Paper Title: Database Scan Acceleration Using Computational Storage Devices

Paper Abstract: Host computation bottleneck has limited the use of high-performance SSDs in big data analysis applications. While lots of data movements from storage devices to host memory are incurred in these applications, storage devices often remain idle because their compute overheads are too high to generate burst IOs to SSDs. A carefully designed in-storage data compute routines can scale with its internal IO bandwidth. Samsung Smart SSD 2.0 is developed for this and based on NVMe TP4091 standards and SNIA CS APIs. It utilizes both FPGA and user-downloaded programs running on an ARM processor to achieve both programmability and energy-efficient computing. DB SCAN processing engine in FPGA for Smart SSD 2.0 is demonstrated that the data can be internally scanned at the speed of a SSD, and thus highly enhance performance, energy efficiency, and scalability of the system. This minimizes host resource usages for IO and SCAN operation allowing a greater number of devices to provide their full performance, possibly leveraging high scalable performance SSDs. We will present the system architecture and scalable performance enhancement on the cutting edge computational storage devices.

Paper Author: Changho Choi, Sr. Principal Engineer, Samsung Semiconductor

Author Bio: Dr. Choi leads Samsung SSD research and new SSD features initiatives including computational storage devices (i.e., Samsung Smart SSD) in Samsung Memory Solutions Lab. He also leads open source initiatives and standards group engagement including OCP, NVMe and SNIA. He has successfully contributed to complete multiple standards in storage standards group, NVMe and SNIA. In addition, he has delivered many presentations at leading storage conferences such as Flash Memory Summit, Storage Developer Conference, etc. He has solid knowledge of SSD internals, storage system and infrastructure, and datacenter solutions including database systems.