Sensitivity Analysis of List Scheduling Heuristics

Abstract

When jobs have to be processed on a set of identical parallel machines so as to minimize the makespan of the schedule, list scheduling rules form a popular class of heuristics. The order in which jobs appear on the list is assumed here to be determined by the relative size of their processing times; well known special cases are the LPT rule and the SPT rule, in which the jobs are ordered according to non-increasing and non-decreasing processing time respectively. When one of the job processing times is gradually increased, the schedule produced by a list scheduling rule will be affected in a manner reflecting its sensitivity to data perturbations. We analyze this phenomenon and obtain analytical support for the intuitively plausible notion that the sensitivity of a list scheduling rule increases with the quality of the schedule produced.