On dynamic data allocation

Dynamic data allocation in distributed database systems automatically reallocates file copies on network nodes according to the alterations of both data utilization and topology. It is necessary to maintain high performance and data availability, especially for high diverse topologies. Dynamic data allocation is accomplished in three phases: performance assessment to decide reallocation; assignment generation to create new assignment; and data reorganization to realize the assignment. This dissertation contributes methodology on analysis and implementation. Strategies for reallocating data when network topology alterations are presented. Benefit functions are defined to evaluate assignment efficiently. An analytical approach to data assignment generation establishes the lower bound on algorithm performance since allocation is NP hard. Then, implemented heuristic approaches by test approximate the bound to generate sub-optimal assignments. A non-blocking strategy presented resolves any conflicts with ongoing normal database activities when reorganizing data. The proposed dependency test and schedule algorithms can implement the new data assignment in a parallel manner.