Carbon Conscious Scheduling in Kubernetes to Cut Energy Use and Emissions

With the growing energy consumption of hyperscale data centers, cloud computing is also growing and the proliferation of cloud computing is increasing its greenhouse gas (GHG) emissions. Kubernetes, the strongly recommended container orchestration platform through the multi-cloud environments, heavily impacts the deployment of compute resources. Traditional scheduling focuses on throughput, latency, and resource usage and does not pay attention to the carbon intensity (gCO 2 /kWh) of the consumed electricity. This paper postulates the use of carbon-sensitive scheduling throughout Kubernetes to dynamically schedule operations according with low-carbon energy production times. By using realizable, predictive carbon intensity values available through APIs like ElectricityMap or WattTime, workloads can be either geographically scheduled to be run in regions with higher renewables penetration, or scheduled so that it runs during periods when the grid is predicted to be cleaner. Placement makes use of native Kubernetes mechanisms node affinity, taints and tolerations, and custom scheduling policy to optimise placement. This paper describes an architectural framework, system assessment methods based on carbon intensity trace data, and deployment case studies of innovative technologies at major technology companies and research laboratories. The results show that carbon-aware workload placement can save at least 10 and up to 30 percent of CO 2 emissions and do so without any perceptible performance impact. There are still difficulties of granularity of carbon data, interoperability standards, and adoption of the enterprise