Cross-layer Application-aware Power/Energy Management for Extreme Scale Science

TL;DR

This paper discusses the need for a comprehensive, cross-layer, application-aware approach to power and energy management in extreme-scale HPC systems to achieve significant efficiency improvements while maintaining performance and reliability.

Contribution

It introduces a novel cross-layer, application-aware framework for power and energy management tailored for exascale HPC systems, addressing multiple objectives and tradeoffs.

Findings

Highlighting the limitations of current HPC power solutions.

Proposing a multi-layer, application-aware management strategy.

Emphasizing the importance of integrated approaches for exascale energy efficiency.

Abstract

High Performance Computing (HPC) has evolved over the past decades into increasingly complex and powerful systems. Current HPC systems consume several MWs of power, enough to power small towns, and are in fact soon approaching the limits of the power available to them. Estimates are with the given current technology, achieving exascale will require hundreds of MW, which is not feasible from multiple perspectives. Architecture and technology researchers are aggressively addressing this; however as past history is shown, innovation at these levels are not sufficient and have to be accompanied with innovations at higher levels (algorithms, programming, runtime, OS) to achieve the multiple orders of magnitude reduction - i.e., a comprehensive cross-layer and application-aware strategy is required. Furthermore, energy/power-efficiency has to be addressed in combination with quality of solutions, performance and reliability and other objectives and appropriate tradeoffs are required.