The HPC PowerStack: A Community-Driven Collaboration Toward Power-Aware, Energy-Efficient, and Sustainable HPC in the Exascale Era

While there exist several standalone efforts that attempt to tackle power, energy, and sustainability challenges in this exascale era, the majority of the implemented techniques have been designed to meet site-specific needs. There has been no consensus among the stakeholders in academia, research and industry on which software components of a modern HPC stack should be deployed and how they should interoperate. Coordination among these components is critical for optimizing for a target metric (e.g., FLOPS per watt) while meeting operational constraints (e.g., power). This led to the formation of the PowerStack Community in 2016 (https://www.osti.gov/biblio/1466153/). Since then, its charter has been to identify the list of power management software components, to establish protocols between them to streamline efficiency goals within a site, and to prototype solutions with existing open-source production software for a cost-effective yet cohesive, cross-platform implementation of the software stack. These concepts were followed and inherited by several ongoing research projects (e.g., Regale, EE-HPC, etc.) and are being updated based on their experiences. The long-term vision is to work towards reaching a community-wide consensus in terms of standardization of power management capabilities across the stack. This interactive vendor-neutral BoF will provide a platform to bring together the stakeholders from academia, research, and industry to participate in this effort and work towards fine-tuning the stack. The goal of this BoF is to share lessons learned through prototyping efforts, provide progress updates, invite audience feedback on current directions, brainstorm solutions to open questions and foster collaboration in the community.