Journal of Parallel and Distributed Computing 61, 520535 (2001) Heterogeneous Distribution of Computations Solving Linear Algebra Problems on Networks of Heterogeneous Computers Alexey Kalinov and Alexey Lastovetsky Institute for System Programming, Russian Academy of Sciences, 25, Bolshaya Kommunisticheskaya str., Moscow 109004, Russia E-mail: kaispras.ru, lastovispras.ru Received December 21, 1998; revised April 3, 2000; accepted September 18, 2000 This paper presents and analyzes two different strategies of heterogeneous distribution of computations solving dense linear algebra problems on heter- ogeneous networks of computers. The first strategy is based on heterogeneous distribution of processes over processors and homogeneous block cyclic distri- bution of data over the processes. The second is based on homogeneous distribu- tion of processes over processors and heterogeneous block cyclic distribution of data over the processes. Both strategies were implemented in the mpC languagea dedicated parallel extension of ANSI C for efficient and portable programming of heterogeneous networks of computers. The first strategy was implemented using calls to ScaLAPACK; the second strategy was implemented with calls to LAPACK and BLAS. Cholesky factorization on a heterogeneous network of workstations is used to demonstrate that the heterogeneous distribu- tions have an advantage over the traditional homogeneous distribution.  2001 Academic Press Key Words: parallel programming tools; parallel linear algebra software; ScaLAPACK; heterogeneous computing; parallel languages. 1. INTRODUCTION Nowadays, high-performance scientific computations concentrate mostly on distributedmemory supercomputers consisting of identical processors. Therefore, it is no wonder that the main efforts of developers of parallel numerical libraries have been aimed at the achievement of the best performance on such machines. As a rule, in computing on such homogeneous computer systems, a strategy of homo- geneous distribution of computations over processors is used [1, 6, 9]. The strategy will be referred to as the HoHo strategy, ``homogeneous distribution of processes over processorshomogeneous distribution of data over the processes,'' with each doi:10.1006jpdc.2000.1686, available online at http:www.idealibrary.com on 520 0743-731501 35.00 Copyright  2001 by Academic Press All rights of reproduction in any form reserved.