1 \documentclass{article}
2 \usepackage[utf8]{inputenc}
3 \usepackage{amsfonts,amssymb}
7 \title{A scalable multisplitting algorithm for solving large sparse linear systems}
12 \author{Raphaël Couturier \and Lilia Ziane Khodja}
17 %%%%%%%%%%%%%%%%%%%%%%%%
18 %%%%%%%%%%%%%%%%%%%%%%%%
22 In this paper we revist the krylov multisplitting algorithm presented in
23 \cite{huang1993krylov} which uses a scalar method to minimize the krylov
24 iterations computed by a multisplitting algorithm. Our new algorithm is based on
25 a parallel multisplitting algorithm with few blocks of large size using a
26 parallel GMRES method inside each block and on a parallel krylov minimization in
27 order to improve the convergence. Some large scale experiments with a 3D Poisson
28 problem are presented. They show the obtained improvements compared to a
29 classical GMRES both in terms of number of iterations and execution times.
33 %%%%%%%%%%%%%%%%%%%%%%%%
34 %%%%%%%%%%%%%%%%%%%%%%%%
37 \section{Introduction}
39 Iterative methods are used to solve large sparse linear systems of equations of
40 the form $Ax=b$ because they are easier to parallelize than direct ones. Many
41 iterative methods have been proposed and adapted by many researchers. For
42 example, the GMRES method and the Conjugate Gradient method are very well known
43 and used by many researchers ~\cite{S96}. Both the method are based on the
44 Krylov subspace which consists in forming a basis of the sequence of successive
45 matrix powers times the initial residual.
47 When solving large linear systems with many cores, iterative methods often
48 suffer from scalability problems. This is due to their need for collective
49 communications to perform matrix-vector products and reduction operations.
50 Preconditionners can be used in order to increase the convergence of iterative
51 solvers. However, most of the good preconditionners are not sclalable when
52 thousands of cores are used.
56 On ne peut pas parler de tout...\\
61 %%%%%%%%%%%%%%%%%%%%%%%
63 %%%%%%%%%%%%%%%%%%%%%%%
64 The key idea of the multisplitting method for solving a large system
65 of linear equations $Ax=b$ consists in partitioning the matrix $A$ in
68 A = M_l - N_l,~l\in\{1,\ldots,L\},
71 where $M_l$ are nonsingular matrices. Then the linear system is solved
72 by iteration based on the multisplittings as follows
74 x^{k+1}=\displaystyle\sum^L_{l=1} E_l M^{-1}_l (N_l x^k + b),~k=1,2,3,\ldots
77 where $E_l$ are non-negative and diagonal weighting matrices such that
78 $\sum^L_{l=1}E_l=I$ ($I$ is an identity matrix). Thus the convergence
79 of such a method is dependent on the condition
81 \rho(\displaystyle\sum^L_{l=1}E_l M^{-1}_l N_l)<1.
85 The advantage of the multisplitting method is that at each iteration
86 $k$ there are $L$ different linear sub-systems
88 v_l^k=M^{-1}_l N_l x_l^{k-1} + M^{-1}_l b,~l\in\{1,\ldots,L\},
91 to be solved independently by a direct or an iterative method, where
92 $v_l^k$ is the solution of the local sub-system. A multisplitting
93 method using an iterative method for solving the $L$ linear
94 sub-systems is called an inner-outer iterative method or a two-stage
95 method. The results $v_l^k$ obtained from the different
96 splittings~(\ref{eq04}) are combined to compute the solution $x^k$ of
97 the linear system by using the diagonal weighting matrices
99 x^k = \displaystyle\sum^L_{l=1} E_l v_l^k,
102 In the case where the diagonal weighting matrices $E_l$ have only zero
103 and one factors (i.e. $v_l^k$ are disjoint vectors), the
104 multisplitting method is non-overlapping and corresponds to the block
106 %%%%%%%%%%%%%%%%%%%%%%%
108 %%%%%%%%%%%%%%%%%%%%%%%
110 \section{Related works}
113 A general framework for studying parallel multisplitting has been presented in
114 \cite{o1985multi} by O'Leary and White. Convergence conditions are given for the
115 most general case. Many authors improved multisplitting algorithms by proposing
116 for example an asynchronous version \cite{bru1995parallel} and convergence
117 conditions \cite{bai1999block,bahi2000asynchronous} in this case or other
118 two-stage algorithms~\cite{frommer1992h,bru1995parallel}.
