Investigatiing the performance scalability & reliability of a distributed file system: Ceph

Abstract

File system is the very crucial part of all computer systems. It has been improved from local to network and now to distributed file systems. NFS, SAN, and NAS are becoming obsolete as they all pursue the traditional block based storage systems and impose the whole work on a single server. The recently innovated DFSs pursue object-based storage system which helps them to decouple metadata from actual data and hence avoid single point of failure and increase performance, scalability, and reliability. The recently added to Linux DFSs family is a POSIX compatible software called Ceph. It has developed and implemented some special concepts and unique features in its architecture (dynamic metadata subtree partitioning, intelligent objectbased storage, data replication across nodes, fault tolerance, assuming nodes failure as a norm rather than exception etc). This research paper investigates and evaluates the author’s claim that ceph is capable of handling more than petabyte-scale of storage in heterogeneous systems with excellent performance, high scalability and high reliability