GOLAP Accepted to SIGMOD'25
2024/10/24
GOLAP: A GPU-Accelerated system for high-speed analytics on SSD-resident data
Authors: Nils Boeschen, Tobias Ziegler, Carsten Binnig
We are happy to announce that GOLAP has been accepted SIGMOD'25!
The paper proposes a GPU-in-data-path architecture to accelerate OLAP queries by using GPUs to decompress data streamed directly from SSDs during table scans. This approach allows near in-memory bandwidth performance using SSDs and includes novel GPU-optimized pruning techniques to further boost read bandwidth. The system achieves up to 100 GiB/s in effective bandwidth, surpassing current in-memory systems.