Performance analysis on distributed storage systems in ring networks

摘要

Distributed storage system can provide reliable service of data storage. Lots of applications, such as data center, P2P network, now use distributed storage systems. It's important to guarantee the reliability of data in these systems. Compared with simple replication, erasure coding can reduce storage space and repair bandwidth when a node in the system fails, and are thus intensively investigated. In reality, the communication between two nodes in the system may be restricted by some physical limitations, such as distance, or economic considerations, etc. In this paper, we propose a new model to account for the connectivity limitation. We consider a ring network where nodes can only communicate to neighbor nodes in a given range. Inspired by the concept of regenerating codes [1], we investigate the theoretical necessary condition of the existence of coding scheme. A trade-off between storage and repair bandwidth is derived. Two coding schemes are studied. One is minimum-storage regenerating (MSR) code, and the other is minimum-bandwidth regenerating (MBR) code. Two applications, corresponding to the coding schemes, are also presented in the paper.