On the Minimization of Workload Balancing Criteria on Identical Parallel Machines

This paper studies the problem of scheduling jobs on identical parallel processors to minimize workload balancing criteria. While workload balancing is certainly an important criterion given the need of production systems to use all of their resources efficiently, there is no established measure of performance in the scheduling literature that characterizes total workload balance. In this paper, we propose the normalized standard deviation, the normalized mean deviation, and the normalized mean difference as criteria that measure the balance of the workloads among the machines associated to a schedule. A local search heuristic, which performs multiple exchanges of jobs among machines, is presented.