Modular Adaptive Query Processing for Service-Based Gridsa

Distributed and heterogeneous environments present significant challenges to complex software systems, which must operate in the context of continuously changing loads, with partial or out-of-date information on resource capabilities. A distributed query processor (DQP) can be used to access and integrate data from distributed sources, as well as for combining data access with data analysis. However, in heterogeneous environments, statically constructed query plans may commit a query evaluator to following significantly suboptimal strategies. As such, there is considerable interest in using adaptive query processors (AQPs) in such settings to provide self-optimizing behaviour. However, with many possible adaptive strategies available, it is important that AQPs can be constructed in a systematic and efficient manner. This paper presents an approach to the development of AQPs in which adaptive behaviour is implemented using cooperating monitoring, assessment and response components. It is shown how this decomposition has been applied in the development of an adaptive DQP system for service-based grids, which reallocates load at query runtime, thereby supporting self-optimization. Experimental results for query processing in wide-area grids show that the approach can be effective in practice, with queries making local decisions to change their behaviour in the light of changing resource throughput.