Parallel computing on personal computers

Abstract

A set of Personal Computers is connected to form a ring-structured parallel system. Each computer has access to its local memory and can communicate with its two neighbours in the ring. This network of asynchronous processors is used to solve in parallel combinatorial optimization problems that are too time- and space-consuming to be handled on a single personal computer. Heuristics are developed to simulate in distributed memory the typical data structures needed in branch-and-bound-algorithms: A single priority queue is maintained and updated in several heaps with very little synchronization overhead. To show the performance of the ring a parallel version of the Travelling-Salesman-Problem is implemented. Execution times and speedups for 50 random graphs solved with up to 16 ring members are discussed.