传统文件系统中的Ｃａｃｈｅ和预取技术是两种降低访问延迟的有效方法．在并行科学计算应用的Ｉ／Ｏ访问模式下，简单的Ｃａｃｈｅ和预取技术已无法提供较高的Ｃａｃｈｅ 命中率．该文在分析该Ｉ／Ｏ模式的基础上提出了适度贪婪的Ｃａｃｈｅ 和预取一体化算法（ＰＧＩ）．该算法充分利用了并行文件系统环境的特点，采用了适度贪婪的动态滑窗技术，可以有效地消除预取时的抖动，降低系统处理开销；并同时采用了Ｃａｃｈｅ 和预取一体化的淘汰损失估计算法，使淘汰的损失降到最低点，在整体上提供了较短的Ｉ／Ｏ 服务时间 Cache and prefetch in traditional file systems are two effective ways to reduce latency. In parallel I / O access for scientific computing applications, simple Cache and prefetch have failed to provide high Cache hit rates. Based on the analysis of the I / O mode, this paper proposes a moderately greedy Cache and Prefetch Integrated Algorithm (PGI). This algorithm takes full advantage of the parallel file system environment and adopts a graceful and greedy dynamic sliding window technology, which can effectively eliminate the jitter during prefetching and reduce the processing overhead of the system. At the same time, the algorithm uses the combination of Cache and prefetch to eliminate the loss The estimation algorithm minimizes the loss of obsolescence and provides a shorter overall I / O service time