Probeless and decentralised multi-criteria decision support for IoT computation offloading

In edge computing-enabled Internet of Things (IoT), the choice of whether a task should be processed by the IoT device itself or offloaded at the edge or at a distant cloud infrastructure is largely based on the original design of the system or best effort. Dynamic computation offloading decision mechanisms are able to achieve considerable accuracy at predicting the energy and computation costs of an offloadable task, but are typically reliant on a probing phase, which introduces delays. Here, we show that we can omit probing and its associated overheads by employing simpler lightweight estimation techniques for the cost predictions together with incorporating the concept of Age of Knowledge in the decision. Our solution, a Multi-Criteria, Probeless Dynamic Offloading Decision Support Mechanism (MARPLE), is able to take decisions in a decentralised manner and at near real-time. For benchmarking purposes, we also formulate and solve a mixed integer mathematical program (MILP) optimisation problem that provides the globally optimal offloading solution when all information is a priori known and centrally available. Unlike many decentralised or learning-based offloading methods, MARPLE avoids probing, model training, and inter-device synchronisation, relying instead on lightweight estimation over limited recent historical data-enabling practical deployment on resource-constrained IoT devices. Our results in a real IoT testbed show that our proposed solution outperforms the previous state of the art in terms of both the total response time and energy consumed on the IoT devices and can achieve close to optimal solutions.