Hardware-conscious Query Processing in GPU-accelerated Analytical Engines

In order to improve their power efficiency and computational capacity, modern servers are adopting hardware accelerators, especially GPUs. Modern analytical DMBS engines have been highly optimized for multi-core multi-CPU query execution, but lack the necessary abstractions to support concurrent hardware-conscious query execution over multiple heterogeneous devices and, thus, are unable to take full advantage of the available accelerators. In this work, we present a Heterogeneity-conscious Analytical query Processing Engine (HAPE), a hardware-conscious analytical engines that targets efficient concurrent multi-CPU multi-GPU query execution. HAPE decomposes heterogeneous query execution into i) efficient single-device and ii) concurrent multi-device query execution. It uses hardware-conscious algorithms designed for single-device execution and combines them into efficient intra-device hardware-conscious execution modules, via code generation. HAPE combines these modules to achieve concurrent multi-device execution by handling data and control transfers. We validate our design by building a prototype and evaluate its performance on a co-processing radix-join and TPC-H queries. We show that it achieves up to 10x and 3.5x speed-up on the join against CPU and GPU alternatives and 1.6x-8x against state-of-the-art CPU- and GPU-based commercial DBMS on the queries.


Published in:
Proceesings of the 9th Biennial Conference on Innovative Data Systems Research
Presented at:
9th Biennial Conference on Innovative Data Systems Research, Asilomar, California, USA, January 13-16, 2019
Year:
2019
Laboratories:




 Record created 2018-12-18, last modified 2019-03-17

Final:
Download fulltext
PDF

Rate this document:

Rate this document:
1
2
3
 
(Not yet reviewed)