Learning-Based Service Selection for Latency Optimization in Heterogeneous Fog Computing

Edge and fog computing has become an important sample for supporting the latency sensitive application like the Internet of Things (IoT) today. However, heterogeneous network conditions and dynamic workload make the static service selections have unexpected performance in practical deployments. This study researches the self-adaptive service selection problem. It describes the problem as a Multi-Armed-Bandit (MAB) decision process and implements an Upper-Confidence-Bound (UCB) strategy in the Yet Another Fog Simulator (YAFS) simulated framework and compares with the Round Robin baseline under the heterogeneous network conditions. The result shows that the UCB-based method achieves lower long-term latency and better behavior in service selection. These findings are consistent with the recent research in routing and learning-driven service placement in fog and edge environments.