*********************************
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
*********************************
Title: Overcoming Memory Capacity Constraints for Large Graph Applications on GPUs
Prasun Gera
Ph.D. student
School of Computer Science
Georgia Institute of Technology
Date: Friday, Nov 8, 2019
Time: 12 p.m. - 2 p.m. (Eastern Time)
Location: KACB 3100
Committee:
---------------
Dr. Hyesoon Kim (Advisor, School of Computer Science, Georgia Institute of Technology)
Dr. Santosh Pande (School of Computer Science, Georgia Institute of Technology)
Dr. Richard Vuduc (School of Computer Science and Engineering, Georgia Institute of Technology)
Abstract:
------------
Due to the limited capacity of GPU memory, most prior work on graph applications on GPUs has been restricted to graphs of modest sizes that fit in memory. Real world graphs such as social networks and web graphs require tens to hundreds of gigabytes of storage whereas GPU memory is typically in the order of a few gigabytes.
In this thesis proposal, we investigate the following question: How can we accelerate graph applications on GPUs when the graphs do not fit in memory? This question opens up two lines of inquiry. First, we consider the system architecture where the GPU can address larger, albeit slower host memory that is behind an interconnect such as PCI-e. While this increases the total addressable memory, graph applications have poor locality that makes efficient use of this architecture challenging. We formulate the locality problem as a graph ordering problem and propose efficient reordering methods for large graphs. Next, we consider graph compression as a complementary approach. We propose techniques that enable graph applications to decompress data at run-time efficiently and in parallel on GPUs.
