MPI动态负载平衡策略的研究与实现  

Dynamic Load Balancing Strategies in MPI Parallel Environment

作　　者：卢照[1] 张锦娟[1] 师军[1] 鱼佳欣[1] 

机构地区：[1]陕西师范大学计算机科学学院,陕西西安710062

出　　处：《计算机技术与发展》2010年第5期132-135,149,共5页Computer Technology and Development

摘　　要：集群环境下的并行计算越来越被广泛应用,MPI是集群系统中最重要的编程工具。在并行处理过程中,负载平衡起着很重要的作用,它直接影响到整个算法的效率。文中结合MPI编程环境下的具体特点,提出了基于负载益处估价的方法来判断是否进行任务迁移,给出了负载实时监测和调度的算法,并在每个节点机上间隔性地进行测试。最后在搭建的MPI环境下,运用并行排序方法进行了验证。实验结果表明采用负载前后有了很明显的提高,特别是随着任务量不断增大的情况下提高的效果更加明显。Cluster parallel computing environment is used more and more widely.MPI is the most important programming tool.In parallel processing,load balancing plays a very important role,it directly affects the efficiency of the entire algorithm.In this paper,under MPI programming environment specific characteristics,and based on the benefits of the method of valuation to determine whether trasfer to the task,given the real-time monitoring of load and scheduling algorithms.In each node interval test and compute.Finally,in the MPI environment,the use of parallel sequencing methods to verify the scheduling algorithms.The experimental results show that the load before and after the improvement has been obvious,especially in the larger task,result is more obvious.

关 键 词：集群 负载平衡 任务迁移 MPI并行程序 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]