A robust and flexible microeconomic scheduler for parallel computers

We compare the behavior of a microeconomic scheduler with three commonly used scheduling policies under a systematically designed set of experiments. We experiment with three different income distribution policies within the economic scheduling algorithm. We show that the other scheduling policies can be considered as limiting cases of the microeconomic scheduling policy. Our results show that the economic scheduler performs the best overall when mutually antagonistic criteria, such as system and user response times on one hand, and system utilization and maximum waiting times on the other hand, are considered. We conclude that the microeconomic scheduler exhibits robust performance across a broad range of parameters, and is exible in permitting trade-o s between antagonistic goals. 1This paper is available on the WWW at URL http://www.cs.odu.edu/~pothen/papers.html. This work was supported by National Science Foundation grants CCR-9412698 and DMS-9505110, by U. S. Department of Energy grant DE-FG05-94ER25216, and by the National Aeronautics and Space Administration under NASA Contract NAS1-19480 while the second author was in residence at the Institute for Computer Applications in Science and Engineering.