Advances in hardware technology have turned the classical assumptions behind database systems upside-down. Large DRAMs, deep memory hierarchies, and the need to exploit heterogeneous hardware challenge the database field. At TU Dortmund University, we address these challenges with several projects that make database systems resource aware.
In the talk, I will report on two of our efforts. I will begin with early results on the use of FPGAs to accelerate the processing of WAH-encoded bitmap indexes. CPU-based systems need to compromise compression rates for decompression complexity, thereby wasting precious bandwidth. By off-loading to hardware, such compromises are no longer necessary. As a second example of resource awareness, I will demonstrate a mechanism to partition caches in multi-core systems. Our technique—meanwhile implemented in SAP HANA—requires little changes to the database software; increases query performance for important workloads; and improves robustness, but never causes negative effects to other queries.
This work has been supported by Deutsche Forschungsgemeinschaft (DFG).
The work on cache partitioning has been done jointly with SAP SE and ICD.
ED, lecture hall, Hörsalsvägen 11, EDIT trappa C, D och H
04 December, 2019, 14:00
04 December, 2019, 15:00