蓝鲸分布式文件系统的客户端元数据缓存模型

The Model of Meta-data Cache on Client Node in Blue Whale Distributed File System

在蓝鲸分布式文件系统中,客户端的所有无数据操作都是通过远程过程调用由元数据服务器完成,所有数据读写都是直接与存储服务器交换完成的。由于通信延迟,在客户端进行频繁数据读写时,元数据信息交换影响了整个系统的性能。我们设计了一种在客户端尽量缓存文件元数据信息的模型,有效地减少了元数据通信,缩短了整个读写过程的延迟,极大地提高了蓝鲸分布式文件系统的性能。

In Blue Whale distributed file system, the meta-data server handles all the meta-data through remote procedure calls, and all the data flows directly between clients and storage nodes. Because of communication delay, the metadata exchanging between clients and servers has great impact on the whole system performance especially when the load is heavy. We propose a model to maximize the meta-data cache on client node that dramatically reduces the RPC messages when itreads and writes data. This model greatly improves the reading and writing performance of Blue Whale distributed file system.