In this video from the PBS Works User Group 2017, Jignesh Bhadaliya from Dell EMC presents: Storage Aware Job Scheduling.
“In any HPC environment, storage is an integral and often unaccounted performance component of the infrastructure alongside compute. The health and performance of the overall infrastructure can be highly dependent on storage. Storage related performance issues can lead to adverse effects on overall job throughput, often resulting in business impact. In a reactive manner, the issue is often addressed by throttling the job flow. A more elegant and proactive approach would be to establish a feedback loop between storage and compute with the assistance of a job scheduler. This talk will explore the need for such a feedback loop, its benefits, and potential implementation scenarios for embarrassingly parallel HPC workloads.”
Jignesh (Jig) Bhadaliya is the CTO for HPC/EDA at Unstructured Storage Division. Jig has over twenty years of experience in HPC/EDA, having worked at Dell EMC, Broadcom, Rockwell Semiconductors and University of California. Jig has been working in high performance computing and storage for over a decade, and is passionate about quantifying, characterizing, synthesizing, and replaying workloads to understand their effect on the overall HPC infrastructure. His most recent contribution defines the SFS2014 SP2 EDA benchmark suite, to be released later this year. Jig is based out of Orange County, California, where he has spent the majority of his life. When not working, Jig enjoys spending time with his family, bicycling, or hacking embedded devices.
