[Ecn-sane] [bbr-dev] cwnd 4 in BBRv2?

[Neal Cardwell]

On Thu, Mar 28, 2019 at 6:10 PM Dave Taht <dave.taht at gmail.com> wrote:
>
> I just finished watching the sessions in ietf iccrg...
>
> and although I have many questions about BBRv2, I am curious if the
> cwnd in it is capped to a minimum of 4, not 2, still? BBRv1 is...
>
> https://github.com/torvalds/linux/blob/master/net/ipv4/tcp_bbr.c#L385

So far that part of the algorithm has not changed in BBRv2. So, except
for certain corner cases in loss recovery (e.g. cwnd=1 on RTO), the
min cwnd is generally 4.

But keep in mind that BBR generally tries to be pacing-limited, rather
than cwnd-limited. And below a cwnd of 4 delayed ACKs can inhibit
pipelining, causing stop-and-wait-type problems that significantly
hurt throughput.

best,
neal