BBR's Sharing Behavior with CUBIC and Reno

TL;DR

This work assesses how well two foundational theoretical models—Ware's steady-state model and Scherrer’s fluid model—predict BBR's sharing behavior with loss-based flows across BBR versions. By conducting large-scale FABRIC experiments for BBRv1, BBRv2, and BBRv3, the authors identify regimes where each model is accurate (notably steady-state for BBRv1 with single loss-based flows, and fluid for BBRv1/v2 in shallow/moderate buffers) and where they fail (deep buffers, many flows, intra-flow fairness, and especially BBRv3). The results show clear discrepancies for BBRv3, suggesting the need for updated or new modeling approaches to accurately capture its behavior. Overall, the study confirms the relevance of the older models in specific contexts while highlighting their limitations for the latest BBR version and prompting further investigation and model development.

Abstract

TCP BBR's behavior has been explained by various theoretical models, and in particular those that describe how it co-exists with other types of flows. However, as new versions of the BBR protocol have emerged, it remains unclear to what extent the high-level behaviors described by these models apply to the newer versions. In this paper, we systematically evaluate the most influential steady-state and fluid models describing BBR's coexistence with loss-based flows over shared bottleneck links. Our experiments, conducted on a new experimental platform (FABRIC), extend previous evaluations to additional network scenarios, enabling comparisons between the two models and include the recently introduced BBRv3. Our findings confirm that the steady-state model accurately captures BBRv1 behavior, especially against single loss-based flows. The fluid model successfully captures several key behaviors of BBRv1 and BBRv2 but shows limitations, in scenarios involving deep buffers, large numbers of flows, or intra-flow fairness. Importantly, we observe clear discrepancies between existing model predictions and BBRv3 behavior, suggesting the need for an updated or entirely new modeling approach for this latest version. We hope these results validate and strengthen the research community's confidence in these models and identify scenarios where they do not apply.

Figures (6)

• Figure 1: Topologies.
• Figure 2: Aggregate link fraction captured by BBRv1 flows sharing a bottleneck with a loss-based flow. The gray dashed line indicates fair sharing. Experimental results are compared with predictions from the steady-state model 2019IMC and fluid model scherrer.
• Figure 3: Aggregate link fraction captured by BBRv2 flows sharing a bottleneck with a loss-based flow. The gray dashed line indicates fair sharing. We compare the predictions of the fluid model for BBRv2 scherrer to experimental results.
• Figure 4: Fluid model prediction and experiment results when BBR shares a bottleneck with an equal number of flows of another type (different BBR version or loss-based CC), or when all flows sharing the bottleneck are the same.
• Figure 5: Aggregate link fraction captured by BBRv3 flows sharing a bottleneck with a loss-based flow. The gray dashed line indicates fair sharing. We compare results to BBRv1's steady-state model 2019IMC and BBRv2's fluid model scherrer.