Re: TCP RFC 6298 RTO handling

[Greg Troxel <gdt%lexort.com@localhost>]

Kevin Bowling <kevin.bowling%kev009.com@localhost> writes:

> Versus the LoC, this is a high value change that improves TCP
> performance in common [, tail] loss situations.
>
> https://www.netbsd.org/~kbowling/tcp-rfc6298.patch
>
> Even on high latency links, 1s is a good long while and a spurious RTO
> may be profitable since links are generally less bandwidth constrained
> while latency is what it is if you are dealing with tail loss in a
> buffering scheme.

I read the diff and the RFC.   Assuming a machine with the patch behaves
ok and more or less you've watched some connections with tcpdump, I
think is is good to commit.


Not related to this change, but I believe there is at least one issue in
our TCP code, where retransmission is less aggressive than recommended
by standards.  So if you are looking at traces, and you think you see
something, don't think you are wrong because the code must be right,
even more than you might have.