pNFS for Everyone: An Empirical Study of a Low-cost, Highly Scalable Networked Storage

This paper presents empirical experience on a scalable, high-performance storage with two existing low-cost technologies. One technology is Ethernet-Attached Disks (E-AD) which are iSCSI-like disks but with low cost controllers. Another is Network File System (NFS) which is an open standard supported through open source. In particular, NFS version 4.0 (NFSv4.0) has been the de-facto distributed file system protocol for generations. Unfortunately, in this age of mass data, scalability of NFSv4.0 has become an issue. The NFS community has developed the pNFS protocol that takes advantage of parallel accesses to the pool of storage, and announced NFS version 4.1 including the pNFS protocol. We evaluated the scalability and performance of our storage system under various environments and benchmarks. The results show that NFSv4.1 has substantial benefits on large I/O units due to direct disk access and parallelismclos