Computing Multidimensional Aggregates in Parallel

Abstract

Computing multiple related group-bys and aggregates is one of the core operations of On-Line Analytical Processing (OLAP) applications. This kind of computation involves a huge volume of data operations (megabytes or treabytes). The response time for such applications is crucial, so, using parallel processing techniques to handle such computation is inevitable. In this paper we present several parallel algorithms for computing a collection of group-by aggregations based on a multiprocessor system with sharing disks. We focus on a special case of the aggregation problem-'Cube' operator which computes group-by aggregations over all possible combinations of a list of attributes. The proposed algorithms introduce a novel processor scheduling policy and a non-trivial decomposition approach for the problem in the parallel environment. Particularly, we believe the proposed hybrid algorithm has the best performance potential among the four proposed algorithms. All the proposed algorithms are scalable. Show more