一种基于预测的负载平衡策略

A predicted load balancing strategy for distributed and parallel systems

提出了一种基于预测的动态负载平衡算法 ,算法以本地负载信息为基础预测结点到达空闲状态的时间 ,并且在此之前发出任务请求 ,从而保证系统中各结点都处于忙碌状态 ,以提高系统资源的利用率与系统整体性能 .由于算法在实施负载平衡时不需要在各进程之间进行同步操作 ,而且各结点自主决定负载平衡参数 ,对系统中其他结点的信息要求少 ,因此负载平衡的开销较小 .实验证明 ,该算法对于科学计算类问题具有较好的效果 .

A prediction based algorithm for dynamic load balancing is proposed, which is capable of predicting the time for a node to reach the idle state based on the local loading information and issuing a task request before this so as to ensure that all nodes in the system will remain busy to enhance the utilization rate of the system resources and the overall system performance. As there is no need for synchronization among the various processes when implementing load balancing and the nodes are free to independently determine the load balancing parameters, which does not require too much information on the other nodes, the expenses of load balancing are relatively low. It has been proved by experiment that this algorithm is fairly effective for problems of scientific calculations.