ECO: E cient Collective Operations for Communication on Heterogeneous Networks

PVM and other distributed computing systems have enabled the use of networks of workstations for parallel computation, but their approach of treating all networks as collections of point-to-point connections does not promote e cient communication|particularly collective communication. The E cient Collective Operations package (ECO) contains programs which solve this problem by analyzing the network and establishing e cient communication patterns. These patterns are used by ECO's library of collective operations. The analysis is done o -line, so that, after paying the one-time cost of analyzing the network, the execution of application programs is not delayed. This paper describes ECO and gives performance results of using ECO to implement the collective communication in CHARMM, a widely used macromolecular dynamics package. ECO substantially improves the performance of CHARMM on a heterogeneous network. ECO facilitates the development of data parallel applications by providing a simple interface to routines which use the available heterogeneous networks e ciently. This approach gives a programmer the ability to use the available networks to their full potential without acquiring any knowledge of the network structure. 1