Postgres-R(SI): combining replica control with concurrency control based on snapshot isolation

Replicating data over a cluster of workstations is a powerful tool to increase performance, and provide fault-tolerance for demanding database applications. The big challenge in such systems is to combine replica control (keeping the copies consistent) with concurrency control. Most of the research so far has focused on providing the traditional correctness criteria serializability. However, more and more database systems, e.g., Oracle and PostgreSQL, use multi-version concurrency control providing the isolation level snapshot isolation. In this paper, we present Postgres-R(SI), an extension of PostgreSQL offering transparent replication. Our replication tool is designed to work smoothly with PostgreSQL's concurrency control providing snapshot isolation for the entire replicated system. We present a detailed description of the replica control algorithm, and how it is combined with PostgreSQL's concurrency control component. Furthermore, we discuss some challenges we encountered when implementing the protocol. Our performance analysis based on the TPC-W benchmark shows that this approach exhibits excellent performance for real-life applications even if they are update intensive.