Transaction management mechanisms for active and real-time databases: A comprehensive protocol and a performance study

Active and real-time databases (ARTDB) have a variety of applications in electronic brokerages in ®nancial markets, stock trading, network management and manufacturing process control. Transaction processing (TP) in ARTDB is extremely complicated since transactions may trigger other real-time transactions to an arbitrary depth with various types of dependencies (coupling modes). Therefore, transaction processing must be cognizant of not only the time deadlines but also the types of semantic dependencies with other transactions. The con ̄ict resolution between two transactions cannot be considered in isolation since a€ecting one transaction may a€ect every other semantically dependent transaction. Similarly, transaction scheduling needs to be compatible with the concurrency control to avoid unnecessary restarts. In this paper we argue that transaction pre-analysis using the pre-declaration paradigm is an ecient mechanism to integrate the various issues of transaction processing such as concurrency control, scheduling, and semantic dependencies. The pre-analysis is possible since in many applications transactions repeat from a set of transaction classes, and the con ̄icts can be easily determined at a logical level by partitioning relations into mutually exclusive subset (e.g., by stock-id in ®nancial applications). We develop a pre-analysis based transaction processing mechanism called OCCWB. OCCWB is an extended optimistic concurrency control protocol with blocking that combines the bene®ts of both optimistic and lock based protocols. Such an approach also has an implicit overload management mechanism required in many applications. OCCWB consists of four phases, namely, transaction pre-analysis, serialization ordering, priority adjustment and priority wait. Our protocol is validated using simulation and is shown to outperform existing protocols under various workload and parameter settings. Ó 1998 Elsevier Science Inc. All rights reserved.