Using Multicast to Reduce Deadlock in Replicated Databases
Abstract
Obtaining good performance from a distributed replicated
database that allows update transactions to originate at any site while
ensuring one-copy serializability is a challenge. A popular
analysis of deadlock probabilities in replicated databases shows that the
deadlock rate for the system is high and increases as the third power of
the number of replicas. We show how a replica management protocol
that uses atomic broadcast for replica update reduces the occurrence of
deadlocks and the dependency on the number of replicas. The rate
of deadlock is shown to be a function of both the number of sites as well
as the cost of broadcast. The analysis is confirmed by simulation
experiments.
Back