Reducing the effect of global communication in GMRES(m) and CG on parallel distributed memory computers

by: E de Sturler, HA van der Vorst

Applied Numerical Mathematics, Vol. 18, No. 4. (October 1995), pp. 441-459.

View FullText article

DOI, ScienceDirect

Reviews [Write a review of this article]

There are no reviews of this article

Find related articles from these CiteULike users

rblake

Find related articles with these CiteULike tags

algorithm, cg, parallel

Abstract

In this paper we study possibilities for the reduction of communication overhead introduced by inner products in the iterative solution methods CG and GMRES(m). The performance of these methods on massively parallel distributed memory machines is often limited because of the global communication required for the inner products. We investigate two ways of improvement. One is to assemble the results of a number of inner products collectively. The other is to create situations where communication can be overlapped with computation. The matrix-vector products may also introduce some communication overhead, but for many relevant problems this involves only communication with a few nearby processors that is easily overlapped as well. So this may, but does not necessarily, further degrade the performance of the algorithm.

@article{citeulike:2431301, abstract = {In this paper we study possibilities for the reduction of communication overhead introduced by inner products in the iterative solution methods CG and GMRES(m). The performance of these methods on massively parallel distributed memory machines is often limited because of the global communication required for the inner products. We investigate two ways of improvement. One is to assemble the results of a number of inner products collectively. The other is to create situations where communication can be overlapped with computation. The matrix-vector products may also introduce some communication overhead, but for many relevant problems this involves only communication with a few nearby processors that is easily overlapped as well. So this may, but does not necessarily, further degrade the performance of the algorithm.}, author = {de Sturler, E. and van der Vorst, H. A. }, citeulike-article-id = {2431301}, doi = {10.1016/0168-9274(95)00079-A}, journal = {Applied Numerical Mathematics}, keywords = {algorithm, cg, parallel}, month = {October}, number = {4}, pages = {441--459}, posted-at = {2008-02-26 20:35:54}, priority = {2}, title = {Reducing the effect of global communication in GMRES(m) and CG on parallel distributed memory computers}, url = {http://dx.doi.org/10.1016/0168-9274(95)00079-A}, volume = {18}, year = {1995} }

RIS record

TY - JOUR ID - citeulike:2431301 L3 - citeulike-article-id:2431301 TI - Reducing the effect of global communication in GMRES(m) and CG on parallel distributed memory computers JF - Applied Numerical Mathematics VL - 18 IS - 4 SP - 441 EP - 459 N2 - In this paper we study possibilities for the reduction of communication overhead introduced by inner products in the iterative solution methods CG and GMRES(m). The performance of these methods on massively parallel distributed memory machines is often limited because of the global communication required for the inner products. We investigate two ways of improvement. One is to assemble the results of a number of inner products collectively. The other is to create situations where communication can be overlapped with computation. The matrix-vector products may also introduce some communication overhead, but for many relevant problems this involves only communication with a few nearby processors that is easily overlapped as well. So this may, but does not necessarily, further degrade the performance of the algorithm. KW - algorithm KW - cg KW - parallel AU - de Sturler, E AU - van der Vorst, HA PY - 1995/10// UR - http://dx.doi.org/10.1016/0168-9274(95)00079-A ER -

RIS BibTeX