Doktorandské kolokvium KAI - Negin Bagheri (27.4.2026)
v pondelok 27.4.2026 o 13:10 hod. v miestnosti I 9
Prednášajúci: Negin Bagheri
Názov: CADO: A Lightweight QoS-Aware Runtime Offloading Mechanism for Multi-Layer Service Execution across the Device–Fog–MEC–Cloud Continuum
Termín: 27.4.2026, 13:10 hod., I/9
Abstrakt:
Modern distributed IoT and AI-oriented applications increasingly rely on latency-sensitive service execution across heterogeneous edge–cloud infrastructures. In such environments, deciding where a task should run is a system-level QoS challenge, since runtime performance depends not only on computational capacity, but also on communication delay, queue buildup, congestion, and deadline feasibility. This work presents CADO (Congestion-Aware Deadline-Driven Offloading), a lightweight and interpretable runtime offloading mechanism for a four-layer Device–Fog–MEC–Cloud continuum. For each arriving task, CADO estimates end-to-end completion time at each candidate layer and first selects the best deadline-feasible option; when no safe option exists, it applies a risk-aware fallback rule. Unlike learning-based approaches, CADO requires no training phase or centralized optimization. The mechanism is evaluated in a discrete-event simulation under nominal conditions and two stressed scenarios, including infrastructure degradation and AI-heavy workloads. The results show that CADO remains competitive under nominal conditions while improving latency and deadline robustness under stress. This work forms an important part of my doctoral research and provides the system-level foundation for future extensions toward Federated Learning over the same multi-layer infrastructure.