Efficient Analytical Query Processing on Multi-GPGPU Systems

EDIC candidacy exam
Exam president: Prof. Robert West
Thesis advisor: Prof. Anastasia Ailamaki
Co-examiner: Prof. George Candea

Abstract
Conventional wisdom suggests that handling the explosion in terms of data volume only requires scaling out modern database systems. Still, besides scaling out, database systems must consider how to fully exploit the abundant resources of modern servers, such as multiple cores and large memories. In addition, servers are becoming more and more heterogeneous and the processing power is usually scattered across multiple processing nodes, such as different CPU cores and sockets or even accelerators like GPGPUs. Analytical query processing has to overcome the challenges introduced by heterogeneity and use the available resources as efficiently as possible to be able to keep up with the endless increase of data volume.

This proposal discusses three existing works. The first two are aiming to efficiently process queries in a single-threaded setup and extend such an engine into a parallel one, respectively. The third work profiles a GPU database to give a better insight on where time goes when GPU databases process analytical queries. Finally, we conclude with our research proposal regarding analytical query processing in multi-GPU systems.

Background papers
Efficiently compiling efficient query plans for modern hardware, by Neumann, T. Proceedings of the VLDB Endowment 4, 9 (2011), 539-550.
Encapsulation of parallelism in the volcano query processing system, by Graefe, G. In Proceedings of the 1990 ACM SIGMOD International Conference on Management of Data (New York, NY, USA, 1990), SIGMOD ’90, ACM, pp. 102-111.
The yin and yang of processing data warehousing queries on gpu devices, by Yuan, Y., Lee, R., and Zhang, X.  Proceedings of the VLDB Endowment 6, 10 (2013), 817-828.