PERFORMANCE STUDY OF TWO PHASE LOCKING IN SINGLE-SITE DATABASE SYSTEMS.

Abstract

Queuing network models are used to study the problem for the preclaim and the claim-as-needed locking profiles. A wide variety of realistic conditions, such as different database-size/transaction-size ratios, and different granule placement assumptions, are considered. Using simple combinatorial arguments and elementary probability theory, the authors show that expressions for quantities such as the probability of a lock being granted, the probability of deadlock, and the mean wait time for a blocked request can all be derived analytically. The analysis is straightforward and the models are easy to understand. Validation of the models is done by comparing the models' predictions with simulation results, which shows very good agreement. Finally, the problem for a claim-with-ordering locking policy is studied. Because analytic modeling of this policy is difficult, a simulation model is used instead.