Re-Thinking the Hardware-Software Interface for Data-Centric Systems
, Carnegie Mellon University
Date: Monday, May 16, 2022
Time: 11:00 AM to 12:00 PM Note: all times are in the Eastern Time Zone
Event Type: Seminar
Host: Professor Daniel Sanchez, CSG - CSAIL - MIT
Contact: Sally O. Lee, 3-6837, email@example.com
Speaker URL: None
TALK: Re-Thinking the Hardware-Software Interface for Data-Centric Systems
With the slowing of technology scaling, computer systems must look higher in the stack for performance and energy efficiency. Data movement is usually to blame for poor scaling, but current systems give software too few options to optimize data movement. As a result, many systems have turned to specialized solutions, often requiring custom hardware, to deal with the data movement problem.
This talk will argue that the hardware-software interface is the problem, and data-centric design is the solution. We will consider three data-centric systems that optimize data movement while remaining general-purpose: (i) Polymorphic cache hierarchies allow software to modify or expand the memory interface on general-purpose hardware. (ii) Energy-minimal dataflow architectures massively reduce instruction and data movement, letting software compete with ASIC efficiency. And (iii) General-purpose flash-based caches reduce cost by an order of magnitude in the datacenter behind a simple interface.
Nathan Beckmann is an assistant professor at Carnegie Mellon University in the Computer Science and (by courtesy) Electrical and Computer Engineering Departments. His research focuses on general-purpose, data-centric computer architectures and systems. His awards include a Google Research Scholar Award in 2021 and the NSF CAREER Award in 2019. He graduated from MIT advised by Daniel Sanchez, receiving the George M. Sprowls Award for an outstanding PhD dissertation in computer science in 2015.
Join Zoom Meeting
Meeting ID: 978 0240 8593
Created by Sally O. Lee at Wednesday, April 27, 2022 at 3:14 PM.