Sixteenth International Workshop on Accelerating Analytics and Data Management Systems Using Modern Processor and Storage Architectures in conjunction with VLDB 2025, London, U.K. Monday, September 1, 2025.
Christos Laspias, Andrew Pavlo, Jignesh M. Patel
Carnegie Mellon University
http://www.pdl.cmu.edu/
For as long as database management systems (DBMSs) have existed, there have been efforts to develop specialized hardware to accelerate their workloads. The goal is clear: to offload the DBMS’s most common and repetitive tasks to hardware, thereby improving efficiency and performance. Recently, Intel has released CPUs with new accelerators located on the same die, such as the In-Memory Analytics Accelerator (IAA) that targets data processing tasks. In this work, we examine the Intel IAA’s ability to optimize data compression and decompression operations for online analytical processing (OLAP) workloads. To evaluate the benefits of this accelerator, we added support for IAA compression into DuckDB. Our experiments comparing IAA with DuckDB’s existing compression method (Snappy) show that it improves decompression speeds by up to 3.15× in microbenchmarks and the end-to-end TPC-H query latencies by up to 38%.
FULL PAPER: pdf