Application-aware Congestion Mitigation for High-Performance Computing Systems

TL;DR

This paper introduces Netscope, an ML-driven framework that dynamically mitigates network congestion in HPC systems by considering application-specific network characteristics, significantly reducing runtime variability and improving system utility.

Contribution

Netscope is a novel, automated ML framework that accurately predicts congestion impacts and adapts mitigation strategies in real-time for HPC applications.

Findings

Netscope achieves a correlation of 0.7 to 0.9 in estimating congestion impact.

It reduces tail runtime variability by up to 14.9 times.

It improves median system utility by 12%.

Abstract

High-performance computing (HPC) systems frequently experience congestion leading to significant application performance variation. However, the impact of congestion on application runtime differs from application to application depending on their network characteristics (such as bandwidth and latency requirements). We leverage this insight to develop Netscope, an automated ML-driven framework that considers those network characteristics to dynamically mitigate congestion. We evaluate Netscope on four Cray Aries systems, including a production supercomputer on real scientific applications. Netscope has a lower training cost and accurately estimates the impact of congestion on application runtime with a correlation between 0.7and 0.9 for common scientific applications. Moreover, we find that Netscope reduces tail runtime variability by up to 14.9 times while improving median system utility by 12%.