Orchestration Mechanism Impact on Virtual Network Function Throughput

Abstract

Virtual Network Function (VNF) has gained importance in the IT industry, especially in the telecommunication industry, because a VNF runs network services in commodity hardware instead of dedicated hardware, thereby increasing the scalability and agility. The container technology is a useful tool for the VNF because it is lightweight, portable and scalable. The container technology shortens the product development cycle by easing the service deployment and maintenance. The telecommunication industry uses service uptime as an important gauge to evaluate if a service is of carrier grade, and keeping services up and running generates most of the maintenance costs. These costs can be reduced by container orchestration such as Kubernetes. Kubernetes handles the automation of deployment, scaling and management for applications with the help of orchestration mechanisms, such as the scheduler and load-balancers. As a result of those mechanisms, the VNFs running in a Kubernetes cluster can reach high availability and flexibility. However, the impact of the mechanisms on VNF throughput has not been studied in detail.

The objective of this thesis is to evaluate the influence of Kubernetes orchestration mechanisms on VNF throughput and Quality of Service (QoS). This objective is achieved by means of measurements run with a packet-forwarding service in a Kubernetes cluster.

Based on the evaluations, it is concluded that the VNF throughput is dependent on 6 parameters: CPU types, CPU isolation, number of Pods, location of Pods, location of load-balancer controllers, and load-balancing techniques.