A Replica ControlMethod for Improving Availability for Read-only Transaction

Abstract

Data replication is often considered in distributed database systems to enhance availability and performance. The benefit of data replication, however, can only be realized at the cost of maintaining the consistency of data. In particular, network partition failures make it more difficult to achieve high data availability while ensuring strong correctness criteria such as 1-copy serializability. In this paper, we propose a replica control method to improve the availability of data in the presence of network partition failures. Our method extends the traditional primary copy method by using the relaxed correctness criterion called insular consistency for large-scale distributed systems, where partition failures frequently occur. We focus on increasing the availability of data for read-only transactions. We introduce a version vector as a tool for guaranteeing insular consistency and present a mechanism that allows readonly transactions to be executed at any partition as long as the insular consistency is satisfied. An asynchronous update propagation mechanism is also employed to improve the performance of update operations. We also show that the proposed method is correct and give some performance considerations.