Threshold-based load balancing policies in heterogeneous distributed systems

This paper presents a new loan balancing policy for heterogeneous distributed systems, in which the servers with heavy load can distribute their tasks to other servers. The proposed policy is based on thresholds to determine whether the current server is heavily loaded or not. The threshold is defined as the ratio of service rates between adjacent server pairs. The current load of a server is indicated by the queue length. The scheduler compares the current load with the thresholds to decide the loan status, and then determines if some tasks should be migrated to other servers. For studying the performance of response time, the Markov process is used to model the behaviour of the heterogeneous distributed system under the proposed threshold policy. Additionally, the equations derived from this model are solved by using the Matrix-Geometric solution technique. Finally, the performance of the proposed policy is compared with that of the generic loan balancing policy, the policy with no load balancing, and the ideal load balancing policy. The results show that the proposed policy is better than that of other policies and close to the ideal loan balancing policy.