Parallel Algorithms for Queries with Aggregate Functions in the Presence of Data Skew

摘要

Queries with aggregate functions often involve more than one relation and thus join is needed. In this paper, the effect of the sequence of aggregation and join operations is identified in a parallel processing environent. Three parallel methods of processing aggregate function queries are developed which mainly differ in the selection of partitioning attribute. Their cost models that incorporate the effect of data skew are provided.