The Research On Task Schedule And Load Balance For Heterogeneous System

With the development of network technique, through the medium of high performance network interconnect 'cluster system become a valid construction to provide high-scalability and high-availability network service. But owing to the dynamics and complexity of node and application and versatility of resource style, load unbalance phenomena often occur in the cluster system at practical running, so debase systemic performance. Consequently, the key to improve the whole performance of cluster system is the technique of load balance and task schedule .After deep analyzing correlation research status quo of task schedule of domestic and overseas, aim at different condition of systemic environment, a load balancing algorithm based on tree spreading thoughts and an adaptive dynamic load balancing algorithm for heterogeneous system is put forward respectively: under environment of advanced cluster management system, the algorithm warrant diffuse path last every last intermediate node loading condition decision local surplus propagation path,in such a way that tree spreading thoughts scheduler maintain load balancing of cluster system interior transport network.An adaptive dynamic load balance algorithm includes two parts. At first, through collecting dynamically the load messages of all nodes in system, by which calculate load number of each node after system balance, this paper use a quintic formula of load specimen distribution to judge the opportunity of starting the load balance algorithm. And then the value function of the best time to move and a formula of load migration are given by load balance algorithm qualitatively, the nodes can be districted to three sets as heavy load,light load and normal load. The nodes selected in heavy load set by using adaptive tactics are carried on the load movement with the nodes in light load set, and consequently the dynamic load balance of the system is realized. Otherwise, the algorithm can regulate dynamically the constant coefficient of load message according to the quality of application problem. Consequently the system is optimized and the performance is improved.Passing GridSim simulator constructing cluster system with eight nodes, and startup adaptive dynamic load balance algorithm, experimental result make show, along with extra load increasing, runtime of parallel program with adaptive dynamic load balance algorithm increase slower than that of without load balance.