一种集群计算系统中并行I/O文件存储分配策略

现代计算机系统性能已经由受限于CPU转变为受限于I／O，近年来研究人员对高性能计算中的并行I／O问题进行了深入研究．本论文的主要工作是对集群计算中的并行I／O子系统进行研究，提出一种文件拆分与存储分配的新策略，即：已知知识的文件拆分与分配方法(KKFDA)，该方法保证文件存储方式与访问方式的一致性，达到了提高文件访问时的本地命中率、缩短访盘响应时间的目的．算法设计与实验证明了方法的有效性．

KKFDA:a New Approach of File Stored Assignment for Parallel I/O on Cluster Computing System

The technologies of processor and network develop quickly and that of I/O subsystem is much slower. Based on such fact, a new strategy named Knowledge Known File Declustering and Access (KKFDA) is proposed especially aimed to the limitation about high performance computing. The main idea about the strategy is that the file access way will be known firstly before the big file is declustered and stored on disks in distributed way. The strategy uses the access information and ensures the consistency way among file declustering, storage and access. At last, the result of the higher hits in local disks, low traffic in network, and rapid response time is reachable. The designed algorithm and corresponding experiments have testified the validity of KKFDA.