Evolution of on-Node Adaptive Power Tuning

The presentation discusses the evolution of on-node adaptive power tuning to optimize available resources and decrease power use and cooling costs of systems.

• Adaptive on-node systems are needed to optimize available resources and decrease power use and cooling costs of systems
• Current options available require setting options for power use ahead of time or reliance on kernel-level governors
• TuneD provides a framework to configure certain power optimization settings on different hardware using profiles
• B States and C States are important capabilities to concentrate on for power optimization
• Intel's Kubernetes Power Manager improves what can be done with power optimization
• The presentation includes a pre-recorded demo of the discussed concepts

The speaker mentions that power optimization is becoming a hot topic and has two advantages over other topics that fall under the umbrella of ESG, which are environmental, social, and governance. It is quantifiable and can potentially save money. The speaker also mentions that there is some work happening to do more power savings optimizations in a wider context, such as for the whole cluster or multi-clusters.

To decrease power use and cooling costs of our systems, we need more adaptive on-node systems in order to optimize the available resources. Current options available require setting options for power use ahead of time or reliance on kernel-level governors. We can do better. Charles Darwin once said that “Intelligence is based on how efficient a species became at doing the things they need to survive.” By this definition, most of our systems are primitive and wasting many resources, consuming calories that they do not need and throwing them off as heat energy, which we then have to consume more calories to remediate in our datacenters. In this talk, we give a vision of more adaptive power-tuning models. We will use a combination of TuneD and the Kubernetes Power manager to demonstrate a model of how we can design more intelligent systems.