Kaan Kara gave an invited talk at the OpenPOWER Summit Europe

Kaan Kara gave an invited talk entiteld "Specialized Hardware Solutions for in-Database Analytics Enhanced by OpenCAPI+HBM" the OpenPOWER Summit Europe

Abstract:  

Driven by the increasing complexity of data processing tasks, specialized hardware solutions are emerging as a way to increase performance and efficiency for both relational data processing and advanced analytics such as machine learning. Recently increased usage of FPGAs as a target deployment platform in the datacenter opens up new opportunities in specializing hardware and rethinking the system architecture.

In the first part of this talk, we present an overview of our previous efforts in this domain, focusing on doppioDB: A branch of MonetDB with FPGA-based data processing capabilities. We introduce the hardware-software layer built to transparently use FPGA resources in a multi-tenant system, followed by an overview of already implemented FPGA-based operators, their advantages, and limitations. In the second part, we shift our focus to recently available OpenCAPI-attached FPGA platforms with high-bandwidth-memory (HBM). We discuss some preliminary results on how OpenCAPI and HBM can be utilized to improve FPGA-based processing within doppioDB.