Performance Models of Timestamp-Ordering Concurrency Control Algorithms in Distributed Databases

A distributed database (DDB) consists of copies of data files (usually redundant) geographically distributed and managed on a computer network. One important problem in DDB research is that of concurrency control. This paper develops a performance model of timestamp-ordering concurrency control algorithms in a DDB. The performance model consists of five components: input data collection, transaction processing model, communication subnetwork model, conflict model, and performance measures estimation. In this paper we describe the conflict model in detail. We first determine the probability of transaction restarts, the probability of transaction blocking, and the delay due to blocking for the basic timestamp-ordering algorithm. We then develop conflict models for variations of the basic algorithm. These conflict models are illustrated by numerical examples.