Reducing the Load of Metadata Server by Changing Cache Policy Dynamically in Distributed File System

A distributed file system is a scalable and reliable storage system. However, it has a problem that performance of the system is limited because a large number of metadata operations are concentrated in a single Metadata Server (MDS). To solve this problem, it is effective to cache metadata in client side memory but, consistency of the metadata cache needs to be guaranteed. There are mainly two methods to guarantee the consistency: MDS oriented and client oriented. However, in both cases, the performance of the MDS degrades due to intensive control messages in certain access patterns. In this paper, we propose a hybrid method to apply the two cache policies separately to each file and change the cache policy dynamically depending on the access patterns. An experimental measurement showed that the proposed method can reduce the load of the MDS by alleviating control messages, and can improve the performance of the system.