TCP BBR default in Ubuntu 24.04 (Linux 6.8) possibility?

[Daniel J Blueman]

Hi Neil et al,

Firstly, I appreciate all the fine work that has gone into BBR and QUIC to overcome the challenges of reasonable goodput and lower latency in the face of non-deterministic path delays, packetloss and related. The benefits are even higher in countries further from key datacenters vs CUBIC, especially with more density/radio frequency congestion (WiFi).

As the BBR implementation stands today in the Linux kernel (moreover the 6.8 release that will be used), is there an argument for making this the default congestion control algo in Ubuntu 24.04, or some BBRv2+ changes need backporting for improved fairness?

Thanks,
  Daniel

[Neal Cardwell]

Hi Daniel,

To make sure we're all on the same page: currently the TCP BBR code in Linux is BBRv1. We are working on getting BBRv3 upstream into Linux TCP.

BBRv1 is definitely not ready to be the default on any Linux distribution. Whether BBRv3 is ready to be a distribution default is arguable. Personally I would prefer to see BBRv3 improve with respect to Reno/CUBIC coexistence and keeping loss rates low before arguing that it is ready to be the default.

best regards,

neal

--
You received this message because you are subscribed to the Google Groups "BBR Development" group.
To unsubscribe from this group and stop receiving emails from it, send an email to bbr-dev+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/bbr-dev/1db89c14-39b2-4ceb-862f-813df834f3bbn%40googlegroups.com.

[Neal Cardwell]

Hi Daniel,

Thanks for closing the loop. I appreciate it!

best,

neal

On Wed, Jan 31, 2024 at 7:14 AM Daniel J Blueman <dan...@quora.org> wrote:

Hi Neal,

Well characterised, and agreed the fairness/coexistence should be a
good reason to hold off making this default until BBRv3 is mainlined.

I'll keep an eye out for the kernel patches and we can review in a
later distro cycle.

Many thanks!
  Dan

--
Daniel J Blueman