A new approach for directory management in GlusterFS

Distributed filesystems provide possibility of storing and sharing large amount of data. They also provide an environment for users to store and maintain their data simply. Unlike most distributed filesystems, GlusterFS stores and manages data and metadata together. GlusterFS is a highly scalable filesystem that provides access to a large number of clients. One of the problems that GlusterFS faces is metadata operations that can be considered as a result of directory metadata distribution among all servers. The purpose of this distribution is to provide quick and easy access to data. In this study we proposed a new indexing method to improve directory management and operation. We used incremental directory distribution to avoid small directory distributing. Experiment results demonstrate improvement of locality and decreasing of sequential access to directory entries about 17%. GlusterFS Rebalance operation for adding one server to filesystem is very costly. Our method reduced this cost approximately 74% in average.