Semantics-Based Transaction Processing for Real-Time Databases: The Case of Automated Stock Trading

Abstract

In the future, automated stock trading (AST) databases will need to support execution of transactions with time constraints similar to that discussed in real-time databases. Transactions in AST databases repeat from a predictable set of transaction classes where the logical read-sets and write-sets of these classes can be established a priori. We propose a novel transaction management mechanism that exploits this a priori knowledge of data requirements to minimize wasted resources and transaction restarts inherent in existing real-time transaction processing mechanisms based on dynamic acquisition of resources. Preanalysis provides a mechanism to incorporate transaction semantics, identify potentially detrimental conflicts, and induce favorable serialization orders before transactions execute. By blocking potentially conflicting transactions, we incorporate an implicit overload management technique. Scheduling decisions can be made efficient to complement serialization orders, thus reducing unnecessary transaction restarts. Our protocol is evaluated using simulation and is shown to outperform existing real-time concurrency control mechanisms under various workload and resource parameters.