Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces a formal framework for assessing functional stability in edge computing systems, addressing the inadequacies of traditional reliability models that focus on binary states. By shifting the analysis from the system level to individual functions, the authors define strong and weak forms of stability through a continuous quality function and establish key parameters and aggregate metrics for evaluation. The framework is validated through case studies involving UAV swarms and edge AI inference services, revealing critical distinctions in architectural performance that classical models fail to capture.
Edge computing systems can achieve nuanced performance insights through a novel framework that quantifies functional stability beyond traditional reliability metrics.
Edge computing systems operate in conditions where component failures, intermittent backhaul connectivity, resource exhaustion and adversarial disturbances are part of the normal operating envelope rather than rare events. Classical reliability and fault-tolerance models, oriented toward monolithic systems and binary working/failed states, do not capture the differentiated quality requirements of services co-located on a resource-constrained edge node. This article proposes a formal framework of functional stability that shifts the unit of analysis from the system to the individual function. Strong and weak forms of stability are defined through a continuous quality function and per-function quality thresholds. A system of base parameters (disturbance tolerance, recovery time, degradation depth, admissible degradation) and aggregate metrics (cumulative quality, functional availability, integral stability) is introduced and connected to the formal definitions by nine theorems. The framework is instantiated for edge computing scenarios and illustrated through two analytical case studies: a UAV swarm and an edge AI inference service with cloud fallback. The proposed metrics distinguish architectural alternatives that classical availability cannot separate and expose the dependence of architectural ranking on the explicitly selected disturbance-class set.