Pipelined Query Processing in Coprocessor Environments
Henning Funke, Sebastian Breß, Stefan Noll, Volker Markl, and Jens Teubner
Proc. of the 2018 ACM SIGMOD Conference on Management of Data, Houston, TX, USA, June 2018.
Query processing on GPU-style coprocessors is severely lim- ited by the movement of data. With teraflops of compute throughput in one device, even high-bandwidth memory can not provision enough data for a reasonable utilization. Query compilation is a proven technique to improve the memory efficiency. However, its inherent tuple-at-a-time pro- cessing style does not suit the massively parallel execution model of GPU-style coprocessors. This compromises the im- provements in efficiency offered by query compilation. In this paper, we present a new approach to query compilation that allows to extend the scope of operations that can be included in a single pipeline kernel. We show that our approach is bet- ter suited for GPU-style coprocessors than previous work that generates kernels only for coherent functionality. Compared to operator-at-a-time, we show reductions of memory access volumes by factors of up to 7.5x. This facilitates shorter kernel execution times by factors of up to 9.5x.
Energy Awareness in Database Algorithms and Systems (SFB 876, A2)