Naiad and Timely Dataflow

Speaker Name: 
Michael Isard
Speaker Organization: 
Google Brain
Start Time: 
Friday, November 30, 2018 - 1:20pm
End Time: 
Friday, November 30, 2018 - 2:25pm
Location: 
Engineering 2, Room 215
Organizer: 
Lindsey Kuper

Abstract: 

I will give a meta-research talk about our SOSP 2013 paper "Naiad: A Timely DataFlow System". I will talk about some details of the system we built, but mostly I will talk about how the paper came about as part of our research program; what problems we were trying to solve; how well I think the system succeeded at solving them; how well I think the paper succeeded at explaining our solution; and comment on what if any impact the work seems to have had since its publication.
 
Bio:
 
Michael Isard received a D.Phil in computer vision in 1998, and gradually started doing more systems research over 13 years at Microsoft Research Silicon Valley. While at Microsoft he participated in a variety of distributed systems research including leading the Dryad distributed dataflow project. Since 2015 he has worked at Google Brain, mostly on the TPU back-end for TensorFlow. Much of his systems research is related to the interaction between programming languages and systems, and how language features can make it easier to write high-performance distributed and concurrent programs.