tech-net archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: [patch] bug fix & TCP networking performance improvements

Matthias Drochner <> writes:

> said:
>> I believe this does in fact fix the bug we brought up in our original
>> discussion -- that shifting srtt left by 2 as the first step of the
>> computation means it is adjusted by 1/2 RTT rather than 1/8.
> Looking at the patch and the CVS history I'm getting the impression
> the the change which introduced it (tcp_input.crev.1.16) was an attempt
> to change the representation of srtt so that it keeps 5 bits after
> the point, thus increasing precision. It was just not communicated
> correctly, and likely incomplete.
> But if done correctly, it shouldn't have bad effects.

I think that's what happened.

It's almost correct; I'm about to check in a large patch that changes
only comments and explains things - it's quite hard to figure out.

I believe there are only three things wrong; I'll post another note
about that post-commit.


Attachment: pgpwGKiQvdVr9.pgp
Description: PGP signature

Home | Main Index | Thread Index | Old Index