The performance of replica control protocols in the presence of site failures

Although managing multiple copies of a database has been the subject of intensive research for quite some time now, it has yet to fulfill its promise in practical applications. In the current state of distributed database technology, data replication, if implemented at all, is typically enforced by the read-one-write-all protocol. More complicated but less restrictive replica control protocols, though a popular topic in research, are not implemented in any widely-used systems. A major reason for this lack of acceptance is that the performance impact of these protocols cannot be easily quantified, since very little existing performance figures of replicated commercial database systems are available. This paper investigates the overhead of data replication, and compares the performance of a seldom implemented protocol, quorum consensus, with the performance of a widely accepted protocol, primary copy