Abstract Fog computing extends cloud services to the network edge, enabling low-latency processing for time-sensitive applications. However, scheduling complexity significantly increases due to heterogeneous resources, dynamic workloads, and strict Quality-of-Service (QoS) constraints. Although numerous scheduling techniques have been proposed, existing studies often assess only a narrow subset of algorithms or rely on offline metaheuristics unsuitable for real-time environments. This paper presents a comprehensive comparative evaluation of twelve scheduling algorithms, including five classical, one heuristic, and six metaheuristic-inspired approaches, within a realistic 20-node multi-tier fog-cloud topology. Across 1,365 experiments spanning seven utilization levels, we analyze each algorithm’s deadline adherence, load distribution, and scheduling efficiency. We adapt RT-MOMFO, a real-time variant of Multi-Objective Moth-Flame Optimization (Salehnia et al., 2023), for online fog scheduling through parameter reduction (population=20, iterations=8), ready queue management, and periodic tier preference cycling. This engineering adaptation achieves an 8.1% deadline miss ratio at U=0.8 with near-perfect load balance (0.988), ranking second overall. We also develop GMO (Greedy Multi-Objective), a lightweight greedy scheduler that eliminates population-based mechanisms while retaining load-aware routing and periodic tier cycling. GMO achieves a 9.4% DMR at U=0.8 with 47 times faster decisions (2 ms vs. 94 ms), demonstrating that speed advantages can offset algorithmic simplicity. The results establish a clear performance hierarchy, highlight the importance of explicit fog-cloud tier management, and show that simple greedy approaches remain competitive when perfect load balance is prioritized. Statistical analysis confirms top-tier algorithms (GAMMR, PSG-M, RT-MOMFO) are statistically indistinguishable at critical utilizations, with comprehensive ablation studies validating each design choice. This study provides one of the most extensive evaluations to date with rigorous statistical validation.
Azmi et al. (Tue,) studied this question.