The design and implementation of a user-oriented replicated file system for UNIX

Replication is a way to protect files from becoming inaccessible or destroyed by hardware and/or software failures. Several algorithms have been proposed to maintain the consistency of replicated files, but few authors consider the implementation details of these algorithms. The absence of guidelines for the implementation of a replicated file system (as well as other factors) makes it difficult for the average academic Unix user to exploit file replication.

This thesis identifies the major design issues found during the implementation of a replicated file system for Unix. Besides providing some useful guidelines for the design of such system, the main result is a set of portable, ready to use, C Shell scripts (plus a user manual), that allow any group of Unix users to replicate, and share, any number of files and/or subdirectories.

Static weighted voting with no read quorums, and the optional use of witnesses, is used to maintain the consistency of the replicated files. Replication is done transparently and at the file level. In other words, the Intervention of the system manager or any special privileges are not required, and the user does not have to modify the commands, editors, application programs, etc. to use replicated files.