Staci Smith, a fourth-year Ph.D student, has had her paper, “Mitigating Inter-Job Interference Using Adaptive Flow-Aware Routing”, nominated for the prestigious Best Student Paper award at Supercomputing 2018---the premiere conference in the area of high-performance computing (HPC). Only 68 of 288 (24%) of submissions were accepted, and of those, five papers have been selected as candidates for the Best Student Paper award. Full details are available here.
The paper is based on part of Smith’s Ph.D dissertation research, which focuses on developing techniques for mitigating inter-job interference on supercomputers. Her research will allow multiple jobs to execute on supercomputers without degradation that would otherwise occur due to competition for the shared interconnect. Her novel technique, called “Adaptive Flow-Aware Routing” (or AFAR), re-routes traffic from heavily-loaded routers to lightly-loaded ones---at the granularity of an entire flow. Results on a 1,296-node supercomputer show improvements of up to 46% for applications that communicate extensively. The paper is co-authored with Smith’s advisor, Professor David Lowenthal, and former Masters student Clara Cromey, as well as collaborators from Lawrence Livermore National Laboratory and Toyko Institute of Technology.
Smith’s general research interests are in performance analysis, performance modeling, and runtime systems for HPC machines. She currently focuses on communication and network performance, with interest in using routing, software-defined networking, and job scheduling techniques to decrease network congestion. She received her B.S in Mathematics and Computer Science from the University of Arizona in 2014, where she was selected as Outstanding Senior by both departments. She was a computer science Galileo Circle Scholar in 2013 and again in 2017, and she is currently the 2018-2019 ARCS Foundation Crawford Endowment Scholar.