Recently, cluster computers become a viable and less expensive alternative to multiprocessor systems. However, a serious difficulty in concurrent programming of a cluster computer system is how to deal with scheduling and load balancing of such a system which may consist of heterogeneous computers. Based on the four-tuple (dynamic load balancing model) which is proposed by other scholars, this paper introduces a five-tuple to better solve the above problems. The five-tuple involves balancing environment, task allotment, load estimate, scheduling strategy and scheduling evaluation. The five-tuple interprets the full aspects of load balancing and makes the logical relationship between the five tuples more clear by using formalized language. According to the five-tuple model and the corresponding solution of each tuple, this paper realizes the model under certain experimental conditions in the Linux cluster system. Experiments are given testing the five-tuple based on tasks that are independent, having a little or no relativity, uncertain executing time. The result shows that the model proposed in this paper can make the load balancing better. What’s more, scheduling evaluation shows that the strategies are reasonable.

Parallel, Cluster computers, Load balancing model, Load estimate