Autonomous Transaction Managers in Responsive Computing

Abstract

Transaction systems are designed to facilitate the design of applications in distributed, failure-prone environments. Typically, mechanisms are provided as part of the transaction management system to relieve the application designers of managing concurrent executions and unexpected failures. Additionally, providing adequate real-time responsiveness in such systems necessitates the distributed autonomous management of the transaction executions — due to the reason that a centralized approach would be inefficient and highly sensitive to the failure of a single site. Ensuring the logical correctness of concurrent transaction executions in terms of maintaining serializability in an environment consisting of several autonomous sites is a current research effort. While several schemes have been proposed that are sufficient to ensure serializability, the particular conditions necessary have not been made precise. We explore this question in an attempt to delineate those classes of concurrency control protocols that can be used at each local site in the context of autonomous transaction managers. Our research provides the characterization and techniques for integrating the transaction schedules arising from the autonomous sites in a manner similar to the approach used for centralized concurrency control.