FILE REPLICATION IN DISTRIBUTED SYSTEMS (PERFORMANCE ANALYSIS, SITE RECOVERY)
Abstract
This thesis studies the problem of file replication in distributed systems. File replication is desirable for improving availability and reliability. However, there are pros and cons from the view of performance. The efficiency of a distributed file system depends highly on the replication technique used. This thesis develops a systematic method for modeling and analysis of replicated distributed file systems, and investigates several different file replication mechanisms, namely, remote access (no replication), prereplication (RAWA), weighted voting, and two variants of demand replication, polling and staling. Under realistic assumptions, our results show that the demand replication schemes provide the most efficient file access. This thesis then studies the replication control problems. It develops the logical-serializability theory, and shows that polling and staling, as well as RAWA and weighted voting, are correct replication schemes in the sense that replicated files controlled by these schemes can behave just as single-copied files in supporting transactions. The thesis also discusses the replication control problem in the presence of site failures, and proposes a new scheme, RAWAA, to improve availability. This thesis also reports on the prototype of a distributed file and directory system, IBIS, implemented on top of UNIX. The design decisions and implementation issues in supporting both remote access and demand replication are presented.
Degree
Ph.D.
Subject Area
Computer science
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.