Panel Discussion on Network Congestion

Rockport Networks logo
Hyperion Research logo

Quieting Noisy Neighbors: Addressing Network Congestion in Multi-Workload Environments

Monitoring and controlling network congestion is now a critical element of operating large-scale HPC, ML, and AI clusters. Amdahl’s law predicts an effective maximum scale for HPC workloads related to the time taken for the parts of the workload that must be executed serially. However, Amdahl’s law ignores the cost associated with inter-process communication by assuming that the serial part of a workload is fixed. In fact, network congestion caused by parallelization increases the serial part of a workload, further limiting workload scale.

This roundtable, moderated by Hyperion Research, will offer real-world insights on the challenges of congestion in multi-workload environments. We will discuss the root causes of congestion and the resulting ripple effect it can have on performance and scale. We will share benchmarking approaches to help measure and even predict the impact in production. And finally, we will explore the strategies and techniques available to mitigate or eliminate these impacts altogether.

headshot of Jeff Kirk

Jeff Kirk

HPC and AI Strategist, Office of the CTO
Dell Technologies 

headshot of Hari Subramoni

Hari Subramoni

Research Scientist, Department of Computer Science and Engineering,
Ohio State University (OSU)

headshot of John McCalpin

John D. McCalpin

Research Scientist, HPC Performance and Architectures Group
Texas Advanced Computing Center, TACC

headshot of Matthew Williams

Matthew Williams

Chief Technology Officer
Rockport Networks

headshot of Bob Sorensen

Bob Sorensen

Senior Vice President of Research and Chief Analyst for Quantum Computing,
Hyperion Research

Access the Panel Discussion

© 2021 Tabor Events. All Rights Reserved. A Division of Tabor Communications Inc.