On a Cooperative VNS Parallelization Strategy for the Capacitated Vehicle Routing Problem

Abstract

It is generally accepted that cooperation-based strategies in parallel metaheuristics exhibit better performances in contrast with non-cooperative approaches. In this paper, we study how the cooperation between processes affects the performance and solution quality of parallel algorithms. The purpose of this study is to provide researchers with a practical starting point for designing better cooperation strategies in parallel metaheuristics. To achieve that, we propose two parallel models based on the general variable neighborhood search (GVNS) to solve the capacitated vehicle routing problem (CVRP). Both models scan the search space by using multiple search processes in parallel. The first model lacks communication, while on the other hand, the second model follows a strategy based on information exchange. The received solutions are utilized to guide the search. We conduct an experimental study using well-known benchmark instances of the CVRP, in which the usefulness of communication throughout the search process is assessed. The findings confirm that careful design of the cooperation strategy in parallel metaheuristics can yield better results.