120 In \cite{huang1993krylov}, the authors proposed a parallel multisplitting
121 algorithm in which all the tasks except one are devoted to solve a sub-block of
122 the splitting and to send their local solution to the first task which is in
123 charge to combine the vectors at each iteration. These vectors form a Krylov
124 basis for which the first task minimizes the error function over the basis to
125 increase the convergence, then the other tasks receive the update solution until
126 convergence of the global system.
130 In \cite{couturier2008gremlins}, the authors proposed practical implementations
131 of multisplitting algorithms that take benefit from multisplitting algorithms to
132 solve large scale linear systems. Inner solvers could be based on scalar direct
133 method with the LU method or scalar iterative one with GMRES.
135 In~\cite{prace-multi}, the authors have proposed a parallel multisplitting
136 algorithm in which large block are solved using a GMRES solver. The authors have
137 performed large scale experimentations upto 32.768 cores and they conclude that
138 asynchronous multisplitting algorithm could more efficient than traditionnal
139 solvers on exascale architecture with hunders of thousands of cores.
142 %%%%%%%%%%%%%%%%%%%%%%%%
143 %%%%%%%%%%%%%%%%%%%%%%%%
146 \section{A two-stage method with a minimization}
147 Let $Ax=b$ be a given sparse and large linear system of $n$ equations
148 to solve in parallel on $L$ clusters, physically adjacent or geographically
149 distant, where $A\in\mathbb{R}^{n\times n}$ is a square and nonsingular
150 matrix, $x\in\mathbb{R}^{n}$ is the solution vector and $b\in\mathbb{R}^{n}$
151 is the right-hand side vector. The multisplitting of this linear system
152 is defined as follows:
156 A & = & [A_{1}, \ldots, A_{L}]\\
157 x & = & [X_{1}, \ldots, X_{L}]\\
158 b & = & [B_{1}, \ldots, B_{L}]
163 where for all $l\in\{1,\ldots,L\}$ $A_l$ is a rectangular block of size $n_l\times n$
164 and $X_l$ and $B_l$ are sub-vectors of size $n_l$, such that $\sum_ln_l=n$. In this
165 case, we use a row-by-row splitting without overlapping in such a way that successive
166 rows of the sparse matrix $A$ and both vectors $x$ and $b$ are assigned to one cluster.
167 So, the multisplitting format of the linear system is defined as follows:
169 \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,
172 where $A_{li}$ is a block of size $n_l\times n_i$ of the rectangular matrix $A_l$, $X_i\neq X_l$
173 is a sub-vector of size $n_i$ of the solution vector $x$ and $\sum_{i<l}n_i+\sum_{i>l}n_i+n_l=n$,
174 for all $i\in\{1,\ldots,l-1,l+1,\ldots,L\}$.
176 The multisplitting method proceeds by iteration for solving the linear system in such a
181 A_{ll}X_l = Y_l \mbox{,~such that}\\
182 Y_l = B_l - \displaystyle\sum_{i=1,i\neq l}^{L}A_{li}X_i,
187 is solved independently by a cluster of processors and communication are required to
188 update the right-hand side vectors $Y_l$, such that the vectors $X_i$ represent the data
189 dependencies between the clusters. In this case, the parallel GMRES method is used
190 as an inner iteration method for solving the linear sub-systems~(\ref{sec03:eq03}).
199 %%%%%%%%%%%%%%%%%%%%%%%%
200 %%%%%%%%%%%%%%%%%%%%%%%%
202 \bibliographystyle{plain}
203 \bibliography{biblio}