Towards Unifying Database Scheduling and Concurrency Control: A Frontier Approach

Abstract

Real-time databases have numerous applications, including commodities trading, military command and control, patient monitoring, air traffic control, and flexible manufacturing. All of these applications require that data be processed in such a way that the data and output of the database remains logically consistent. They also require that deadlines imposed on database operations are met (temporal consistency). Thus, the ultimate goal of scheduling and concurrency control for real-time databases is to maintain both temporal and logical consistency of the transactions and of the data. In situations, such as overload, where both temporal consistency and logical consistency cannot be met, a transaction scheduler may have to sacrifice one requirement for the other and make a trade-off between the two. For example, suppose that a transaction t read is reading some sensor data in a real-time database, and another transaction, t update , needs to update that data in order to maintain the temporal consistency of the data and/or the transaction. If t update were allowed to execute, the logical consistency of t read would be violated, but if t update were blocked, the temporal consistency of the sensor data or of t update could be violated. Thus, one requirement must be traded-off for the other.