A time-cognizant dynamic crash recovery scheme suitable for distributed real-time main memory databases

Abstract

Rapid and efficient recovery in the event of site crash is very important for distributed real-time main memory database system. In this paper, the recovery correctness criteria of distributed real-time main memory databases are first given. Then, a time-cognizant dynamic crash recovery scheme (TCDCRS) based on log is presented. TCDCRS uses nonvolatile RAM as logging store and integrates the properties of partitioned logging, ephemeral logging and real-time logging in order to reduce the logging cost as possible during the normal running. During restart recovery after site crashes, a dynamic recovery method based on the classification recovery strategy, which supports concurrent of system services and recovery processing, is adopted to decrease the downtime to the most extent. Experiments and evaluations show that TCDCRS has better performances than traditional recovery schemes in two aspects: the missing deadlines ratio of transactions and the time of system denying services after crashes. © Springer-Verlag Berlin Heidelberg 2006.