An Architecture for Guaranteeing Real-Time Databases Available Ceaselessly

Abstract

Real-time databases are widely applied in time-critical applications, such as autopilot systems, robot navigation, real-time monitoring, programmed stock trading, etc. These applications usually require real-time databases can provide 7times24times60times60s' ceaseless service weekly. However, real-time databases cannot completely avoid all kinds of failures. The traditional recovery-processing scheme cannot already satisfy the requirement for the above applications. In this paper, we present a fault tolerant architecture based on self-perceptive, self-diagnosing and self-adaptive recovery mechanisms. This architecture can prevent or delay failures effectively. On the basis of this architecture, we propose a novel prediction recovery scheme. The prediction recovery scheme allows the execution of transactions during recovery by providing predictive values of damaged data items to these transactions that need immediate access to those data items.