Abstract: The task schedule in PVM is not load balance. This paper presents a dynamic load-balance scheduling policy by combining load practical measuring and load prediction on artificial neutral network and by using check point and process migration. The experiment results on parallel matrix multiplication demonstrate that the response time of improved PVM is less than that of conventional PVM.

摘要： 针对PVM的任务调度不能实现负载平衡的问题,介绍了一种对结点负载采取实测和利用人工神经网络进行预测相结合的策略,并利用检查点技术和进程迁移技术,改进PVM的任务调度机制,实现了动态负载平衡。在并行矩阵乘上的对照实验表明,改进的PVM具有更短的响应时间。

关键词: 负载平衡, 任务调度, 负载预测

CLC Number: 

• TP311