Sustainability Through Accountability in a CNCF Ecosystem

The presentation discusses the urgent need for energy efficiency in the data center industry and proposes a solution using Kepler and Rook technologies to measure and correlate workload energy consumption with carbon emissions.

• The integration of AI into search engines will lead to a significant increase in energy consumption in the data center industry
• The industry's carbon footprint is comparable to that of the airline industry and is growing faster
• Metrics and measurements are needed to accurately assess energy efficiency and carbon emissions
• Kepler and Rook technologies can be used to measure and correlate workload energy consumption with carbon emissions
• Linear and non-linear models can be used to create energy consumption models
• The created models can be uploaded to GitHub and used in a Prometheus container
• The solution can be run on both virtual and physical machines

The presenter showed a graph of energy consumption over time for different namespaces, with the highest energy consumption being from the Rook namespace. The solution was demonstrated to be able to break down energy consumption at the part level.

Carbon footprint and energy consumption accounting is essential to sustainable Cloud native computing. However, such capability is quite challenging in multi-tenant services such as compute and storage. This session explains how to use CNCF projects to achieve this goal. Specifically, the Rook operator provides cloud-native storage to applications, orchestrating the Ceph storage system. Recently, tracing was added to Ceph, using an Open Telemetry client and Jaeger backend. In this talk, we would show how we combine per pod energy consumption data coming from Kepler, together with the tracing information coming from Jaeger to estimate the energy consumption of each user in the storage system, even when the consumption is spread among multiple pods. This solution highlights the feasibility of building sustainable computing futures in the CNCF ecosystems. It will benefit both end users and developers and inspire more innovations.