Performance analysis of optimistic concurrency control schemes for system with large memory

Abstract

The authors examine the impact of a large buffer memory on the comparative performance of different Optimistic Concurrency Control (OCC) schemes. They develop analytic models to analyze the performance of different schemes. The modelling partitions the analysis into two interacting components: hardware resource contention and data contention. Queueing models are used to estimate the resource contention effects. The data contention results in transaction abort, leading to resource consumption for transaction rerun, and consequently higher resource contention, leading in turn to higher data contention. Such interaction between the data and resource contention is captured using an iteration. The accuracy of the analysis is validated through simulations.