Explore the accuracy and platform-specific performance of Kepler, the Kubernetes Efficient Power Level Exporter, in this 33-minute CNCF conference talk by Jie Ren and Ken Lu from Intel. Learn about Kepler's use of eBPF for probing energy-related system statistics in cloud-native containers and its export of Prometheus metrics. Discover the limitations of current white-box oriented test cases and the introduction of a new platform validation framework. Understand how this framework complements existing tests, enables developers and end-users to validate project features on specific hardware platforms, and allows for the integration of vendor-specific validation cases. Gain insights into evaluating platform support, identifying gaps and limitations, and assessing the accuracy of power modeling across different platforms. Examine the benefits of this fully automated, runner system agnostic framework for improving Kepler's functionality and reliability.
Overview
Syllabus
Is Kepler Accurate on Specific Platforms? - Jie Ren & Ken Lu, Intel
Taught by
CNCF [Cloud Native Computing Foundation]