Transaction processing in real-time database systems

Abstract

Scheduling transactions in a real-time database requires an integrated approach in which the schedule does not only guarantee execution before the deadline, but also maintains data consistency. The problem has been studied under a common framework which considers both concurrency control issues and the real-time constraints in centralized and distributed transaction processing. A real-time transaction processing model has been defined for a centralized system. The proposed protocols use a unified approach to maximize concurrency while meeting real-time constraints at the same time. In order to test the behavior of the model and the proposed protocols, a real-time transaction processing testbed has been developed using discrete event simulation techniques. The results indicate that different protocols work better under different load scenarios and that the overall performance can be significantly enhanced by modifying the underlying system configuration. Among other system and transaction parameters, the effect of data partitioning, buffer management, preemption, disk contention, locking mode and multiprocessing has been studied. For the distributed environment, new concepts of real-time nested transactions and priority propagation have been proposed. Real-time nested transactions incorporate the deadline requirements in the hierarchical structure of nested transactions. Priority propagation addresses the issues related to transaction aborts in real-time nested transaction processing. The notion of priority ceiling has been used to avoid the priority inversion problem. The proposed protocols exhibit freedom from deadlock and have tightly bounded waiting period. Both of these properties make them very suitable for distributed real-time transaction processing environment.