Multi-criteria decision support for energy-efficient IoT edge computing offloading

Computation offloading is one of the primary technological enablers of the Internet of Things (IoT), as it helps address individual devices’ resource restrictions (e.g. process- ing and memory). In the past, offloading would always utilise remote cloud infrastruc- tures, but the increasing size of IoT data traffic and the real-time response requirements of modern and future IoT applications have led to the adoption of the edge computing paradigm, where the data is processed at the edge of the network, closer to the IoT devices. The decision as to whether cloud or edge resources will be utilised is typically taken at the design stage, based on the type of the IoT device. Yet, the conditions that determine the optimality of this decision, such as the arrival rate, nature and sizes of the tasks, and crucially the real-time conditions of the networks involved, keep changing. At the same time, the energy consumption of IoT devices is usually a key requirement, which is affected primarily by the time it takes to complete tasks, whether for the actual computation or for offloading them through the network. This thesis presents a dynamic computation offloading mechanism, which improves the performance (i.e. in terms of response time) and energy consumption of IoT de- vices in a decentralised and autonomous manner. We initially propose the Multi-critEria DecIsion support meChanism for IoT offloading(MEDICI), which runs independently on an IoT device, enabling it to make offloading decisions dynamically, based on multiple criteria, such as the state of the IoT, edge or cloud devices and the conditions of the net- work connecting them. It provides mathematical models of the expected time and energy costs for the different options of offloading a task (i.e. to the edge or the cloud or the IoT device itself). To evaluate its effectiveness, we provide simulation results, by extending the EdgeCloudSim simulator, comparing it against previous families of approaches used in the literature. Our simulations on four different types of IoT applications show that allowing customisation and dynamic offloading decision support can improve drastically the response time of time-critical and small-size applications, such as IoT cyber intrusion detection, and the energy consumption not only of the individual IoT devices but also of the system as a whole. Furthermore, we present an enhancement of our MEDICI mechanism, the ProbeLess Multi-critEria DecIsion support meChanism for IoT offloading (PL-MEDICI), which en- ables MEDICI to operate in real IoT environments without the need for probing or having pre-defined parameters in order to estimate or model the network conditions or the com- putation capabilities of the different devices involved. This is the first probeless dynamic and decentralised offloading decision support mechanism for IoT environments. The probeless property is achieved by combining lightweight statistical techniques with the concept of age of knowledge (AoK) to allow us to have accurate enough information to use for our estimations. We provide experimental results performed in a real IoT testbed with three real IoT applications, showcasing that PL-MEDICI outperforms existing techniques in terms of both response time and energy consumption. Finally, in order to further evaluate our PL-MEDICI mechanism, we formulate a mixed- integer linear program optimisation problem that provides the theoretical optimal cen- tralised solution to our problem. This is used to compare our PL-MEDICI against the theoretical optimum, given the same estimated input. Our results showed that our of- floading mechanism is close to the obtained optimal solution in terms of both the re- sponse time and energy consumptio