Receive Buffer Pre-division Based Flow Control for MPTCP

Han, Jiangping; Xue, Kaiping; Hong, Peilin; Yu, Nenghai; Li, Fenghua

doi:10.1007/978-981-10-8890-2_2

Cited by 1 publication

(1 citation statement)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Packet scheduling policies are designed for distributing data packets over multiple paths to alleviate the packet reordering issue, and thus improve the flow completion time and network throughput. Several packet scheduling schemes have been proposed to achieve these goals, such as Blocking Estimation-based MPTCP Scheduler (BLEST) [17], Out-of-order transmission for in-order arrival scheduling policy (OTIAS) [18], Delay-Aware Packet Scheduling (DAPS) [19], Forward Prediction Scheduling (FPS) [20], Fine-grained Forward Prediction based Dynamic Packet Scheduling (F 2 P − DPS) [21], Offset Compensation based Packet Scheduling (OCPS) [22], DMPTCP [23], Receive Buffer Pre-division based flow control mechanism (RBP) [24], Forward Prediction based Dynamic Packet Scheduling and Adjusting with Feedback (DPSAF) [25].…”

Section: Related Work and Existing Problemsmentioning

confidence: 99%

Tuning the Aggressive Slow-Start Behavior of MPTCP for Short Flows

et al. 2019

Self Cite

View full text Add to dashboard Cite

With the widespread availability of multi-homed devices, enabling multiple paths by utilizing multipath TCP (MPTCP) in current networks is a common practice to improve the performance and robustness. Although MPTCP improves both bandwidth efficiency and network reliability, the transmission performance for short flows can become worse than regular single path TCP when the concurrent subflows transferred through a shared bottleneck due to MPTCP's aggressive slow start behavior, which is uncoupled, and each subflow behaves independently as regular TCP, affecting MPTCP and concurrent traffic at the bottleneck. In this paper, we first reveal that the MPTCP's aggressive behavior in slow start causes timeouts and throughput collapse. We further present the design and implementation of GSAM, which employs the theoretical analysis to derive the appropriate threshold for smoothing the congestion window growth in GSAM according to the network conditions in slow start phase and leverages congestion detection and control at end-host to avoid buffer overflow under concurrent MPTCP connections with multiple subflows sharing the bottleneck. The experimental results based on the real implementations show that the GSAM reduces the completion time by up to 80% while retaining high Goodput for large flows as MPTCP.

show abstract