The performance of database replication with group multicast

Replication with update-anywhere capability while maintaining global synchronization and isolation has long been thought impractical. Protocols have been proposed for distributed replicated databases that take advantage of atomic broadcast systems to simplify message passing and conflict resolution in hopes of making replication efficient. This paper presents performance measurements on a simulation of a replicated database using those protocols. The results show that with the proper group broadcast mechanism, replication with update-anywhere capability is indeed practical.