Subject: Re: perhaps time to check our TCP against spec?
To: Jonathan Stone <jonathan@DSG.Stanford.EDU>
From: Kevin M. Lahey <firstname.lastname@example.org>
Date: 04/07/1998 23:50:04
In message <199804080223.TAA06694@Kowhai.Stanford.EDU>Jonathan Stone writes
>>Since you'll all just run out and read the archive, I know
>>I better point out that it looked to me like the consensus on
>>a conservative value was the interface MTU, rather than
>But where I'm coming from is that the correct thing to do _is_ the
>conservative option. I'm not saying those who, like you guys at NAS,
>have HIPPI and FDDI around, cannot turn on if_maxmtu; but the default
>should be ``Be conservative in what you send'', &c.
>That's so fundamental to what's considered good Internet engineering
>that I really dont understand why we're fighting about it.
I really thought that discussion would dry up after my message,
rather than get worse.
I think it would be instructive to go back and look at the discussion
in the mail archive I pointed everybody at. I thought that there
was enough of a case made for the interface MTU advertisement
that it was appropriate to support it. I'm not sure that the
case was made that it was THE right thing to do.
Link layers are supposed to be able to fragment.
What are you going to do with your Metricom when IPv6 requires
1280 byte packets? Send a smaller MSS? Not an option!
>That may be so. But those other TCP wisards don't have to put up with
>Jason Thorpe accusing them of being unprofessional, or of being
>incompptent, a liar, or of faking their data at every turn, now do
Jason and you obviously have your problems. However, if instead of,
"NetBSD's TCP is broken and the sky is falling and it is Jason's fault",
you'd written something like:
Hmmm. Check out this tcpdump output for the loopback interface:
I noticed that the MSS advertised was only 1.5K, when it used to be 32K.
I think this changed about the time that the syn cache changes were
committed. Shouldn't it have remained 32K?
Also, BTW, I noticed that it looks like the MSS advertised is the
largest of the attached interfaces. This breaks for me when dealing
with my Metricom radio modems in the following topology. Is there
any way you can think of to fix this? I'm not sure about this
I suspect that this would've been fixed up, or mitigated in some
way, without two full days of angst-filled messages! If you'd
supplied the few dozen lines of patches that it'll take me to
implement this, all the better, and you'd have saved yourself
six months of problems! If you'd taken the time, as Jason and
I have, to follow tcp-impl for the last year, you'd have been far
better equipped to discuss this intelligently.
I think that, just like me, you're a well meaning person who wants
the NetBSD stack to be solid. Alas, you've gone about making this
happen in the very hardest way I can imagine, and created lots
of hard feelings and unpleasant conflict.