This thesis treats the intersection between two of the largest transformations we are seeing within our society today; the cloud and the Internet-of-Things (IoT). The aim of this thesis is to investigate different ways to model and control a network of cloud services so that timing-critical IoT applications can make use of them. Examples of such applications can be autonomous and mobile robots, smart production plants, or massive multi-player augmented-reality games. The main motivational use-case, however, comes from the industrial side, and their digitalization, the drive towards industrial IoT (IIoT). We wish to enable smart robots to offload some of their computations to the cloud in order to allow for better and smarter control and collaboration. For instance, using the cloud, it would become possible for them to collaborate and make use of smarter analytics, artificial intelligence, and machine learning, in order to improve efficiency and safety.
To address this problem the thesis combines concepts and theory from different fields, most notably from control theory, real-time systems, and network calculus. Examples are: modeling of dynamic systems and the use of feedback and feedforward control from control theory, the goal of ensuring that end-to-end deadlines are met, from real-time systems, and finally the principles of modeling traffic from network calculus.
The thesis begins with an introduction to provide some background on cloud, IIoT, and to set the scope of the thesis. Following this, we begin by treating the problem of controlling a single cloud service with the goal of ensuring that the traffic flowing through the node is guaranteed to meet a deadline. Following this, we study a chain of connected cloud nodes, investigating how to provide end-to-end deadline guarantees for the traffic flowing through the chain. The chain is finally generalized to a network of cloud nodes, with multiple flows traversing it. For this problem we study how to ensure that the end-to-end deadline of every single flow in the network is guaranteed. We also provide a set of protocols controlling how cloud nodes and flows are allowed to dynamically join and leave the network, such that no end-to-end deadline is violated.