On the Performance of PC Clusters in Solving Partial Differential Equations.
Xing Hui Cai, Åsmund Ødegård · 2001
Introduction For low-cost PC clusters that are made up of only o-the-shelf components, a common notion is that they can not be real competitors for conventional supercomputers in respect of delivering sustained computing power. We aim to show in the present paper that this needs not be the case, especially when PC clusters are used to solve partial dierential equations (PDEs). More specically, we will look at the issue of how to achieve good PDE solving performance on PC clusters, and present several case studies for a particular Linux cluster of 48 Pentium-III processors. The study of many physical phenomena resorts to the numerical solution of certain PDEs. To achieve a satisfactory resolution, it is often necessary to use millions of degrees of freedom. Such large scale computations demand huge computing power, which is traditionally provided by supercomputers. However, recent research in the eld of cluster computing has indicated that PC clusters can be an attractive