Skip to content
BBR: Congestion-Based Congestion Control 4 min

Neal Cardwell, Yuchung Cheng, C. Stephen Gunn, Soheil Hassas Yeganeh, Van Jacobson · ACM Queue 2016

BBR: Congestion-Based Congestion Control

Google's way to send fast without filling buffers: measure the path's width and length instead of waiting for loss. Race it against loss-based control.

Read the original paper

Networking · Yellow Belt · Pro · 4 min read

BBR: Congestion-Based Congestion Control

Every time you stream a video or load a page, your computer and the server must decide how fast to send. Too slow wastes the connection. Too fast clogs it, and packets are delayed or dropped. The rule that decides this is called congestion control, and for decades the standard answer was simple: speed up until a packet is lost, then slow down.

A team at Google wrote an article in 2016 arguing that this old idea no longer fits today's networks. When network buffers are large, loss based control fills them, causing long delays called bufferbloat. When buffers are small, it mistakes ordinary drops for congestion and slows down too much. Their answer, BBR, does not wait for loss. It measures the path.

BBR continuously estimates two numbers: the bottleneck bandwidth (BtlBw) and the round-trip propagation time (RTprop). It sends at the bottleneck rate and keeps about one bandwidth-delay product of data in flight. That fills the pipe without filling the buffer, so throughput stays high and delay stays low.

A path is a pipe

Whatever a connection crosses, the paper says, from TCP's point of view it behaves like a single link with the same round-trip time. Two physical facts describe it. RTprop is the round-trip time when nothing is queued. BtlBw is the speed of the slowest link, the bottleneck.

The paper describes any network path as a single link with the same round-trip time. Two numbers explain it: the bottleneck bandwidth BtlBw and the round-trip propagation time RTprop. Their product, the BDP, is how much data fits in the pipe.

Their product, BtlBw x RTprop, is the bandwidth-delay product (BDP). It is exactly the amount of data that fits in the pipe. With less than a BDP in flight, the link sits partly idle. With more, no extra data gets through. The excess just waits in the bottleneck buffer.

The paper points out that the left edge of the bandwidth-limited region, at exactly one BDP in flight, is the best place to run. Leonard Kleinrock showed in 1979 that this operating point maximizes delivered bandwidth while minimizing delay and loss. Loss-based control sits at the right edge instead, with the buffer full.

Keep reading with Pro

This paper is part of Pro. Pro and Architect open the full breakdown, with animated diagrams, interactive elements you can play with and a quiz. A few papers stay free to read in full.

Still to read

2 more sections, 2 more diagrams and 2 interactive elements to play with

  • Measure the two numbers
  • What it did in practice

Found this breakdown useful?

Share it with someone else wrestling with this paper.

Discussion0

Join the Discussion

Sign in to leave comments, reply to others, or like insights.

Sign In to ScaleDojo

No comments yet. Be the first to start the thread!

Enjoyed this? Get more like it.

New paper breakdowns, levels, and one concept worth knowing, straight to your inbox.

No spam, ever. Unsubscribe in one click.

Enjoyed this content?

Your support keeps us creating free resources

We put a lot of hours into researching and writing these guides. If it helped you, consider buying us a coffee. Every bit goes toward keeping ScaleDojo's content free and growing.

$

One-time payment via Stripe. ScaleDojo account required.