A LATENCY-AWARE TASK OFFLOADING AND RESOURCE ALLOCATION FRAMEWORK FOR FOG-EDGE-CLOUD COLLABORATIVE COMPUTING IN REAL-TIME IOT APPLICATIONS

Abstract

The exponential growth of latency-sensitive Internet of Things (IoT) applications such as autonomous vehicles, remote surgery, industrial automation, and real-time healthcare monitoring has exposed the limitations of centralized cloud computing, where geographic distance induces unacceptable transmission delays. This review presents a latency-aware task offloading and resource allocation framework for fog-edge-cloud collaborative computing, designed to minimize end-to-end latency while optimizing resource utilization in heterogeneous, multi-tier architectures. The proposed framework integrates dynamic offloading decisions based on task urgency, device mobility, network conditions, and computational load, employing predictive models (LSTM, spatio-temporal graph neural networks), deep reinforcement learning (DRL), and heuristic algorithms to orchestrate task partitioning, service migration, and proactive container relocation. Key mechanisms include mobility-aware service migration using Chebyshev graph convolutional networks, priority-based queuing at edge nodes, and adaptive resource provisioning that balances energy, bandwidth, and latency constraints. Simulation and real-world evaluations demonstrate 30–65% reductions in average latency, 20–45% improvements in task success rates under high mobility, and enhanced system throughput compared to baseline cloud-only or static offloading schemes. The framework addresses critical challenges such as intermittent connectivity, resource heterogeneity, and security in fog-edge environments, offering a scalable, proactive approach to support ultra-reliable low-latency communication (URLLC) requirements in 5G/6G-enabled IoT ecosystems.