一种机群文件系统的缓存模型

A Caching Model in Cluster File System

提出了一种新的机群文件系统缓存模型 .它充分利用机群系统累积的系统资源和高速的互联网络 ,将文件系统元数据和内容数据分离 ,分别使用分布式元数据缓存和统一缓存模型进行管理 .元数据缓存使用改进的广播一致性协议和 L RU替换算法 .内容数据统一缓存则将磁盘缓存整个文件和内存缓存文件块相结合组成一个单一映像的多层次分布协作缓存 ,并使用单拷贝优先 L RU和向前传递调度缓存块替换算法以及一种贪心的数据预取方法 .实验结果表明 。

The different characteristics of metadata and content in cluster file system motivate us to employs distinct caching mechanics to manage them. What is more, cluster of workstation provides accumulative system resources. Based on the analysis of those facts, the design and implementation of a distributed metadata caching model and a content data unique caching system are presented in the paper. The metadata caching uses a modified broadcasting caching coherence protocol and LRU caching block replacing algorithm. While the unique caching model cooperates whole file disk caching with dada block memory caching efficiently, using unique copy prior and forward algorithm to replace caching blocks and a greedy strategy to prefetch caching data. According to the experiment result, we conclude that metadata caching and unique caching can improve the cluster file system performance substantially.