Modeling Dynamic Trust for Cooperative Operations in Edge Computing Environments

---

Résumé

Collaborative edge computing promises low-latency intelligence by orchestrating microservices across heterogeneous, mobile, and intermittently connected nodes. Yet sustained cooperation is fragile without explicit, adaptive trust management. Existing trust models for edge computing, such as lightweight estimation approaches or blockchain-based collaborative schemes, address important aspects of trust computation but remain limited by the lack of contextual reasoning, insufficient robustness to advanced attacks, or the absence of integration within decision-making mechanisms such as cooperative orchestration or task migration. To fill this gap, we introduce DyTrust, a dynamic and context-aware trust model that combines behavioural analytics, anomaly detection, and trust-aware orchestration to enhance secure collaboration among edge nodes. DyTrust supports resistance to on–off behaviour, collusion, Sybil-like manipulations, and poisoning in distributed learning. We integrate DyTrust into a trust-aware, many-objective formulation of service placement and migration that balances latency, energy, migration cost, and trust; the search is solved with NSGA-III using reference-point selection. An implementation in iFogSim2 evaluates distributed DNN inference and federated learning workloads under realistic mobility and network conditions, with attack scenarios for recommender collusion and federated backdoors. The simulations show that DyTrust yields consistently stronger Pareto fronts (higher hypervolume, lower IGD), higher trust satisfaction at comparable QoS, and improved robustness to adversarial dynamics, while imposing modest control-plane overhead.We discuss limitations and outline directions for risk-aware objectives, privacy-preserving evidence gathering, and hardware-in-the-loop validation.

Mots-clés

Aucun mot-clé renseigné.