Measuring Sustainability via Project Kepler, with Niki Manoledaki

BLUF: This episode explores the critical intersection of cloud-native infrastructure and environmental sustainability, highlighting GreenOps as a methodology for optimizing resource efficiency and introducing Project Kepler as a key tool for measuring energy consumption in Kubernetes environments 3:55.

Key Takeaways:

• GreenOps and Sustainability: GreenOps emerged around 2022 to focus on operational efficiency and reducing the carbon footprint of cloud resources, distinct from but related to FinOps 4:02. Organizations adopt sustainability practices driven by regulatory requirements (like EU disclosure laws), corporate values, and cost optimization, though cost reduction does not always equate to carbon reduction 6:45.
• Project Kepler Functionality: Project Kepler is a Kubernetes-based tool that acts as an exporter for monitoring energy usage at the node, pod, and container levels 20:09. It utilizes an attribution model to map hardware power readings to specific workloads, allowing engineers to identify inefficiencies and optimize their infrastructure 20:40.
• Measurement vs. Estimation in Cloud: A major challenge in sustainability is the lack of direct visibility into hardware energy usage in public cloud VMs, often due to security restrictions like the blocking of RAPL (Running Average Power Limit) data 26:10. Consequently, Kepler often relies on estimation models for virtualized environments, requiring users to balance the need for accuracy with the practical necessity of taking action based on available data 29:23.
• AI and Carbon Tracking: The rise of Large Language Models (LLMs) introduces new complexity, as energy usage is tied to token counts rather than traditional CPU utilization metrics 33:50. Tracking the carbon impact of proprietary, cloud-hosted AI models is difficult due to "black box" nature of API providers, leading to recommendations for using specialized, smaller models where possible to reduce overhead 37:40.
• Developer Action: Engineers can contribute to sustainability by optimizing code, reducing unnecessary logging, and utilizing platform-level tools like autoscalers (e.g., Kubernetes Cluster Autoscaler or Carpenter) to ensure resources match demand efficiently 11:20.

Closing statement: As cloud-native technologies evolve, integrating sustainability into the development lifecycle is becoming as important as performance and cost. By leveraging tools like Project Kepler and adopting a mindset of operational efficiency, teams can reduce their environmental impact while maintaining robust infrastructure.

Sources:

• 3:55 Definition of GreenOps and its emergence in 2022.
• 6:45 Drivers for sustainability: cost, values, and regulation.
• 20:09 Introduction to Project Kepler and its role in energy monitoring.