Workload Volatility Management in Datacenters and Cloud

Abstract

With increasing adoption of cloud by enterprises, resource management has never been more important than it is now. Efficient utilization of infrastructure can not only improve application performance but also help enterprises save significantly on their operational expenditure. For features like Cluster Autoscaling, inefficient resource management can lead to cluster resources getting scaled in or out more frequently, thereby increasing operational expenses. Traditional resource management engines distribute workloads across the nodes based on their utilization. This approach is most effective when the workloads are more or less stable. If the workloads are unstable or volatile, this approach of distributing the load is not optimal as workloads keep varying and the resource management engine can end up moving workloads from one node to another continuously to keep the load distributed. This results in inefficient resource utilisation, overloading of hosts and wastage of resources due to continuously moving workloads. With new workloads like container applications and Kubernetes, clusters are more prone to workload volatility since these workloads are short-lived and can frequently switch between the containers. In the future, handling this workload volatility will be critical to support these new container workloads. To address this problem, in this paper, we introduce a new Volatility Balancer which can work with any Load Balancer and distribute workload volatility along with workload utilization. Volatility Balancer, together with a Load Balancer, can provide an efficient resource management model for all the new generation workloads.