Objections to TCP BBR by default in 24.04?

[Andrea Righi]

On Sat, Jan 20, 2024 at 06:39:20PM +0800, Daniel J Blueman wrote:
> Ubuntu has a distro-leadership opportunity to improve the end-user
> experience for much of the world on higher-latency connections. With
> TCP BBRv2 being materially unchanged for some time, it is widely
> considered mature.
> 
> If there isn't good rationale against this, I propose we enable TCP
> BBRv2 by default for the 24.04 development phase, moving back to Cubic
> if any issues are identified.
> 
> Comments/objections?

Hi Daniel, sorry for the late response, I had to investigate a bit about
BBRv2 (I was not very familiar with it).

Overall, it definitely looks interesting, but as far as I can see it has
not been merged upstream yet (I assume you are requesting to include
BBRv2, not just enable BBR by default, right?).

This can be a little problematic, because it puts on us the burden of
maintaining it (even if it is stable, mature and unchanged, we still
need to actively check for potential updates / security fixes).

Of course we can make exceptions to this rule, if there is a valid
motivation.

Hence, some questions:

 - Is there an open tracking bug for this in launchpad? (if not I can
   open it, I'm just trying to collect as many information as possible
   and put everything to a single place, so that we can better keep
   track of it)

 - Is there a active plan to include this upstream?

 - Can you provide some details/links about the advantages of BBRv2 vs
   Cubic? What could be a valid motivation for us to justify the
   inclusion of this patch/change?

 - Is there a specific test case / benchmark that we can run to get some
   numbers and prove the benefits of BBRv2 vs Cubic?

 - What is the regression potential of including this change and
   enabling it by default? What kind of specific workload(s) is(are)
   expected to regress?

Thanks,
-Andrea