Leveraging Edge Computing Resources for Ultra-Low Latency Services

Abstract

The rapid growth of applications requiring extremely low latency, such as immersive AR/VR, autonomous vehicle platooning, and remote multi-agent surgery, necessitates computational approaches beyond centralized cloud models. Edge Computing (EC) reduces communication delays by utilizing resources that are closer to end-users. However, meeting the strict demands of Ultra-Reliable Low-Latency Communication (URLLC) often requires pushing computation even further to the network edge, engaging end-user resources directly. This dissertation leverages the utilization of Extreme Edge Computing (XEC), an evolution of EC that leverages the often idle computational power within end-user devices such as smartphones, wearables, and IoT sensors. XEC aims to decrease service latency, democratize edge access by reducing reliance on centralized clouds, and rapidly scale computational capacity without large capital investments. This paradigm also offers opportunities to incentivize end-user participation. However, provisioning XEC faces many challenges due to the unpredictable churn and limitations of user-owned edge nodes, including constrained processing power, battery life, intermittent connectivity, and unpredictable availability, all of which introduce volatility and lower reliability. To address these complexities, this dissertation introduces a robust, adaptive provisioning framework designed for XEC’s dynamic conditions. The framework utilizes agile, context-aware resource allocation, predictive workload optimization, and proactive resilience mechanisms to ensure service continuity and stability amid uncertainty. It navigates fluctuations in device capabilities, network quality, energy levels, and user demands, significantly enhancing system reliability and fault tolerance. This dissertation presents the proposed framework’s architecture, development methodology, and evaluation metrics for EC and XEC contexts. By systematically mitigating challenges from resource heterogeneity and uncertainty, this research enables reliable, scalable, ultra-low latency services in extreme edge scenarios. It advances the state-of-the-art by offering a viable path to dependable computation at the network edge, enabling the next generation of URLLC applications.