Maintaining logical and temporal consistency in time critical databases.

摘要

Whereas transaction processing under deadline constraints has been the main focus of prior real-time database research, in this thesis, we focus on issues raised by the presence of temporal data and replicated data. In particular, we study the problem of data consistency maintenance and transaction scheduling in real-time databases dealing with temporal data and replicated data.; Unlike traditional real-time databases, timing constraints of transactions operating with temporal data quite often have their origins in the temporal properties of the data. Such data is sampled periodically and usable only within certain time limits. It is a very important task of a real-time database to satisfy the temporal consistency of data. Therefore we have investigated two specific problems for real-time transactions dealing with temporal data: (a) given transactions with deadlines, and given the temporal properties of the data, how should transactions be processed so that they are not aborted for lack of temporally valid data; (b) how should deadlines and periods (for transactions that update the database) be assigned so that temporal validity of data is maintained while the load imposed by the transactions is minimized. To address problem (a) the concept of data-deadline is developed, and time cognizant transaction scheduling algorithms based on data-deadline, forced wait and similarity protocols are proposed. It is shown that these algorithms produce considerable performance improvement. To address problem (b) a novel approach, More-Less, is proposed. Our analysis and experiments show that More-Less can provide better schedulability and reduce update transaction workload while guaranteeing data timing constraints.; We then investigated the problem of replicated data consistency in distributed real-time databases. Data replication can help database systems meet the stringent temporal constraints of current real-time applications. In this thesis, we present MIRROR, a concurrency control protocol specifically designed for firm-deadline applications operating on replicated real-time databases. MIRROR augments the classical O2PL concurrency control protocol with a novel state-based real-time conflict resolution mechanism. Our performance studies show that MIRROR provides the best performance in both fully and partially replicated environments for real-time applications with moderate update frequencies.