DPAgg: A Dynamic Partition Aggregation on Multicore Processor in Main-Memory Database

Multicore processor provides an opportunity of query performance enhancement in database by parallel processing. However, the increasing of cores and computing performance of multicore processor can´t be utilized efficiently because of the bottleneck of memory access. It is known to all that aggregation is one of the most important operators in database. In this paper, we present a new dynamic partition aggregation algorithm. It´s a group by value based partition algorithm like most the state-of-art parallel aggregation algorithms. First, the algorithm introduces dynamic partition strategy into aggregation in order to make it more adaptive to different characteristic data. Second, it sets the optimized cache level to L2 not last level cache. Third, the improvements that are proposed for Radix-Clustered hash join are also been modified and applied into aggregation in order to reduce overhead cost of partition operations. Our experiments show the superiority of our dynamic partition aggregation algorithm on multicore processor.