{X}_l^k$
 \State Exchange the local minimizer $\tilde{X}_l^k$ with the neighboring clusters
 \State Minimization $\|b-R\alpha\|_2$: \Call{UpdateMinimizer}{$S_l$, $R$, $b$, $k$}
 \State Local solution of the linear system $Ax=b$: $X_l^k=\tilde{X}_l^k$
 \State Exchange the local minimizer $\tilde{X}_l^k$ with the neighboring clusters


@@ -289,49 +222,30 @@ normal equations CGNR~\cite{S96,refCGNR}.
 \Statex
 
 \Function {InnerSolver}{$x^0$, $j$}
 \Statex
 
 \Function {InnerSolver}{$x^0$, $j$}

-\State Compute the local right-hand side: $Y_l = B_l - \sum^L_{i=1,i\neq l}A_{li}X_i^0$
-\State Solving the local splitting $A_{ll}X_l^j=Y_l$ using the parallel GMRES method, such that $X_l^0$ is the initial guess
+\State Compute local right-hand side $Y_l = B_l - \sum^L_{\substack{m=1\\m\neq l}}A_{lm}X_m^0$
+\State Solving local splitting $A_{ll}X_l^j=Y_l$ using parallel GMRES method, such that $X_l^0$ is the initial guess

 \State \Return $X_l^j$
 \EndFunction
 
 \Statex
 
 \Function {UpdateMinimizer}{$S_l$, $R$, $b$, $k$}
 \State \Return $X_l^j$
 \EndFunction
 
 \Statex
 
 \Function {UpdateMinimizer}{$S_l$, $R$, $b$, $k$}

-\State Solving the normal equations $(R^k)^TR^k\alpha^k=(R^k)^Tb$ in parallel by $L$ clusters using the parallel CGNR method
-\State Compute the local minimizer: $\tilde{X}_l^k=S_l^k\alpha^k$
+\State Solving normal equations $(R^k)^TR^k\alpha^k=(R^k)^Tb$ in parallel by $L$ clusters using parallel CGNR method
+\State Compute local minimizer $\tilde{X}_l^k=S_l^k\alpha^k$

 \State \Return $\tilde{X}_l^k$
 \EndFunction
 \end{algorithmic}
 \label{algo:01}
 \end{algorithm}
 
 \State \Return $\tilde{X}_l^k$
 \EndFunction
 \end{algorithmic}
 \label{algo:01}
 \end{algorithm}
 

-The  main key points  of the  multisplitting method  to solve  a large
-sparse  linear  system  are  given in  Algorithm~\ref{algo:01}.   This
-algorithm is based on a two-stage method with a minimization using the
-GMRES iterative method as an  inner solver. It is executed in parallel
-by  each cluster  of processors.   The matrices  and vectors  with the
-subscript  $l$ represent  the local  data for  the cluster  $l$, where
-$l\in\{1,\ldots,L\}$. The two-stage solver uses two different parallel
-iterative algorithms:  the GMRES method  to solve each splitting  on a
-cluster of processors, and the CGNR method executed in parallel by all
-clusters  to minimize  the  function error  over  the Krylov  subspace
-spanned by  $S$.  The  algorithm requires two  global synchronizations
-between the $L$  clusters. The first one is  performed at line~$12$ in
-Algorithm~\ref{algo:01}  to exchange  the local  values of  the vector
-solution $x$ (i.e. the  minimizer $\tilde{x}$) required to restart the
-multisplitting  solver. The  second  one is  needed  to construct  the
-matrix $R$ of  the Krylov subspace.  We choose  to perform this latter
-synchronization $s$  times in every  outer iteration $k$  (line~$7$ in
-Algorithm~\ref{algo:01}). This is a straightforward way to compute the
-matrix-matrix    multiplication     $R=AS$.     We    implement    all
-synchronizations   by   using   the   MPI   collective   communication
-subroutines.
+The main key points of our multisplitting method to solve a large sparse linear system are given in Algorithm~\ref{algo:01}. This algorithm is based on a two-stage method with a minimization using restarted GMRES iterative method as an inner solver. It is executed in parallel by each cluster of processors. Matrices and vectors with the subscript $l$ represent the local data for cluster  $l$, where $l\in\{1,\ldots,L\}$. The two-stage solver uses two different parallel iterative algorithms: GMRES method to solve each splitting~(\ref{sec03:eq03}) on a cluster of processors, and CGNR method executed in parallel by all clusters to minimize the function error~(\ref{sec03:eq07}) over the Krylov subspace spanned by $S$. The algorithm requires two global synchronizations between $L$ clusters. The first one is performed at line~$12$ in Algorithm~\ref{algo:01} to exchange local values of vector solution $x$ (i.e. the minimizer $\tilde{x}$) required to restart the multisplitting solver. The second one is needed to construct the matrix $R$ of the Krylov subspace. We chose to perform this latter synchronization $s$ times in every outer iteration $k$ (line~$7$ in Algorithm~\ref{algo:01}). This is a straightforward way to compute the sparse matrix-dense matrix multiplication $R=AS$. We implemented all synchronizations by using message passing collective communications of MPI library.

