Algorithm for optimally distributing quantized load on transputers with unequal speed: An application to the detection of gravitational wave signals from coalescing binaries

Abstract

In a parallel computing system, we work with a network of a large number of processors wherein the performance characteristics each processer may have are different. This leads to a situation that when there is equal load on all the processer, some complete the job before the others. To make the optimum use of the available computing facility and optimise on time, it is necessary to balance the load on the processers according to there characteristics like speed etc. Here we present an algorithm to optimse to on ‘ time ‘ when difference processer have difference speed and the load is quantised in integral multiples of a given unit of load. The algorithm distribute the load in such a manner that all the processer work optimally and the processing time is minimal. The optimal distribution of the load is achived by employing the well known bisection technique for finding the rots of an equation. We discuss this algorithm in the context of our application for filtering the coalescing binary gravitational wave signals. Numerical result are finally discussed for the 64 transputer machine ( PARAM ) .