Model-Driven Simulation for Performance Engineering of Kubernetes-Style Cloud Cluster Architectures

We propose a performance engineering technique for self-adaptive container cluster management, often used in cloud environments now. We focus here on an abstract model that can be used by simulation tools to identify an optimal configuration for such a system, capable of providing reliable performance to service consumers. The aim of the model-based tool is to identify and analyse a set of rules capable of balancing resource demands for this platform. We present an executable model for a simulation environment that allows container cluster architectures to be studied. We have selected the Kubernetes cluster management platform as the target. Our models reflect the current Kubernetes platform, but we also introduce an advanced controller model going beyond current Kubernetes capabilities. We use the Palladio Eclipse plugin as the simulation environment. The outcome is a working simulator, that applied to a concrete container-based cluster architecture could be used by developers to understand and configure self-adaptive system behavior.