 
 

+%%%%%%%%%%%%%%%%%%%%%%%%
+%%%%%%%%%%%%%%%%%%%%%%%%

 
 \section{Experiments}
 
 \section{Experiments}

-
-In order  to illustrate  the interest  of our algorithm.   We have  compared our
-algorithm  with  the  GMRES  method  which  a very  well  used  method  in  many
+In order to illustrate  the interest  of our algorithm. We have  compared our
+algorithm  with  the  GMRES  method  which  is a very  well  used  method  in  many

 situations.  We have chosen to focus on only one problem which is very simple to
 implement: a 3 dimension Poisson problem.
 
 situations.  We have chosen to focus on only one problem which is very simple to
 implement: a 3 dimension Poisson problem.
 


@@ -345,14 +259,83 @@ implement: a 3 dimension Poisson problem.
 \end{equation}
 
 After discretization, with a finite  difference scheme, a seven point stencil is
 \end{equation}
 
 After discretization, with a finite  difference scheme, a seven point stencil is

-obtained.
+used. It  is well-known that the  spectral radius of  matrices representing such
+problems are very close to 1.  Moreover, the larger the number of discretization
+points is,  the closer to 1  the spectral radius  is.  Hence, to solve  a matrix
+obtained for  a 3D Poisson  problem, the number  of iterations is high.  Using a
+preconditioner  it  is   possible  to  reduce  the  number   of  iterations  but
+preconditioners are not scalable when using many cores.
+
+Doing many experiments  with many cores is  not easy and requires to  access to a
+supercomputer  with several  hours for  developing  a code  and then  improving
+it. In the following we presented  some experiments we could achieved out on the
+Hector architecture,  the previous UK's  high-end computing resource,  funded by
+the UK Research Councils, which has been stopped in the early 2014.
+
+Table~\ref{tab1} shows  the result of  the experiments.  The first  column shows
+the  size of  the  3D Poisson  problem.  The size  is chosen  in  order to  have
+approximately  50,000 components  per core.   The second  column  represents the
+number of  cores used. In parenthesis,  there is the decomposition  used for the
+Krylov multisplitting. The  third column and the sixth  column respectively show
+the execution time for the GMRES  and the Kyrlov multisplitting codes. The fourth
+and  the   seventh  column  describes   the  number  of  iterations.    For  the
+multisplitting  code, the  total number  of inner  iterations is  represented in
+parenthesis. For  the GMRES code (alone  and in the  multisplitting version) the
+restart parameter is fixed to 16. The precision of the GMRES version is fixed to
+1e-6. For  the multisplitting,  there are two  precisions, one for  the external
+solver which is fixed to 1e-6 and another one for the inner solver (GMRES) which
+is fixed to 1e-10. It should be noted  that a high precision is used but we also
+fixed a  maximum number of  iterations for each  internal step. In  practise, we
+limit the  number of internal step to  10. So an internal  iteration is finished
+when the precision is reached or  when the maximum internal number of iterations
+is reached. The precision and the maximum number of iterations of CGNR method are fixed to 1e-25 and 20, respectively. The size of the Krylov subspace basis $S$ is fixed to 10 vectors.
+\LZK{J'ai ajouté les paramètres concernant la résolution du problème de moindres carrés. Confirmer leur valeurs.}
+
+
+
+\begin{table}[htbp]
+\begin{center}
+\begin{tabular}{|c|c||c|c|c||c|c|c||c|} 
+\hline
