Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BouquetFL: Emulating diverse participant hardware in Federated Learning

Arno Geimer

TL;DR

BouquetFL addresses the lack of hardware heterogeneity in Federated Learning experiments by enabling CPU, RAM, and GPU constraint emulation on a single host within the Flower framework. It offers a hardware sampler based on the Steam survey to generate realistic client profiles and enforces hardware limits via subprocess environments, making heterogeneous FL experiments accessible and reproducible. Quantitative validation shows BouquetFL accurately preserves the relative performance ordering across consumer GPUs, evidenced by a high Spearman correlation ($\rho = 0.92$) and Kendall's $\tau = 0.80$, while also capturing memory-related failures and data-loading variations. This approach provides a practical, scalable alternative to multi-device testbeds and can inform hardware requirements and deployment strategies for FL in real-world settings.

Abstract

In Federated Learning (FL), multiple parties collaboratively train a shared Machine Learning model to encapsulate all private knowledge without exchange of information. While it has seen application in several industrial projects, most FL research considers simulations on a central machine, without considering potential hardware heterogeneity between the involved parties. In this paper, we present BouquetFL, a framework designed to address this methodological gap by simulating heterogeneous client hardware on a single physical machine. By programmatically emulating diverse hardware configurations through resource restriction, BouquetFL enables controlled FL experimentation under realistic hardware diversity. Our tool provides an accessible way to study system heterogeneity in FL without requiring multiple physical devices, thereby bringing experimental practice closer to practical deployment conditions. The target audience are FL researchers studying highly heterogeneous federations. We include a wide range of profiles derived from commonly available consumer and small-lab devices, as well as a custom hardware sampler built on real-world hardware popularity, allowing users to configure the federation according to their preference.

BouquetFL: Emulating diverse participant hardware in Federated Learning

TL;DR

BouquetFL addresses the lack of hardware heterogeneity in Federated Learning experiments by enabling CPU, RAM, and GPU constraint emulation on a single host within the Flower framework. It offers a hardware sampler based on the Steam survey to generate realistic client profiles and enforces hardware limits via subprocess environments, making heterogeneous FL experiments accessible and reproducible. Quantitative validation shows BouquetFL accurately preserves the relative performance ordering across consumer GPUs, evidenced by a high Spearman correlation () and Kendall's , while also capturing memory-related failures and data-loading variations. This approach provides a practical, scalable alternative to multi-device testbeds and can inform hardware requirements and deployment strategies for FL in real-world settings.

Abstract

In Federated Learning (FL), multiple parties collaboratively train a shared Machine Learning model to encapsulate all private knowledge without exchange of information. While it has seen application in several industrial projects, most FL research considers simulations on a central machine, without considering potential hardware heterogeneity between the involved parties. In this paper, we present BouquetFL, a framework designed to address this methodological gap by simulating heterogeneous client hardware on a single physical machine. By programmatically emulating diverse hardware configurations through resource restriction, BouquetFL enables controlled FL experimentation under realistic hardware diversity. Our tool provides an accessible way to study system heterogeneity in FL without requiring multiple physical devices, thereby bringing experimental practice closer to practical deployment conditions. The target audience are FL researchers studying highly heterogeneous federations. We include a wide range of profiles derived from commonly available consumer and small-lab devices, as well as a custom hardware sampler built on real-world hardware popularity, allowing users to configure the federation according to their preference.
Paper Structure (9 sections, 2 figures)

This paper contains 9 sections, 2 figures.

Table of Contents

  1. Introduction
  2. Functionalities
  3. Integration With Flower
  4. Representative hardware sampler
  5. Limitations
  6. Experimental validation of BouquetFL
  7. Experimental setup
  8. Results
  9. Conclusion

Figures (2)

  • Figure 1: Execution flow of BouquetFL within the Flower framework. The ServerApp distributes global model parameters to each ClientApp as usual. When the client’s fit method is invoked, BouquetFL creates a dedicated subprocess environment that limits effective GPU compute share via CUDA MPS and applies clock speed and memory restrictions. The client performs data loading and local training under these constraints, then forwards the resulting update back to the main Flower process, which resets all hardware limits before the next round.
  • Figure 2: Comparing the relative performance of BouquetFL-simulated GPUs to real-world video game benchmarks, both normalized around their mean. Lower values mean better performance. Left: Scatter plot comparing GPU training performance under BouquetFL to normalized gaming benchmark results. Each point represents one GPU; the diagonal indicates perfect agreement. Right: Normalized performance trends grouped by GPU generation.