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.