This paper presents a new load balancing policy in heterogeneous distributed systems. All nodes in this system connected in point-to-point may have different service rates, and each node can serve not only external arrival tasks but also the tasks migrated from the other nodes. The scheduler takes into account the threshold value, based on the ratio of service rates, and the current queue length in determining whether one task shall be migrated to the other node or left for the local node. A Markov process model is used to describe the behavior of the heterogeneous distributed system under the proposed policies. The model is then solved by using the matrix-geometric solution technique. Performance of the policy is compared to that of the generic load balancing policy, no load balancing policy, and the ideal load balancing policy. The results show that the proposed policy is better and closer to the ideal load balancing policy.
展开▼
