A script-based prototype for dynamic deadlock avoidance

Abstract

Expert systems apply Artificial Intelligence (AI) techniques to an application area, aiming (usually) to mimic the behaviour of a human expert. However, there are some AI techniques which can be used to improve the internal performance of an existing application, not necessarily currently performed by a human. In this paper, we present further research and results of EAGLE (External Advisor for Granting Locks Expertly), an expert system advisor for the lock manager in a database system. By matching lock event sequences received from the lock manager against stored scripted deadlock sequences, EAGLE is able to identify unfolding deadlock sequences. By using this Dynamic Deadlock Avoidance (DDA) approach, EAGLE is able to avoid deadlocks before they occur. Currently, no ideal solution exists to the deadlock problem. Solutions vary in terms of the number of waits for access to resources and the number of deadlock occurrences. By utilising AI techniques, DDA offers a new way of treating the deadlock problem. In this paper we describe the design of EAGLE and present the results, in terms of the number of waits and number of deadlock occurrences, occurring with DDA compared with Deadlock Detection and Resolution.