Quality-based replication of freshness-differentiated Web applications and their back-end databases

Replication is one standard way to meet demand and availability requirements of Internet content. Data-intensive web applications can support multiple service quality levels and may have varying levels of resource needs, data accuracy requirements, and synchronization overheads. Such database-driven applications (DAs) typically consist of a combination of application logic and large back-end databases. Application service providers (ASPs) seek to replicate these quality-differentiated applications on a large network of heterogeneous servers in such a way that response quality and demand are satisfied while minimizing the database synchronization overhead. This NP-hard optimization problem is the Quality-Sensitive DA Replication Problem (QDAREP). This work investigates QDAREP and develops and tests several replication algorithms for its solution.; Presented first is special version of QDAREP in which DAs have just a single quality level. A replication strategy is proposed and is shown to fairly distribute request load to servers and to be robust to changes in user request patterns.; Next, algorithms for solving static and dynamic instances of QDA REP are proposed and evaluated. The static Update Burden Bounding (UBB) algorithms assign replicas to servers by predicting and minimizing update load growth. A dynamic algorithm, DYN, adjusts replica-to-server assignments in response to changing user request loads while ensuring that assignment transitions are deadlock-free. The static and dynamic algorithms are effective and relatively quick.; The ant colony optimization meta-heuristic is adapted to solve QDA REP. Results of an initial investigation of the performance of the subsequent algorithm, AntDA, are reported and plans for further research into this promising approach are given.; Also presented are two software programs developed as part of the dissertation. The Database Assignment Test Engine (DATE) implements the static and dynamic assignment algorithms. The Database Application Simulation (DASIM) is a prototype DA replication environment.