Triggered Updates for Temporal Consistency in Real-TimeDatabases

Abstract

A real-time database system has temporal consistency constraints in addition to timing constraints. The timing constraints require a transaction to be completed by a specified deadline, and the temporal consistency constraints require that temporal data read by a transaction be up-to-date. If a transaction reads out-of-date data, it will become temporally inconsistent. A real-time database system consists of different types of temporal data objects, including derived objects. The value of a derived object is computed from a set of other objects, known as the read-set of the derived object. The derived object may not always reflect the current state of its read-set; a derived object can become out-of-date even if its read-set is up-to-date. Any subsequent transaction reading the derived object will then become temporally inconsistent. In this case, in order to read up-to-date objects, a transaction will have to wait until some other transaction updates the out-of-date object. However, in doing so, the waiting transaction may miss its deadline, particularly if the update is not periodic but instead arrives randomly. We propose to update the outdated objects so that not only is the temporal consistency improved, but also the number of missed deadlines does not increase significantly, and as a result there is an overall improvement in the performance of the system. We propose, implement and study a novel approach, to be known as triggered updates, to improve temporal consistency in firm real-time database systems when updates are not periodic. We identify properties of triggered updates and explain how they work by giving both an intuitive and a probabilistic analysis. We present strategies for generating triggered updates, discuss their suitability in various contexts and perform a detailed simulation study to evaluate their performance. Results show that it is possible to improve temporal consistency without degrading the timeliness of real- time database systems to a great deal.