A Reliability and Availability Model of a Kubernetes Cluster Using SysML

This paper demonstrates the use of Model Based Systems Engineering and SysML for a Kubernetes cluster and how the results of such models can be used for design and architectural decisions. There are two major innovations in this paper. The first is use of SysML for reliabiltiy/availability modeling of a computer system with a containerized software architecture using Kubenetes, the most common orchestration (system management) platform for containerized architectures [1]. By using SysML, it is possible to incorporate reliability and availability models into a Model Based Systems Engineering (MBSE) development process. The benefits are better design decisions and lower cost. The second innovation is a description of how the Litmus Chaos failure simulation testing framework [8] can be used for empirical measurement of reliability/availability model parameters. Systems built on the Kubernetes platform can be developed incrementally so that executable systems with partial functionality can be observed and measured to provide early feedback on system reliability and availability performance. The early feedback enables more accurate assessments and more effective corrective actions if necessary. The numerical results of the SysML model were verified using an independent model. The results of the two models agreed to the 9
^th
significant figure or better.