What comes to mind first with cloud system architectures is the ubiquitous use of microservices, supported by total virtualization from hardware to middleware to applications: virtual machines, micro operating systems, containers, and container orchestration. All of this gives microservices tremendous potential, while making such systems extremely complex. Microservices constitute, at the same time, a pattern for a cloud architecture and a source of concerns to be solved by the architecture at large.

This paper presents a set of proposed “principles and patterns for control-theoretic [1,2], model-based [3] architectures for cloud software [4].” The intended audience includes people interested in cloud computing solutions. The presentation level is suitable for both industry and academia.

The paper has a very systematic approach, taking the challenges and defining from them the principles that lead to the patterns to be implemented. The cloud is a multi-tiered provider of services produced through virtualization, living in a universe dominated by uncertainty: highly dynamic environments with often unpredictable changes. This requires mechanisms of self-adaptation. A full-stack view is necessary for continuous delivery/integration from a DevOps perspective. Thus, the architectural principles can be synthesized as follows: service orientation, virtualization, uncertainty, and adaptivity. The resulting architectural patterns include microservices complemented by packaging (containers) and orchestration (Kubernetes) as the ideal solution for continuous delivery; dynamic runtime models (using various approaches like Markovian chains, Bayesian inferences, fuzzy logic, and so on) to address uncertainty; and controller-based feedback loops (correlated with the runtime models) to ensure adaptivity. The authors discuss model and controller at length, and present three use cases.