A leftover Load Rate-based dynamic balance mechanism is proposed after analyzing static balance technology and dynamic balance technology on the purpose of improving the performance of cluster systems.This mechanism adopts leftover Load Rate as appraisable standard;BP neural network is used to train nodes of cluster systems because of different performance of the nodes;and a new distribution method which is based on static balance technology with a flow table and dynamic balance technology based on load-forecasting is proposed to distribute assignments to nodes of cluster systems.This distribution method reduces times of dispatching assignments between nodes

and improves the performance of cluster systems.The final simulation result suggests this balance mechanism is feasible and valid.