An Operator-Stream-Based Scheduling Engine for Effective GPU Coprocessing

Abstract

Since a decade, the database community researches opportunities to exploit graphics processing units to accelerate query processing. While the developed GPU algorithms often outperform their CPU counterparts, it is not beneficial to keep processing devices idle while over utilizing others. Therefore, an approach is needed that effectively distributes a workload on available co-processors while providing accurate performance estimations for the query optimizer. In this paper, we extend our hybrid query-processing engine with heuristics that optimize query processing for response time and throughput simultaneously via inter-device parallelism. Our empirical evaluation reveals that the new approach doubles the throughput compared to our previous solution and state-of-the-art approaches, because of nearly equal device utilization while preserving accurate performance estimations.