TL;DR
This paper introduces "Coupled BBR," a novel congestion control algorithm, and "AR&P Scheduler," an adaptive packet scheduler, to significantly improve the performance of Multipath TCP (MPTCP) in challenging wireless networks. By combining BBR's bandwidth-aware approach with intelligent subflow management, the proposed system achieves higher throughput, better fairness, and increased robustness in dynamic and lossy environments.
What Problem Does This Paper Address?
Multipath TCP (MPTCP) aims to enhance network performance by utilizing multiple network paths simultaneously. However, its effectiveness is often hampered in wireless networks due to factors like random packet loss and rapidly changing link conditions. Traditional loss-based congestion control algorithms struggle in these environments, leading to inefficient bandwidth utilization and unstable performance. Furthermore, existing MPTCP solutions face challenges in balancing congestion across multiple subflows and ensuring fairness with single-path TCP flows. This paper addresses these limitations by proposing a new coupled congestion control mechanism and an adaptive scheduling strategy tailored for MPTCP in wireless and dynamic network scenarios.
Methods
The core contributions of this paper are:
- Coupled BBR: This is a modified version of the BBR (Bottleneck Bandwidth and Round-trip propagation time) congestion control algorithm, adapted for MPTCP. Unlike standard BBR which treats each subflow independently, Coupled BBR uses detected bandwidth from all subflows to adjust the sending rate of each individual subflow. This approach aims to achieve fairness with single-path TCP BBR flows and balance congestion among MPTCP subflows. It leverages BBR's periodic bandwidth detection to maintain high throughput and stability, even in the presence of packet loss.
- AR&P Scheduler (Adaptively Redundant and Predictive packet scheduler): This scheduler works in conjunction with Coupled BBR. It comprises two key components:
- AR-Scheduling: This component adaptively decides whether to send redundant packets on a subflow. If a subflow is experiencing poor conditions (low bandwidth, high RTT), it sends redundant packets to improve robustness. Otherwise, it focuses on sending non-redundant packets to maximize bandwidth utilization.
- P-Scheduling: This component predicts the arrival time of packets and schedules them one by one. It aims to minimize out-of-order packet delivery, especially in asymmetric networks with varying RTTs across subflows.
Both Coupled BBR and the AR&P scheduler are implemented within the MPTCP Linux kernel.
Datasets
Not discussed in the paper. The evaluation relies on testbed experiments and simulations.
Key Insights and Ideas
- BBR's potential for MPTCP: The paper highlights that BBR's bandwidth-centric approach, which doesn't solely rely on packet loss for congestion signaling, is well-suited for lossy wireless environments where traditional TCP struggles.
- The need for coupling: Simply applying standard BBR to each MPTCP subflow independently doesn't achieve optimal fairness or congestion balance. Coupling the congestion control across subflows is crucial.
- Adaptive redundancy for robustness: Sending redundant packets on problematic subflows, rather than just stopping transmission, can maintain subflow activity and improve overall network resilience during link failures or degradations.
- Predictive scheduling for order: By predicting packet arrival times, P-Scheduling can proactively manage packet ordering, significantly reducing out-of-order packets in asymmetric networks.
- Synergy between congestion control and scheduling: The paper demonstrates that a well-designed scheduler is essential to fully leverage the benefits of a new congestion control algorithm like Coupled BBR, especially in dynamic network conditions.
Key Results
The proposed Coupled BBR and AR&P scheduler demonstrate significant improvements over traditional MPTCP congestion control algorithms (LIA, OLIA, BALIA) and schedulers (Round-Robin, minRTT):
- Throughput: The system achieves higher throughput, with improvements of up to 2.5 times in normal wireless scenarios and over 10 times in scenarios with large RTT and loss.
- Fairness: Coupled BBR ensures fairness with single-path TCP BBR flows, matching their throughput on the best path.
- Robustness: AR&P scheduler shows faster recovery from path failures and better performance in dynamic network conditions.
- Out-of-order packets: The AR&P scheduler significantly reduces out-of-order packets, by up to 80% in asymmetric scenarios, compared to existing methods.
- Bandwidth Utilization: Coupled BBR achieves high bandwidth utilization, especially in lossy and long-delay networks.
Limitations and Future Work
- Complexity: While the greedy method for AR-Scheduling is used to reduce computational complexity, the paper acknowledges that it might not always yield the absolute optimal solution, especially with more than two subflows.
- Slow-start phase: The P-Scheduling algorithm is not used when all subflows are in the slow-start phase due to potential high estimation errors.
- Future Work: The authors suggest that a full BBR network environment would further enhance the benefits of their proposed MPTCP solution. They also note that achieving fairness between BBR and non-BBR flows in a mixed environment remains an area for further exploration. The paper implicitly suggests that further optimization of the AR&P scheduler's parameters and decision-making logic could lead to even greater gains.
Summary updated