The Load Index Strategy in Distributed Systems

• Main Authors: Jian-Rung Chen, 陳健榮
• Other Authors: Jin-Cherng Lin;
• Format: Others
• Language: en_US
• Published: 1993
• Online Access: http://ndltd.ncl.edu.tw/handle/10289319303753270707

碩士 === 大同工學院 === 資訊工程研究所 === 81 === A load sharing algorithm (or load balancing algorithm) can redistribute the workload among nodes in a distributed system to improve system performance. A basic and important element of the load sharing algorithm is the load index strategy, which characterizes the workload of a node and determines the workload heavy or light. The mostly commonly used load index is the CPU queue length. It works well in simple cases of homogeneous systems. But it can not correctly characterize the workload of a node sometimes, especially in the following situations: (1) There are different average service time of jobs for each node. (2) There are irregular surges of workload in some nodes. (3) When heterogeneous systems are considered. We thus propose a new load index strategy in this article. A new-defined multi-dimentional load index strategy, the fuzzy logic load index strategy, is proposed to characterize the workload of a node more exactly. Four load factors -- the CPU queue length, the average sercive time of jobs, the sharp increase or decrease in job size and in job number -- are suggested to characterize the workload status of a node in complex and protean distributed systems. The fuzzy logic is applied to integrate these factors. Simulations show that the proposed load index strategy makes a load sharing algorithm more effective in redistributing the workload among nodes than the CPU queue length strategy does.