An efficient model of dynamic task scheduling for distributed systems

Abstract

The authors propose a distributed task scheduling model using a simple processor architecture and a heuristic scheduling algorithm based on small message exchanges between nearest neighbors. An extensive simulation study was used to analyze the proposed model by taking into account a wide range of practical issues. Comparison with other schemes reported in the literature reveals the superiority of the proposed model in terms of various performance measures. The scheduling algorithm supported by some software components improves the response time to a great extent by trying to keep the network nodes equally busy. The exchange of load status information has low complexity. An interesting trade-off between periodic update and non-periodic information update is exhibited. The performance of the proposed algorithms does not degrade when no transfer limit is used. The communication rate of the network has a major influence on the performance but it is shown that further reduction in response time cannot be achieved after a certain increase in communication rate. Simulation experiments on various network topologies showed the adaptive nature of the proposed algorithms.<>