Kubernetes Horizontal Pod Autoscaler Enhanced Design for Protected Workloads

By adapting the virtual resource reservation to the actual service load, autoscaling permits to reduce the Datacenter energy consumption and the need for additional physical servers. Kubernetes Horizontal Pod Autoscaler (HPA) can be used for Cloud-native Network Functions like the 3GPP components of the Radio Access Network or of the mobile Core Network but does not take into account their need for protection. Based on a discrete time model of the HPA that we first validate against measurements, we propose here the design of a light Kubernetes HPA enhancement to fix this gap. We show the results of a full implementation and evaluate the expected energy efficiency improvements.