Client-server computing paradigm has been successfully adopted as the basis for implementing distributed and object-oriented database systems. Most of the current implementations assume a single server handling requests from multiple clients, In this situation, the SWAT swn becomes a performance bottleneck. We concentrate on the performance implications of designing and implementing multiple page-server architectures. Three performance factors are studied in this thesis - workload characteristics (UNIFORM, ONE-HOTCOLD and Multi-HOTCOLD), declustering algorithm (range declustering, round-robin declustering and balanced declustering) and concurrency control algorithm (2PL, 2PL with wound-wait and optimistic). Both analytical modeling and discrete event simulation have been used to evaluat...[ Read more ]

Client-server computing paradigm has been successfully adopted as the basis for implementing distributed and object-oriented database systems. Most of the current implementations assume a single server handling requests from multiple clients, In this situation, the SWAT swn becomes a performance bottleneck. We concentrate on the performance implications of designing and implementing multiple page-server architectures. Three performance factors are studied in this thesis - workload characteristics (UNIFORM, ONE-HOTCOLD and Multi-HOTCOLD), declustering algorithm (range declustering, round-robin declustering and balanced declustering) and concurrency control algorithm (2PL, 2PL with wound-wait and optimistic). Both analytical modeling and discrete event simulation have been used to evaluate the performance of the system. We show that proper match between declustering algorithm and workload characteristics can improve the system performance. If workload characteristics is not known, the balanced declustering algorithm is a good choice. Results also show that concurrency control algorithms based on two phase locking perform the best in sustaining the performance of the multiple-page server systems.