Decentralized decision-making in adaptive task sharing

Adaptive task sharing in a distributed computing system involves system state elimination and decision-making under uncertainty. The paper presents experimental results of an implementation of a novel adaptive task sharing algorithm on a distributed system consisting of VAX stations connected by a local area network. The adaptive task sharing algorithm incorporates decentralized decision-making and decentralized estimation. It introduces metrics for evaluating the quality of decision-making and the quality of system state estimation. Quantitative results on the performance of decision-making, performance of estimation, and the overall performance are presented. The results indicate that the quality of system state estimation is improved by probing more nodes. The quality of task sharing decisions, and the overall performance of the system in turn, is heavily dependent on the quality of system state estimation when the task size is not very small