Content area
Abstract
With the prevalence of database application and the widespread availability of communication networks, distributed databases systems remains an important topic in the 1990's. While distributed database systems have been the subject of extensive research, their acceptance in the industry remains tenuous. A key reason for this phenomenon is that for such a system to function properly, it must perform correctly in the presence of communication and site failures; that is, it must be fault-tolerant. Achieving fault-tolerance in such a system is non-trivial, as the implementation of system protocols in notoriously difficult, especially if failures are considered. Another reason for the lack of acceptance of distributed database systems is the difficulty involved in assessing the performance impact of the protocols employed in such systems, especially in the presence of failures: the is a myriad of factors to consider, and yet very little published performance data is available.
This dissertation makes two contributions in the area of designing distributed database systems. Our first contribution is the development of a simulation testbed for a distributed database system which encompasses both the details of the protocols and failures in the system. Using this testbed, we investigated the performance of two major classes of protocols in a distributed database system: atomic commitment protocols and replica control protocols. The second contribution is the identification of efficient fault-tolerance protocols for these two classes of protocols.