Experiences with Distributed Computation of Twin Primes Distributions

Abstract

This paper outlines experiences with development of a distributed framework designed for parallel processing of an application using idle processor time on large numbers of heterogeneous computers. The system places no requirements on the performance or availability of the network interconnections. As such, it is most suitable to parallel applications with modest synchronization requirements and a high ratio of computation time to amount of data needed to define the problem domain subspace. We also provide details on our first application using this framework: a parallel computation which counts the distribution of twin primes, calculates Brun's Constant and the maximal distance between pairs of twin primes. Two primes are twins if they differ by two. We have gathered these results for an order of magnitude larger interval than previously computed. We have gathered these results for an order of magnitude larger interval than previously computed. We present an overview of our implementation and provide preliminary results and future directions for the system.