[Other Papers] [Briefing Package]
Title: An NFS Replication Hierarchy Speaker: Brent Callaghan, Sun Microsystems Abstract: NFS is well known as the lingua franca of file access protocols. It is installed, or is available for almost every computing platform. It is well known for its good performance, reliability in the face of network and server outages, and its scalability on large multiprocessor servers. While NFS is commonly used by small LAN-based workgroups to share data on a central server, it is now being used on wide area networks and even on the Internet to provide file access for a much larger audience. NFS version 3 implementations can transfer large blocks of data over TCP connections and new procedures like "readdirplus" can be used to obtain directory information efficiently and WebNFS clients can connect directly to a server through a packet filter firewall or application proxy like SOCKS. NFS clients can be configured to switch automatically from one NFS replica server to another. There is nothing special required of an NFS replica server other than the replica filesystems being consistent. An NFS replication hierarchy is a hierarchical configuration of NFS clients and servers that maintain replicated filesystems. Each member of the hierarchy accepts changes to its replica from a server higher in the tree and may accept changes for several replica filesystems each with its own source server. The presentation will show how the NFS protocol can make efficient use of network bandwidth in delivering changes down a hierarchy with provision for fault tolerance and recovery. A replication hierarchy can be implemented with existing NFS clients and servers.