Provisioning ZFS Pools On Lustre
TimeTuesday, July 3011:30am - 12pm
DescriptionWhile Lustre's parallelism, performance, and scalability make it desirable as a storage solution for clusters, its limitations prevent it from being suitable as a general purpose storage for all of a cluster's needs. In particular, Lustre's relatively poor performance for small, random I/O as well as metadata-intensive workloads make it less suitable for use as storage for users' home areas or working directories for compilation. In this paper, an experiment of deploying a ZFS file system using files pre-allocated on a Lustre file system as the ZFS storage pools is presented. While this adds many management options like snapshots and more fine grained controls over users and quotas, we focus on examining how adding this layer can affect the performance of suboptimal Lustre workloads. Benchmarking of the Lustre file system with and without the ZFS file system is performed, and the results are presented and analyzed.