B3R: A New Approach to BBR Congestion Control for Shallow Buffers

The introduction of Bottleneck Bandwidth and Round-trip time (BBR), a state based Congestion Control Algorithm (CCA), has brought new insights into the way how modern network handles multitude of data across shared links for minimal delay and maximum bandwidth utilization. It claimed to be operating at Kleinrock’s optimal operating point without filling up buffers or having large packet losses. However, evaluation of BBR revealed that it displayed a high loss rate under certain operating conditions. The reason for high losses in BBR is the static increase of sending rate by a factor of 1.25 while probing for more bandwidth under small bottleneck buffer sizes, typically when buffer sizes are less than one times the product of Bandwidth and delay. High packet losses in short buffer sizes can lead to disruptive connections and poor application performance. To tackle this issue, we propose Bottleneck Bandwidth Buffer and Round-trip propagation time (B3R) which aims to actively estimate the bottleneck link buffer and then sets up a proper congestion control in short buffer scenarios to eliminate the high packet loss. The proposed model is evaluated on a testbed network, and the results showed that B3R reduced retransmissions by up to 80% as compared to BBR under short buffer scenarios.

I. Introduction

Congestion Control Algorithms (CCAs) regulate and control the flow of data in modern networks. In a typical network, there are multiple flows of data being transmitted through the same shared resources and CCAs regulate the flows in order to ensure uniform utilization of the available resources and prevent disruptions in the network. Packet loss has been used as the key indicator of congestion along network paths over the Internet for decades. Many CCAs based on packet loss and delay have evolved [ 1 – 5 ] over the years. However, BBR [6] [7] from Google (a state based CCA) was the only CCA that successfully converged the operating point to Kleinrock’s optimal operating point [8] , thus maximizing data transfer rate while reducing Round Trip Transmission Time (RTT). It sequentially probes for bottleneck bandwidth and minimum RTT. BBR then uses those measurements to deliver data at maximum available bottleneck bandwidth without creating an excess queue over the bottleneck link. With no development of an excess queue, buffer bloat [9] issue could be resolved.