Control-over-the-cloud: A performance study for cloud-native, critical control systems
Per Skarin, William Tärneberg, Karl-Erik Årzén, Maria Kihl
Abstract
In the Industry 4.0 era, time-sensitive and mission-critical control applications still have a long way to go, from being tied down and co-located with the systems they control, to taking full advantage of the cloud. Conservatively keeping applications local will deprive these complex applications of abundant compute capacity, wider system integration, and the potential for collaborative control efforts. Feedback control systems are unlike other cloud applications - their performance and objectives can be formally defined, they require timely feedback, and they are sensitive to variations in system performance and noise. Although resources are plentiful, the cloud is a noisy and latency prone execution environment, detrimental to feedback control system. In this paper, we evaluate a set of cloud platforms and infrastructures with the intention of hosting feedback control systems. In lower levels of the software stack, we observe differences between clouds. Further up in the stack we see the disadvantages of applying cloud native platforms. With an understanding of expected performance we proceed to evaluate a simple control strategy and show how the sensitive nature of control can cause a seemingly adequate cloud platform to pose a high risk, while a seemingly inadequate platform can positively affect the performance of our proposed controller.