Parallel Agent Passing Tabu Search Algorithm For Graph Partitioning Problem

Abstract

This paper implements standard Tabu Search (TS) and the Simulated Annealing (SA) algorithm, and examine the performance of the basic type of TS and SA for a multi-graph-partitioning problem. The results show that there is a tendency for the solutions obtained using standard TS and SA to be trapped in bad local optimum and TS is too computationally expensive in its pure form. In this respect, the increasing availability of parallel machines offers an interesting opportunity to explore the possibility of parallel TS algorithms, in order to reduce computational time for TS and to obtain better solutions. However, the parallelism for TS generally induces a loss of quality of the best solution found, because it significantly limits the movement possibilities between different subproblems. Hence, we present how the difficulty caused in parallel algorithm for this problem has been overcome.