Optimisation of Service Placement in Multi-Access Edge Computing

Multi-access edge computing, driven by the development of 5G communications, pushes network and computing resources closer to mobile users. By providing wireless access through base stations, various computation tasks can be offloaded from user equipment and executed on servers in nearby micro-datacentres. The proliferation of Internet of Things sensors, smart devices, and diverse applications leads to a surge of user-submitted queries that must be processed at the network edge, necessitating effective resource management.

In practice, many modern applications are not monolithic but are composed of connected components with distinct processing requirements and data dependencies. To better utilise limited edge resources and schedule these complex applications effectively, such task dependencies must be incorporated into resource allocation decisions.

For service providers, maintaining Quality of Service is a central goal. Beyond providing basic computing and storage, it is crucial to enhance the performance of data-intensive, latency-sensitive, and CPU-intensive applications. Key metrics reflecting Quality of Service include end-to-end latency, service throughput, scalability, and failure rate, all of which present significant challenges in dynamic edge computing environments.

This dissertation first addresses the challenge of scheduling dependent services. User queries are modelled as directed acyclic graphs to capture inter-task dependencies. Two novel heuristic-based algorithms generate placement plans for these application modules. By identifying the critical path of the directed acyclic graph and evaluating server load alongside estimated latencies, the algorithms strategically assign modules to appropriate micro-datacentre servers, co-optimising end-to-end latency and server energy consumption.

The second part explores integrating energy harvesting technologies to establish a sustainable edge computing environment. To address limited and intermittent renewable energy in hybrid multi-access edge computing systems, an online algorithm dynamically manages edge server frequency scaling and cloud offloading. This strategy supplements local generation by offloading tasks to the cloud or purchasing grid power, promoting performance stability while reducing operational costs.

The third part tackles the instability of systems powered purely by harvested energy, where severe mismatches occur between volatile energy supplies and dynamic application demands. A novel online resource scheduling and energy management scheme is proposed to adaptively coordinate Dynamic Voltage and Frequency Scaling with horizontal service migration, maintaining stable performance in an edge system operating entirely on green energy.

The efficacy of the proposed algorithms and management schemes is validated through extensive simulation-based experiments against state-of-the-art baselines. The extended simulation framework is released as an open-source contribution. Results demonstrate that the developed algorithms significantly reduce energy consumption, minimise latency, and lower operational costs, contributing to an efficient and sustainable edge computing ecosystem.