Modern datacenters consume enormous amounts of electricity, a challenge accelerated by the growth of AI workloads. WAO offers a purely software-based approach to reduce server energy consumption by predicting power usage on individual nodes and optimizing workload placement.
We integrated WAO with Kubernetes in two ways:
- WAO-Scheduler uses a custom scheduler plugin to assign Pods to nodes likely to incur the smallest increase in power usage.
- WAO-LoadBalancer is a customized kube-proxy that balances traffic across nodes to minimize power usage from incoming requests.
Both of these components rely on the WAO predictive model:
- It forecasts server power consumption using CPU usage, inlet temperature, and static pressure as inputs.
- A separate model must be developed for each server type through measurement, yet remains independent of the server’s physical deployment environment.
We conducted a series of experiments in a real-world datacenter and found:
- WAO-Scheduler achieved up to a 10–20% reduction in overall power consumption in real-world tests.
- WAO-LoadBalancer is still under active development.
This open-source implementation includes custom resources for per-node configuration, a metrics adapter to collect environmental data, and integrations with both the scheduler plugin and our custom kube-proxy.
Future work involves integrating with cooling systems, enhancing workload rebalancing, and extending support to GPUs. By leveraging Kubernetes extensibility and predictive modeling, we aim to make datacenters more energy-efficient without compromising user experience.
- Presentation at Kubernetes Meetup Tokyo #66 (日本語): [Video] [Slides]
- Experiments in datacenter: Ying-Feng Hsu et al., "Sustainable Data Center Energy Management through Server Workload Allocation Optimization and HVAC System", IEEE Cloud Summit 2024, Washington DC, USA, 2024.
- WAO: R. Douhara et al., "Kubernetes-based Workload Allocation Optimizer for Minimizing Power Consumption of Computing System with Neural Network," 2020 International Conference on Computational Science and Computational Intelligence (CSCI), Las Vegas, NV, USA, 2020.