Towards predictive runtime modelling of Kubernetes microservices

Abstract

Kubernetes is one of the major container management platforms utilised by Cloud Service Providers offering to host applications and services. As cloud based services become more prevalent, platform providers are faced with an increasingly complex problem of trying to meet contracted performance levels. Providers must strike a balance between management of resource allocations and contractual obligations to ensure that their service is profitable, while offering competitive pricing rates for contracts. This research explores performance modelling of microservice application tenants within the Kubernetes container management platform. We present a self-adaptive architecture to achieve modelling at runtime. We establish the potential for automated classification of cloud systems, and utilise a hybridised modelling approach to verify system properties and evaluate performance. We achieve this through the modelling of components as Extended Finite State Machines in WATERS, from which we automate the generating of performance models using the PEPA syntax.