ANALYSIS AND EVALUATION OF HERURISTIC METHODS FOR STATIC TASK-SCHEDULING

Optimal task scheduling in a multiprocessor environment is known to be an NP-hard problem. The research efforts in this area have focused on heuristic methods for efficient distribution of tasks. This paper introduces two static task scheduling algorithms. The first algorithm, called Heavy Node First (HNF), is based on a local analysis of program graph nodes at each level. The second algorithm, called Weighted Length (WL), considers a more global view of the program graph and takes into account the relationship among the nodes at different levels. The two algorithms are compared to the classical Critical Path Method (CPM). For a given Directed Acyclic Graph of n nodes representing a program, it is shown that HNF requires O(n log(n)) steps while WL and CPM require O(n2) steps to accomplish the allocation. The nontrivial worst case performances of the three algorithms are analytically evaluated and their average case performances are evaluated through a simulation study. It is shown that the performance of the three algorithms, which is measured in terms of program execution time and processors' idle time, is almost the same. Therefore, taking into account the time complexity of the task distribution itself, we conclude that a simple and fast heuristic algorithm, such as HNF, may be sufficient to achieve adequate performance. © 1990.