Load Balance Prediction Algorithm Research On Service Cluster System

In the parallel distributed system with multi-processing element (PE), when it is non-uniform for task running on PE, there is load unbalance in every PEs. For effectively and reasonably utilizing resource of each PE, we introduce load balance mechanism to parallel distributed system. Because task run time of a host is close relative to load of this host, we can predict task run time of the host by predicting load of the host. So load prediction of the host is an import substance on load balance system. We can analysis a large of history data, and predict load variance trend of host. The result of prediction can offer the import basis to load balance, thus it decide load balance strategy for effective load balance. At present, on load prediction research of cluster host, there are some problems in: flow prediction of the network is main prediction method on load prediction; load prediction system is not strongly adaptability; only a load index is used; the contradiction between prediction precision and system cost of load prediction is not effectively resolved; most load prediction research focus on load prediction algorithm, but it is ignored on load balance after load prediction. By analyzing defect of load prediction research, we build up a Load Balancing Prediction System based on Mobile Agent----LBPSMA. These systems is based on openMosix parallel distributed system platform, and used classical time series prediction method combining with mobile agent technology, and improve from a simple load prediction algorithm. So it can overcome shortage of actual prediction technology, offer reliable information base to system balancing. This method will be used on this model: Time Series Analysis, Mobile Agent, Hierachical Cluster Analysis. LBPSMA`s system architecture consists of this part: openMosix distributed system platform; Aglets Mobile Agent platform; Load information analyzer; Load predictor; Node selector; Process dispatcher; User interface. In the system, with the object of function modularization, we create a lot of agents, and function of each agent is independence. this agent include: information analysis agent (IAA), load prediction agent (LPA), prediction evaluation agent (PEA), node selection agent (NSA), process scheduling agent (PSA). Based on above architecture, it will be expatiated on LBPSMA`s...