Practical guides for energy saving by power-off - Task 057

Fit to programme

This task has been identified by the working groups as part of the agenda behind WP 2.3.

The task number is 057.

Summary

HPC power consumption is high, often with significant wastage when power is used to keep idle servers active. Applications responding to this project should seek to deliver instructions and documentation for setting up active power saving measures at HPC centres to power off servers when not in use. In particular, the instructions should consider two situations: nodes that are part of a batch queue system, and nodes which are not (either user-accessible, or non-user accessible). The instructions should be drawn from experience of implementing the solutions at HPC sites, so that practical tips can be included. This is important for reducing wastage and emissions at HPC centres.

Methodology

Proposals for this project should first identify target HPC systems where idle nodes consume power unnecessarily. They should then develop and implement e-learning content around evidence-based methods to ensure that these nodes are powered off when not required (e.g. when idle for a short time period), and then also automatically power on these systems upon user demand (e.g. when a user tries to access them). It is expected that this e-learning content will not be theoretical, but based on actual implementation, placed into production, on an HPC facility. Documentation should then be provided detailing the techniques and methods used, and any drawbacks. A summary of potential energy savings should also be presented along with evidence over a relevant period of time.

Outcomes

The key output will be e-learning content, hosted on the SHAREing website, demonstrating how idle HPC nodes can have their power state automatically controlled (on or off) according to user demand. This e-learning content will be evidence-based, developed around a working production method to automatically control the power state of idle nodes on one or more HPC facilities, such that the content is production-tested. The deliverable will be a report on the implementation of this, any drawbacks, and a summary of power saved over a monitored time period. The deliverable should be shared on the SHAREing website.