Cheetah: Accelerating Database Queries with Switch Pruning

Modern database systems are growing increasingly distributed and struggle to reduce the query completion time with a large volume of data. In this poster, we propose to leverage programmable switches in the network to offload part of the query computation to the switch. While switches provide high performance, they also have many resource and programming constraints that make it hard to implement diverse database queries. To fit in these constraints, we introduce the concept of data pruning - filtering out entries which are guaranteed not to affect the output. The database system then runs the same query, but on the pruned data, which significantly reduces the processing time. We propose a set of pruning algorithms for a variety of queries. We implement our system, Cheetah, on a Barefoot Tofino switch and Spark. Our evaluation on the Berkeley AMPLab benchmark shows up to 3x improvement in the query completion time compared to Apache Spark.