Analysis of a dynamic query optimization technique for multijoin queries

Query processing in parallel data base systems is commonly based on the assumption that reliable estimates can be made for the sizes of intermediate results and the load distribution in the system. However, these estimates are difficult to make for nonstandard applications. A solution is to use a dynamic query-processing scheme, which is adaptive to the data and load distribution in a shared-nothing architecture. We propose a novel run-time optimization technique, called task elimination, aimed at reducing the total amount of work for queries over large partitioned data bases. The level of reduction obtained and the total processing cost is determined with a probabilistic model. Finally, we introduce the parallel bottom-up query evaluation strategy, which maximizes the effect of the task elimination optimization technique.