Observability Aware Federation Learning in Fog-IoT Systems Round Time and runtime Metrics Analysis through iFogSim-Based implementation

Federated learning (FL) enables the training of models on distributed Internet of Things (IoT) and localization of them. Practically, Fog-IoT system More infrastructure differences exist in infrastructure, including differences in communication lag times, differences in access to fog resources, and differences in client activity that introduce much uncertainty in their effects on the length of the federation rounds, and the available literature is more focused on the effect of infrastructure dynamics on the execution of learning behavior than on how to optimally execute the federation process itself. This research aims to design an observability-enabled federated learning framework used to examine how varying conditions in heterogeneous Fog-IoT infrastructure affect the performance of federated learning. The proposed framework incorporates TensorFlow-enabled federated learning with the iFogSim2-based Fog-IoT simulation in order to capture heterogeneous devices, variability in communication, constraints in fog resources, as well as passive telemetry from training, systems, and behavioral layers. Experiments shown that communication latency and constrained fog resources greatly affect the execution of federated learning, since the time needed to complete each round is increased from 4.2-5.3 s in baseline conditions to 6.1-8.8 s in high latency conditions and 5.3-7.9 s in constrained fog resources. In addition, increased utilization of fog resources from 30% to 90% causes the ratio of stragglers to increase from 0.08 to 0.38. This set of findings proves that the heterogeneity of infrastructure is among the major predictors of FL performance and shows the viability of calculating observability to Fog-IoT implementation.