(CSE Colloquium Lecture Series)
Speaker: Sheng Li, Staff Research Scientist, Intel Labs
Date: Tuesday, April 25, 2017
Location: Room 1202, CSE Building
Host: Dean Tullsen (firstname.lastname@example.org)
Abstract: Big data, including cloud computing, deep learning, and beyond, is causing a tectonic shift in the computing industry: one that opens up new business and access models, changes the way technology is consumed, and has profound effects on the underlying platforms. With this shift, the abstractions and boundaries that used to separate system research from architecture research are collapsing. As a result, full-stack and wholes system research is critical in designing future platforms for ever-increasing problem sizes, in exposing new opportunities in software and architecture designs, and in defining the new synergic insights and abstractions that will accelerate research in these complementary fields.
In this talk, I will present my recent full stack and whole system research in high performance and efficient big data platforms. Firstly, I will introduce a key-value store platform with record-setting performance and energy efficiency, being 1000X faster than current key-value systems. Then, I will present a hyperscale datacenter infrastructure that improves datacenter efficiency and TCO by 50%. This work has been adopted by HP Moonshot systems, demonstrating a major real world impact of my research. Finally, I will discuss my future research in this rapid changing field, including on-going research on a high-performance deep learning platform.
Bio: Dr. Sheng Li is a staff research scientist and a technical lead on enterprise big data research at Intel Labs. He was a senior research scientist at HP Labs previously, where his work influenced HP moonshot hyperscale server architecture. He obtained Ph.D. in Electrical Engineering from the University of Notre Dame at 2010. Dr. Li has published more than 30 technical papers in world-class venues and holds 34 patents (both awarded and pending). Dr. Li’s work has won multiple awards, including the IEEE micro’s Top Picks'16, the Best Paper Award Honorable Mention from MICRO’13, the HiPEAC'11 paper award, and multiple best paper finalists from top conferences. His publications have been heavily referenced worldwide with 2000+ citations from top tier publications. The McPAT modeling framework for manycore processors developed by him has become one of the de facto standard modeling frameworks for computer architecture and system research. He was also listed as a major contributor to the classic computer architecture textbook: "Computer Architecture: A Quantitative Approach" 5th Edition. Dr. Sheng Li has served on many program committees of top-tier international conferences.
Related Research Publications
Architecting to Achieve a Billion Requests Per Second Throughput on a Single Key-Value Store Server Platform
System-level Integrated Server Architectures for Scale-out Datacenters
Faster CNNs withDirect Sparse Convolutions and Guided Pruning