Unser Artikel "Pipelined Query Processing in Coprocessor Environments" wurde zur Präsentation/Publikation auf der SIGMOD 2018 angenommen. Die SIGMOD findet statt im Juni 2018 in Houston, TX, USA.
Our paper "Pipelined Query Processing in Coprocessor Environments", co-authored by Henning Funke, Sebastian Breß (DFKI Berlin), Stefan Noll (ICD/SAP), Volker Markl (TU Berlin), and Jens Teubner, was accepted for publication/presentation at the ACM SIGMOD Int'l Conference on Management of Data 2018. SIGMOD is a premier venue for world-leading research in database technology. The 2018 edition takes place June 10-15, 2018, in Houston, TX, USA.
The paper discusses a compilation mechanism that converts database query plans into code executable on graphics processors (GPUs). A particular challenge when using GPUs for database acceleration is the limited bandwidth between host and device, but also between processor and memory on the device itself. HorseCQ, the compiler presented in the paper, is carefully optimized for bandwidth efficiency and also takes features of latest GPU hardware into consideration when generating GPU code.