A data priority reload technique for real-time main memory databases

Abstract

A data priority reload algorithm is proposed for real time main memory database systems that allows transaction processing to be resumed before the entire database is recovered from a system failure. Transaction execution priority, reload priority and reload preemption are taken into account during the reload process; this enables transactions with high execution priorities to have more opportunities to meet their deadlines and thus enhances the overall system performance. This algorithm also gives temporal data a higher reload priority than persistent data so that many temporal data can be used before losing their validity, which in turn minimizes the number of transactions aborted or delayed due to invalid data access. Based on the simulation results obtained, we can draw the following conclusions: the data priority reload algorithm provides a significant performance improvement over the conventional reload approach; whether the efficiency of different reload algorithms is crucial to the overall performance mainly depends on the system load, database size and system failure rate; the key factors of the design of real time MMDB reload algorithms are system unavailability, transaction execution priority, and reload threshold. In order to derive an efficient reload scheme, these factors must be studied carefully.