Optimization Plan for Containerized Ecosystems using Learning Systems
Containers in Kubernetes kind of environments place a ‘request’ to specify ‘minimum’ resources needed to operate. However, in many cases, actual utilization of these containers is much lower than what they have requested. This creates an environment where more resources are reserved than what’s actually required. These requests are decided in the first place based on experience of operations admin handling the container. Problems: • What is the optimal request for a given container for a given period? • Are the containers rightly distributed? Could they be rearranged in different nodes to make sure of optimal resource utilization? The paper proposes to collect utilization history of running containers using Kubernetes APIs and feed to learning system ‘R’ and asks for predicted utilization, using it to derive recommended request. This solution applied on a cloud based application in production, could achieve resource consolidation and hence cost savings of up to 80%.