【BAYES COFFEE HOUSE TECH TALK SERIES】System Architectures for Databases on Modern Hardware
Jens Teubner will give a talk, in person and online, for the Coffee House Tech Talk Series. Details of the talk are below. A lunch will be provided after the talk.
Title: System Architectures for Databases on Modern Hardware
Speaker: Jens Teubner
When: 11am Wed 13 Sep 2023
Where: 4th floor Bayes Centre, Coffee House
External link: https://meeting.huaweicloud.com/welink/j/97940104/VrjdBiUFmCrzzjnYf7cidF64B72Q1XH9X
Meeting ID: 97940104
The hardware in modern computing systems has changed significantly over the years: many cores, multiple sockets, large memories, and
heterogeneous architectures have become a commodity. We argue that it is about time to reflect these changes in the way how we architect (database) software. In my talk, I will illustrate our vision of a novel architecture that is ready for the changing hardware landscape.
One key ingredient is a new abstraction for control flow. Specifically, with MxTasking we propose a processing model that is no longer based on classical ‘threads’. Instead, we introduce ‘tasks’ as an abstraction that describes a unit of work (whereas threads describe a sequence of code). Most importantly, the abstraction opens the way to enrich tasks with annotations. Annotations, in turn, enable a lightweight, expressive, and effective interaction between hardware, scheduling, and application logic. In the talk, I will illustrate how this eases the development of “hardware-conscious algorithms”, while better leveraging the features of modern hardware (e.g., through automatic data prefetching) and improving performance.
MxTasking goes particularly well with just-in-time compilation, an idea that has become popular in the database domain during the past years. With ‘Flounder’, we propose a framework that provides best-in-class compilation speeds while keeping code quality high. Unlike previous settings, Flounder can benefit from specifics of an application domain and turn such knowledge into faster compilation and better execution speeds.
Jens Teubner is the head of the Databases and Information Systems Group at TU Dortmund since April 2013. Previously, he held positions as a postdoc at the IBM T.J. Watson Research Lab in New York, USA and as a senior researcher (“Oberassistent”) at the Systems Group of ETH Zürich. He holds a PhD in Computer Science from TU München (graduated in 2006 under the guidance of Torsten Grust; topic: XML query processing) and a Masters degree (“Diplom”) in physics from the University of Konstanz in Germany (advisor: Ernst Bucher).
The focus of Jens Teubner’s research is the implementation of database systems on modern hardware architectures. He is known for his work on FPGA-accelerated data processing and on hardware-conscious database algorithms. His current work includes query compilation for heterogeneous hardware, database/operating system co-design, and scientific data management. During his PhD, Jens Teubner was concerned with scalable evaluation techniques for XML, in particular by using relational database technology as a back-end. The outcome of this work—the XQuery compiler Pathfinder—has become known as part of the open source system MonetDB/XQuery.