Multi-Objective Shortest Paths

Abstract

In this chapter, the authors present some developments in the parallelization of multi-objective shortest path search. It is based on two conference papers – for the theoretical part and for the practical part, and the Master thesis of Stephan Erb. Finding shortest paths in graphs is a classical optimization problem with a large number of applications. The classical sequential label-setting algorithms for multi-objective search define a total order on labels with the property that a minimal label is Pareto optimal among all labels. The basic idea behind the improvements described in this chapter is to exploit that the search and update operations performed by paPaSearch are done in a batched fashion. The parallel algorithm described so far is in some sense already close to being practical. However, it uses very fine-grained and irregularly structured parallelism so that good practical performance is not trivial to obtain. The parallel algorithm described so far is in some sense already close to being practical.