Abstract

With the aim of improved throughput with reduced delay, Google proposed the bottleneck bandwidth and round-trip time (BBR) congestion control algorithm in 2016. Contrasting with the traditional loss-based congestion control algorithms, it operates without bottleneck queue formation and packet losses. However, we find unexpected behaviour in BBR during testbed experiments and network simulator 3 (NS-3) simulations. We observe huge packet losses, retransmissions, and large queue formation in the bottleneck in a congested network scenario. We believe this is because of BBR’s nature of sending extra data during the bandwidth probing without considering the network conditions, and the lack of a proper recovery mechanism. In a congested network, the sent extra data creates a large queue in the bottleneck, which is sustained due to insufficient drain time. BBR lacks a proper mechanism to detect such large bottleneck queues, cannot comply with the critical congestion situation properly, and results in excessive retransmission problems. Based on these observations, we propose a derivative of BBR, called “BBR with advanced congestion detection (BBR-ACD)”, that reduces the excessive retransmissions without losing the merits. We propose a novel method to determine an actual congestion situation by considering the packet loss and delay-gradient of round-trip time, and implement a proper recovery mechanism to handle such a congestion situation. Through extensive test and NS-3 simulations, we confirmed that the proposed BBR-ACD could reduce the retransmissions by about 50% while improving the total goodput of the network.

Highlights

  • Despite rigorous research on the packet loss problem during data transmission [1,2], a significant amount of packet losses are observed, especially in the wireless network

  • bottleneck bandwidth and round-trip time (BBR) steps into a steady-state phase where it cruises while sequentially measuring the bottleneck bandwidth (BtlBW) and minimum round-trip time (minRTT) by the following two states

  • We investigated the excessive retransmission problem of BBR in the case of a congested network scenario

Read more

Summary

Introduction

Despite rigorous research on the packet loss problem during data transmission [1,2], a significant amount of packet losses are observed, especially in the wireless network. BBR does not back-off properly and continues to send data at almost the same rate, the queue sustains and results in enormous packet losses with a significant increase in latency, which leads to a timeout event In such a timeout event, BBR resets its CWND to one MSS (maximum segment size) and starts the recovery process. BBR-ACD detects actual congestion events by considering the delay-gradient of RTT, and packet losses; and slows down the flow of data in the case of actual congestion events by halving the CWND. This allows the bottleneck to get rid of the excess queue, reduces increased latency due to bufferbloat, and ensures high link utilization.

BBR Overview
Motivation behind BBR
BBR Operation
Benefits of BBR
Excessive Retransmission Problem in BBR
Testbed Environment
Testbed Results
Simulation Environment
Evaluation by Simulation
BBR-ACD—A Solution to Reduce the Excessive Retransmissions
Detection of Congestion
Action Taken for Congestion
Performance Evaluation
Testbed results
Conclusions
Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call