集群计算并行I/O中动态文件分配策略

A Dynamic File Allocation Strategy on Parallel I/O of Cluster Computing System

针对集群计算系统中并行I/O文件分配问题进行了研究,提出一种新的动态文件分配算法——启发式负载与服务时间变化综合平衡策略。该策略将所需要的访问服务时间相似的文件分配到集群系统的同一结点磁盘上,并保证每个磁盘上的负载值不超过负载平衡的极限值,从而在所有结点磁盘间既基本实现负载均衡,又使每个磁盘上文件间的服务时间变化最小化。实验结果表明了该策略在性能提高方面的有效性。

The question about the file allocation strategy is researched on the cluster computing system of parallel I/O. A new dynamic file assignment algorithm named heuristic load ＆ variance of service time integrated balancing strategy is proposed. The new strategy allocates the file to the disk of nodes according to the similar service time on the cluster computing system, and ensures the value of load not exceeding a threshold on each disk. The perfect result of the new algorithm is that the load balancing is realized across all the disks in the system, at the same time, heuristic load ＆ variance of service time integrated balancing strategy can minimize the variance of service time among the files on each disk. The experiment has testified the validity of the new algorithm.