A widely used computational model for constructing fault-tolerant distributed applications employs atomic transactions for controlling operations on persistent objects. There has been considerable work on data replication techniques for increasing the availability of persistent data that is manipulated under the control of transactions. Process groups with ordered group communications (process groups for short) has also emerged as a model for building available distributed applications. High service availability can be achieved by replicating the service state on multiple processes managed by a group communication infrastructure. These two models are often seen as rivals. This paper examines whether a distributed transaction system can profit from process groups for supporting replication of objects. A general model of distributed objects is used to investigate how objects can be replicated for availability using a system that supports transactions (but no process groups) and a system...
Mark C. Little, Santosh K. Shrivastava