Graph Metrics for Predicting Speedup in Static Multiprocessor Scheduling

摘要

This paper presents a set of metrics for estimating the speedup achievable in static multiprocessor scheduling using a previously introduced Genetic Algorithm (GA) approach. This is of major importance because, although conventional wisdom suggests that metaheuris-tics such as GAs have the potential to improve over standard heuristics, little research has been conducted on characterizing the sorts of graphs that they should excel at. We describe several metrics and illustrate that four of them can predict the speed up with an accuracy of almost 90%.