IBM Scale Out File Services: Reinventing Network-Attached Storage
This paper focuses on the challenges of creating a scalable Network-Attached Storage (NAS) filer with hundreds of nodes on top of the IBM General Parallel File System (GPFS). The filer is created as a multiheaded data exporter over Ethernet and uses the Common Internet File System (CIFS) as implemented by the Samba open-source project and a Network File System (NFS) protocol server. Special focus is given to ensure the consistency of data in an efficient manner for applications that have complex access patterns from a large number of nodes. Also considered are the challenges of providing an extremely high-performance facility for exporting data in parallel and the demands of storing very large amounts of data.