Abstract

The widespread use of clusters and web farms has increased the importance of data replication. In existing protocols, typical distributed system solutions emphasize fault tolerance at the price of performance while database solutions emphasize performance at the price of consistency. In this paper, we explore the use of data replication in a cluster configuration with the objective of providing both fault tolerance and good performance without compromising consistency. We do this by combining transactional concurrency control with group communication primitives. In our approach, transactions are executed at only one site so that not all nodes incur in the overhead of parsing, optimizing, and producing results. To further reduce latency, we use an optimistic multicast approach that overlaps transaction execution with the total order message delivery. The techniques we present in the paper provide correct executions while minimizing overhead and providing higher scalability.

Citations