In this project we aim to build a task-based runtime for in-memory query processing that makes best use of modern hardware. This will be achieved by either circumventing OS mechanism that stand in the way of good performance and by investigating hybrid runtimes which are applications (a task-based in-memory DBMS runtime in our case) that are combined with a light-weight OS kernels (such as http://cs.iit.edu/~khale/nautilus/). An advantage of hybrid runtimes is that there is no distinction between kernel and userspace (and, thus, no context switching between kernel and application). Furthermore, the application has direct access to the underlying hardware where this is needed for performance.
- Kyle Hale - Illinois Institute of Technology
Playing Fetch with CAT - Composing Cache Partitioning and Prefetching for Task-based Query Processing
Qitian Zeng, Kyle Hale and Boris Glavic
International Workshop on Data Management on New Hardware (2021).