Explore a conference talk on optimizing Kubernetes workload allocation for minimized power consumption. Discover how the Workload Allocation Optimizer (WAO) leverages machine learning to predict power increases and introduces a scoring plugin to the K8s scheduler framework. Learn about the WAO-load balancer's role in assigning Pods to Nodes for optimal power usage. Gain insights into implementing power-saving strategies for cloud-edge computing systems, demonstrated in a real edge data center with over 200 servers. Understand the tradeoffs between service performance and data center power savings, and explore the architecture of WAO-based schedulers and load balancers. Examine evaluation results for power consumption reduction, response time, and the complete K8s-WAO solution's performance in various scenarios.
A K8s-Based Workload Allocation Optimizer for Minimizing Power Consumption
CNCF [Cloud Native Computing Foundation] via YouTube
Overview
Syllabus
Intro
Who are we: Matsuoka Lab at Osaka Univ.
Our technology: Liquid immersion cooling
Our technology: optimal task allocation
The complexity of edge computing systems
Spread of Microservices & Power consumption increases
Our Approach: WAO on K8s
WAO power saving operation strategy
Scheduler Framework on K8s
Architecture of WAO Based Scheduler
Architecture of WAO Based Load Balancer
Setting of Machine Learning Power consumption (PC) model
How does it work? How is the performance? Environment
Preset Temperature of Air Conditioner
Power Consumption Model
Evaluation Value in WAO Based Load Balancer
Evaluation of Power Consumption Reduction
Evaluation of "WAO Scheduler" + MetalLB
Evaluation of Kube Scheduler + "WAOLB"
Evaluation of complete K8s-WAO solution
Evaluation of Kubernetes-based WAO
Evaluation of Response Time
Conclusion
Taught by
CNCF [Cloud Native Computing Foundation]
