Duke Computer Science Seminar

Harnessing Reliable Clocks & Homegrown AI: A Journey through Systems Innovation

September 28, -
Speaker(s): Michael Wei

Lunch

Lunch will be served at 11:45 AM.

Abstract

Part 1: Rethinking Distributed Clocks
For decades, the distributed systems community assumed that computer clocks are unreliable and must be constantly synchronized. In the first part of this talk, I will present Graham, work that challenges that assumption and shows that through an understanding of the electronics and physics in computer clocks, we can  deliver a software solution for producing more reliable clocks that can be synchronized less frequently. I will talk about the challenges we faced in going against state-of-the-art research at the time, and the new applications that are enabled with a reliable clock.

Part 2: Home-inspired Real-time AI Pipelines
Emerging technology areas such as AI can feel foreign to researchers in other disciplines. In the second part of this talk, I’ll talk about how trying to leverage AI in a home-built doorbell led me to a mountain of systems research problems, and how as a systems researcher, “dogfooding” and building a system yourself is the only way to really understand the nuances of a new system design.

Speaker Bio

Michael Wei is a senior research scientist at VMware Research. His work spans across the stack, from hardware to distributed systems, operating systems, and programming languages, and has won multiple awards at venues such as NSDI, Eurosys, Systor and USENIX ATC. His current research focuses on leveraging time as a distributed systems primitive, and building sustainable yet scalable AI systems infrastructure.

 

Contact

Faculty Host: Matt Lentz