FedDES: A Discrete-Event Simulator For Large-Scale Federated Learning
Keywords: Federated Learning, Discrete Event-Driven Simulations
Abstract: We introduce FedDES, a performance simulator for Federated Learning (FL) that leverages Discrete Event Simulation (DES) techniques to model key events—such as client updates, communication delays, and aggregation operations—as discrete occurrences in time. This approach accurately captures the runtime features of FL systems, providing a high-fidelity simulation environment that closely mirrors real-world deployments. FedDES incorporates all three known aggregation settings: Synchronous (e.g., FedAvg and FedProx), Asynchronous (e.g., FedAsync and FedFa), and Semi-Asynchronous (e.g., FedBuff and FedCompass). Designed to be framework-, dataset-, and model-agnostic, FedDES allows researchers and developers to explore various configurations without restrictions. Our evaluations involving over 1,000 clients with heterogeneous computation and communication characteristics demonstrate that FedDES accurately models event distribution and delivers performance estimates within 2% error of real-world measurements. While real-world workloads often take hours to evaluate, FedDES generates detailed, timestamped event logs in just few seconds. As a result, FedDES can significantly accelerate FL developing and debugging cycles, enabling developers to rapidly prototype and evaluate algorithms and system designs, bypassing the need for costly, time-consuming real-world deployments. It offers valuable performance insights—such as identifying bottlenecks, stragglers, fault-tolerance mechanisms, and edge-case scenarios—facilitating the optimization of FL systems for efficiency, scalability, and resilience.
Supplementary Material: pdf
Primary Area: infrastructure, software libraries, hardware, systems, etc.
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8762
Loading