Abstract
An architecture is presented for a file service capable of supporting a very large number of users. It provides each user with a lifetime involvement with the file service while giving the appearance of a single homogeneous system. The architecture is based on the provision of an independent name space for each user and the use of hints for name resolution. File caching is supported and server callbacks are used to aid in maintaining cache consistency. Update-in-place semantics are provided for those applications that required direct access to localized parts of the files. All file service metadata are updated in atomic and recoverable ways using write-ahead logging. The file service also supports user-level distributed atomic actions based on a two-phase commit synchronization protocol. It supports user and resource reassignments in a location-independent way by using the system's user location mechanism during file name resolution. File sharing information remains valid even when participants migrate.

This publication has 0 references indexed in Scilit: