Scheduling When You Do Not Know the Number of Machines

Abstract

Often in a scheduling problem, there is uncertainty about the jobs to be processed. The issue of uncertainty regarding the machines has been much less studied. In this article, we study a scheduling environment in which jobs first need to be grouped into some sets before the number of machines is known, and then the sets need to be scheduled on machines without being separated. To evaluate algorithms in such an environment, we introduce the idea of an α-robust algorithm, one that is guaranteed to return a schedule on any number m of machines that is within an α factor of the optimal schedule on m machine, where the optimum is not subject to the restriction that the sets cannot be separated. Under such environment, we give a (5\3+ϵ)-robust algorithm for scheduling on parallel machines to minimize makespan and show a lower bound 4\3. For the special case when the jobs are infinitesimal, we give a 1.233-robust algorithm with an asymptotic lower bound of 1.207. We also study a case of fair allocation, where the objective is to minimize the difference between the maximum and minimum machine load.