Design and Evaluation of a Decentralized Service Selection Model for Multi-Access Edge Computing

Edge computing redefines today’s service architectures by moving application services in proximity to end users. This lowers latency, mitigates jitter, and reduces packet loss through fewer and less congested communication links. It also satisfies stronger privacy constraints, as user data can be processed at its origin. Multi-Access Edge Computing (MEC) defines an architectural model where edge resources are accessed via multiple access networks such as wired landline, cellular, or Wi-Fi. Mobile users, considered independent of network operators, may be served by multiple operators and access technologies, resulting in a heterogeneous and dynamic network topology. However, frequent topology changes in highly mobile settings (e.g., urban vehicular mobility and dynamic deployments) create special challenges for edge-dependent applications that rely on continuous network performance: the optimal edge is altered suddenly or gradually over time. A supporting service selection method is crucial for the application for fast and robust decisions. In this thesis, simulations and measurements in a Fifth Generation (5G) cellular network show that mobility and environmental impacts cause a significant degradation of the connectivity, also affecting network-based service selection methods such as Domain Name System (DNS) lookups. To address these challenges, a novel decentralized and client-based service selection approach accounting for the specific demands of edge-enabled applications, called meQ, is proposed. The service and network topology is structured as a network graph and distributed to clients using an interest-based model. meQ allows for rapid decisions to identify the currently best service: possible targets are scored using multiple metrics, and selected without recurring remote queries. A system-level simulator is developed to evaluate the model’s performance through several simulation studies. Experimental results show that meQ significantly shortens adaptation times after service deployments and cell handovers: with information being locally present, it reacts within microseconds to changes, saving valuable time compared to DNS lookups. Even in larger networks, meQ remains computationally feasible. Overall, meQ noticeably improves responsiveness and robustness of fast service selection decisions in edge-enabled environments.