HN Theater @HNTheaterMonth

Title: Reconfigurable Computing at HyperScale
Presenter: Andrew Putnam (Microsoft Research)

Abstract:
Microsoft first deployed FPGAs at hyperscale in 2015. Since then, Microsoft’s cloud has continued to grow exponentially, providing developers with a greater capacity of reconfigurable computing logic than has ever been deployed. Accelerated networking (AccelNet), deep neural network inference (BrainWave), and web search (Catapult) are among the most prominent applications deployed in production at hyperscale today.

We have learned a number of lessons in the first few years of deployment – what to do, what not to do, how to structure development teams, and the kinds of problems that only occur at hyperscale. In this talk, I’ll discuss what we’re doing with the current fleet of FPGAs, some of the lessons we’ve learned along the way, what those lessons mean for the future of reconfigurable computing in the Cloud, and what the future holds for reconfigurable computing in the Cloud and Edge. This is especially targeted at graduate researchers, with anecdotes about what it takes to deploy and support HyperScale applications, and fruitful areas for future research.

Bio:

Andrew Putnam is Principal Hardware Development Engineer in Microsoft Research and Azure Networking. He is the co-founder of the Microsoft Catapult project, which was the first to put Field Programmable Gate Arrays (FPGAs) into production hyperscale data centers, doubling the capacity of each server for web search, and creating the fastest network in the cloud. He triple majored in Physics, Computer Science, and Electrical Engineering at the University of San Diego in 2003, and finished his Ph.D. in Computer Science and Engineering at the University of Washington in 2009. He loves hockey, beer, Tim Hortons, and Canada, but is still uncertain about poutine.

Slides available at: https://drive.google.com/open?id=1W060suiqjiVd1wImvnKjqhjBQYkT15t9