Estimating wireless communication delays in federated learning rounds
Hidden in Rounds: Predicting the Time Cost of 802.11 Contention in Federated Learning
Machine LearningDistributed, Parallel, and Cluster Computing
Summary
Federated learning lets many devices work together to train AI models without sharing their data directly, but they must send updates over a shared Wi-Fi channel, which can slow things down. The authors studied how long communication takes when many devices try to send updates over Wi-Fi by running many simulations. They created a simple way to estimate how long each learning round's communication will take without simulating every detail. Their method predicts communication time fairly well, but only in the tested cases. The study also looked at whether some devices get left out and found no clear difference in accuracy, though more data is needed.
What this means in practice
- •For network engineers: Predict Wi-Fi channel delays during federated learning to plan client participation and round timing without full simulations.
- •For machine learning platform developers: Estimate communication overhead for federated model updates over Wi-Fi to optimize federated learning scheduling and resource use.
Authors
Satwat Bashir, Tasos Dagiuklas
Abstract
Federated learning over IEEE~802.11 shares the wireless channel among clients that send model updates. We use ns-3 to measure the frame-delivery ratio and saturation throughput for different client densities and offered loads. A separate FedAvg trainer uses the frame-delivery ratio as a first-order proxy for the update-admission probability and uses an equation to estimate communication time. The method does not simulate the delivery of a complete model update or measure end-to-end training time. Across 720 evaluated runs with two datasets, two data partitions, six client densities, six offered loads, and five seeds, all runs reached their predefined target accuracy within the round budget. Rounds-to-target changed little with offered load, while communication time-to-target increased by about two orders of magnitude across the client-density range. A Bianchi-anchored estimator produced a mean absolute percentage error from $2.3\%$ to $10.2\%$ on held-out configurations. This error is measured against communication time constructed from the same round-duration equation, not against independently measured completion time. We also compare uniform participation with persistent heterogeneous participation. The study does not detect a statistically distinguishable excluded-class accuracy gap over five seeds, but the confidence intervals are wide. The results apply only to the evaluated configurations and do not provide a general convergence or fairness guarantee.