Stale replicas
•In many distributed processing situations, copies of data are kept at several sites
–e.g. to allow cheap/fast local reading
•If updates try to alter all replicas, they become very slow and expensive (they need two-phase commit, and they’ll abort if a remote site is unavailable!)
•So allow replicas to be out-of-date
•Lazy propagation of updates
–Easily managed by shipping the log across from time to time