Paper Title: Accelerating Database Caching and Query with SmartSSD CSDs

Paper Abstract: Greenplum database management system (GPDB) must load large amounts of data from data lakes such as Amazon S3 or Azure WASB, before any analytic type of query processing of data can begin, which requires low latency networking and is a big part of end-to-end processing time. In this talk we will show how utilizing the Samsung SmartSSD Computational Storage Drive's (CSD) storage component as a caching layer between the data lake and GPDB, query processing can start without the time-consuming loading phase. We will also demonstrate the benefits of the SmartSSD to dramatically lower the amount of data transfer to the CPU by offloading compute-heavy query workloads (decompression of compressed database file, parsing, certain query tasks on the cached data-like filter, projection, aggregation) to the SmartSSD's FPGA. Data Engine introduced in SmartSSD as a middle layer to cache data and perform query acceleration can improve overall query performance 3-4x over the typical CPU-based Greenplum Database operations.

Paper Author: Alex Paek, Database Solutions Architect, AMD

Author Bio: Alex Paek is a database solutions architect at AMD (formerly Xilinx). He has extensive experience in developing FPGAs for HPC and database applications. He received his MSEE from the Georgia Institute of Technology.