Database Replication: If You Must be Lazy, be Consistent

Abstract


Due to severe performance penalties associated with synchronous replication, there seems to be a significant interest in asynchronous replica management protocols. In the commercial arena, these protocols are referred to as lazy replication since database transactions are executed locally, and the effects of these transactions are incorporated lazily on remote database copies. Lazy protocols currently in use either do not guarantee consistency and serializability as needed by transactional semantics or they impose restrictions on placement of data and which data object can be updated. In this short paper we investigate an epidemic update protocol that guarantees consistency and serializability in spite of a write-anywhere capability. We conducted experiments on a detailed simulation of a distributed, replicated database to evaluate this protocol. Our results establish that the lazy but consistent approach is indeed viable for a variety of database environments and workloads.

 Back