Staci Smith, David Lowenthal win Best Paper Award at HPDC'21.

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.