Recent Ph.D graduate Staci Smith and her advisor, Professor David Lowenthal, have won the Best Paper Award at the recent High-Performance Distributed Computing (HPDC'21) conference. HPDC is one of the premiere conferences in high-performance computing. The paper authored by Smith and Lowenthal is entitled "Jigsaw: A High-Utilization, Interference-Free Job Scheduler for Fat-Tree Clusters".
The paper, based on Smith's Ph.D dissertation research, describes design and implementation of Jigsaw, a new job-isolating scheduler for three-level fat-trees. Jigsaw proactively enforces network isolation for every job, which completely avoids application performance degradation that would otherwise occur due to inter-job network interference. Unlike existing job-isolating schedulers, which suffer from lowered system utilization, Jigsaw typically achieves system utilization of 95-96%. In scenarios where jobs experience even modest performance improvements from interference-freedom, Jigsaw typically leads to lower job turnaround times and higher throughput than traditional job scheduling.
The award culminates Smith's decorated Ph.D student career. Her first paper, "Mitigating Inter-Job Interference Using Adaptive Flow-Aware Routing", was nominated for Best Student Paper at Supercomputing 2018. She also is the recipient of a 2019 ACM/IEEE-CS George Michael Memorial HPC Fellowship. Smith has taken a position as a software engineer at Google.
The paper is available here.