Measuring the effect of commutative transactions on distributed database performance

Abstract In this paper, we consider algorithms that permit every site in a partitioned system to perform new updates. Because this may result in independent updates to items in different partitions, conflicts among transactions are bound to occur. Commutative transactions can be used to reduce the number of these conflicts. We develop a probabilistic model to estimate the possible reduction in conflicts if we incorporate this notion of commutativity. Our results show that the additional efforts to recognize this commutativity are not beneficial unless the number of transactions that commute with each other is significantly large.