*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
** Joint CERCS / CHiPC Seminar **
"Communication Challenges & Opportunities for Graph Processing"
Dr. Scott Beamer, Lawrence Berkeley National Laboratory
Date: Wednesday, September 20, 2017
Time: 12-1 pm
Location: Klaus 1116-W
* Abstract: For many data-intensive workloads such as graph processing, accessing the necessary data (communication) can often be a bigger challenge than the computation itself. For these communication-bound workloads, optimizing communication can bring many benefits, whether it be improving performance, saving energy, or reducing manufacturing cost. To best optimize communication, one should take a vertically-integrated approach, considering the needs of the algorithms while simultaneously appreciating the capabilities of the underlying architecture.
In this talk, I will describe the challenges and opportunities for improving communication efficiency, particularly for server-class systems communicating with main memory. I focus on graph processing, as its data-intensive nature exposes many communication challenges while the graph abstraction provides a common perspective to analyze its algorithms. To accelerate low-locality graph algorithms, I introduce propagation blocking, a technique that reduces communication by improving spatial locality. I conclude by discussing communication tradeoffs and the potential for specialized hardware to further enhance communication efficiency.
* Bio: Scott is a postdoctoral fellow in the Computer Architecture Group at Lawrence Berkeley National Lab (LBNL). He designs architectures, systems, and algorithms to improve communication efficiency, which in turn saves energy and reduces cost for communication-intensive applications. He received the Kaivalya Dixit Distinguished Dissertation Award from SPEC as well best paper awards from the International Parallel & Distributed Processing Symposium (IPDPS) and the International Symposium on Workload Characterization (IISWC). He earned his Ph.D. in Computer Science at UC Berkeley, where he had previously received his B.S. and M.S. degrees.
* Host: Rich Vuduc <richie@cc.gatech.edu> -- Please contact Rich if you'd like to meet with Scott!
