Equilibria for two parallel links: the strong price of anarchy versus the price of anarchy

Abstract

We consider a job scheduling game with two uniformly related parallel machines (or links). Jobs are atomic players, and the delay of a job is the completion time of the machine running it. The private goal of each job is to minimize its own delay and the social goal is to minimize the maximum delay of any job, that is, to minimize the makespan. We consider the well known price of anarchy as well as the strong price of anarchy, and show that for a wide range of speed ratios these two measures are very different whereas for other speed ratios these two measures give the exact same bound. We extend all our results for models of restricted assignment, where a machine may have an initial load resulting from jobs that can only be assigned to this machine, and show tight results for all variants.