+\multirow{2}{*}{Pb size}&\multirow{2}{*}{Nb. cores} &  \multicolumn{3}{c||}{GMRES} &  \multicolumn{3}{c||}{Krylov Multisplitting} & \multirow{2}{*}{Ratio}\\
+ \cline{3-8}
+           &                   &  Time (s) & nb Iter. & $\Delta$  &   Time (s)& nb Iter. & $\Delta$ & \\
+\hline
+$468^3$ & 2048 (2x1024)        &  299.7    & 41,028    & 5.02e-8  &  48.4    & 691(6,146) & 8.24e-08  & 6.19   \\
+\hline
+$590^3$ & 4096 (2x2048)        &  433.1    & 55,494    & 4.92e-7  &  74.1    & 1,101(8,211) & 6.62e-08  & 5.84   \\
+\hline
+$743^3$ & 8192 (2x4096)        & 704.4     & 87,822    & 4.80e-07 &  151.2   & 3,061(14,914) & 5.87e-08 & 4.65    \\
+\hline
+$743^3$ & 8192 (4x2048)        & 704.4     & 87,822    & 4.80e-07 &  110.3   & 1,531(12,721) & 1.47e-07& 6.39  \\
+\hline
+
+\end{tabular}
+\caption{Results}
+\label{tab1}
+\end{center}
+\end{table}
+
+
+From these  experiments, it can be  observed that the  multisplitting version is
+always  faster   than  the  GMRES   version.   The  acceleration  gain   of  the
+multisplitting version is between 4 and 6.  It can be noticed that the number of
+iterations is drastically reduced with  the multisplitting version even it is not
+neglectable.

 
 \section{Conclusion and perspectives}
 
 \section{Conclusion and perspectives}

+We have implemented a Krylov multisplitting method to solve sparse linear systems on large-scale computing platforms. We have developed a synchronous two-stage method based on the block Jacobi multisplitting and uses GMRES iterative method as an inner iteration. Our contribution in this paper is twofold. First we have constituted a multi-cluster environment based on processors of the large-scale computing platform on which each linear sub-system issued from the splitting is solved in parallel by a cluster of processors. Second, we have implemented the outer iteration of the multisplitting method as a Krylov subspace method which minimizes some error function. This increases the convergence and improves the scalability of the multisplitting method.
+
+We have tested our multisplitting method for solving the sparse linear system issued from the discretization of the 3D Poisson problem. We have compared its performances to GMRES method on a supercomputer composed of 2048 to 8192 cores. The experimental results showed that the multisplitting method is about 4 to 6 times faster than the GMRES method for different sizes of the problem split into 2 or 4 blocks when using the multisplitting method. Indeed, the GMRES method has difficulties to scale with many cores while the Krylov multisplitting method allows to hide latency and reduce the inter-cluster communications.
+
+In future works, we plan to conduct experiments on larger number of cores and test the scalability of our Krylov multisplitting method. It would be interesting to validate its performances for solving other linear/nonlinear and symmetric/nonsymmetric problems. Moreover, we intend to develop multisplitting methods based on asynchronous iteration in which communications are overlapped by computations. These methods would be interesting for platforms composed of distant clusters interconnected by a high-latency network. In addition, we intend to investigate the convergence improvements by using preconditioning techniques and multisplitting methods with overlapping blocks.    
+

 
 

-Other applications (=> other matrices)\\
-Larger experiments\\
-Async\\
-Overlapping
+%Other applications (=> other matrices)\\
+%Larger experiments\\
+%Async\\
+%Overlapping\\
+%preconditioning

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