Modeling and analysis of transaction processing systems

Abstract

In recent years, the demand for on-line transaction processing systems has grown rapidly with ever stringent performance requirements. In this paper, we examine several issues encountered in designing transaction processing systems and report some of the recent advancements in analytical performance modelling methodology on analyzing alternative design trade-offs. First of all, the Concurrency Control (CC) scheme employed can profoundly affect the performance of transaction processing systems. A general analytic modelling approach is presented that can be applied to analyze the various CC schemes under a unified framework, including locking, various optimistic schemes, and hybrid schemes. The analysis can capture the effect of skewed data access, different lock modes, variable length transactions and the buffer retention effect on rerun transactions. Next we consider the analysis of buffer hit probability. In a multi-node environment, whether in a cluster or client server environment, buffer coherency needs to be addressed. The cross invalidation phenomenon can have an adverse effect on the buffer hit probability. A general methodology to analyze various coherency control schemes is examined to predict the buffer hit probability. A hierarchical approach is used to decompose the modelling of transaction processing systems into three components: hardware resource, concurrency control and buffer models. The interaction among the components is then captured through a fixed point iteration.