Dynamic Workload Offloading Framework for IoT-Edge-Cloud Architectures in Smart Cities
Abstract
Smart city infrastructures generate massive streams of time-sensitive data from distributed Internet of Things (IoT) devices, ranging from traffic management sensors and autonomous transit to public safety surveillance. Processing these workloads exclusively in centralized cloud data centers leads to unacceptably high network latency, backhaul bandwidth congestion, and severe energy consumption. Conversely, relying solely on edge nodes introduces dynamic bottlenecks due to constrained compute, memory, and battery capacity. To resolve this dilemma, this paper proposes a novel Dynamic Workload Offloading Framework (DWOF) tailored for multi-tier IoT-Edge-Cloud architectures. DWOF formulates task offloading as a joint constrained optimization problem that minimizes processing latency, energy dissipation, and operational cost under dynamic channel state conditions and heterogeneous task constraints. We introduce a Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm augmented with dynamic task partitioning and priority-aware queueing. Extensive trace-driven simulations demonstrate that DWOF achieves up to a 38.4% reduction in end-to-end task execution delay, a 29.1% improvement in edge node energy efficiency, and a 42.5% decrease in task deadline violations compared to baseline heuristic and static offloading schemes.