Dynamic distributed query processing techniques

Abstract

As optimization of strategies to process queries in a Distributed DataBase (DDB) uses various techniques to estimate the sizes of partial results and other parameters pertaining to the distributed environment, if these estimates are inaccurate the strategies may be far from optimal. Dynamic query execution, which may be used to alleviate this problem, is examined in this paper. Execution of a strategy is assumed to proceed through three phases: (i) monitoring phase in which processors monitor the progress of the strategy execution; (ii) decision making phase in which they may decide to correct the current strategy because it is not optimal due to inaccurate estimates used in its formulation; and (iii) corrective phase in which the current strategy is aborted and a new, corrective strategy is initiated. Methods applicable to each phase and their integration is examined in detail in terms of overhead, complexity and accuracy of information used in correcting a strategy.