CERN Computing Seminar

Hardware acceleration with GPUs and FPGAs

by Shasank Chavan (Oracle Corp.)

Europe/Zurich
513/1-024 (CERN)

513/1-024

CERN

1
Show room on map
Description

Hardware acceleration of database workloads on modern hardware has been at the forefront of Oracle’s Database organization for many years - ever since the acquisition of Sun Microsystems and the development of Exadata Engineered Systems. With the arrival of Oracle Database In-Memory option, analytic workloads were drastically accelerated in part by leveraging SIMD instructions in modern processors. Oracle innovated one step further by supplying a custom ASIC on Sun Microsystems M7/M8 processors where database operations on a column store (e.g. decompression and scans) could be offloaded to. Recently with the Oracle Cloud Infrastructure (OCI) hardware accelerator cards such as GPUs are now readily available. We are starting to look at SQL offload to these hardware accelerators (GPUs and FPGAs), and this talk will go into some detail on what that exactly entails, along with what has currently been done in industry and academia.

About the speaker

Shasank Chavan is the Vice President of the In-Memory Technologies group at Oracle. He leads an amazing team of brilliant engineers in the Database organization who develop customer-facing, performance-critical features for an In-Memory Columnar Store which, as Larry Ellison proclaimed, “processes data at ungodly speeds”. His team implements novel SIMD kernels and hardware acceleration technology for blazing fast columnar data processing, optimized data formats and compression technology for efficient in-memory storage, algorithms and techniques for fast in-memory join and aggregation processing, and optimized in-memory data access and storage solutions in general. His team is currently hyper-focused on leveraging emerging hardware technologies to build the next-generation data storage engine that powers the cloud. Shasank earned his BS/MS in Computer Science at the University of California, San Diego. He has accumulated 20+ patents over a span of 20 years working on systems software technology.

More information