On Wed, 08 Apr 1998 17:29:21 -0700, Jason Thorpe <thorpej@nas.nasa.gov>

On Wed, 08 Apr 1998 17:30:05 -0700 
 Jonathan Stone <jonathan@DSG.Stanford.EDU> wrote:

 > Someone also reminded me in private email that it's not entirely
 > uncommon for firewalls to break PMTU, too....

Again, *WHY* are you still ignoring the valid points I *did* make, in
the same message, and have a shooting party at a minor point?

If you don't answer the main point, what am I supposed to assume?

  * That you still misunderstand the point I'm making?
  * That you've read it and that you now see that you misunderstood earlier?
  * that you've ignored that part of the message?

Is it really that hard to admit you misunderstood and to admit that
you made a mistake?  Why else do you totally ignore my main point, and
answer only a minor point, like the following:


>While the general consensus in the applicable WGs is that such firewalls
>are COMPLETELY BROKEN, Black Hole Discovery is a somewhat orthogonal
>problem to Path MTU Discovery.

Yeah, agreed. Didn't I _say_ that?  But so what?

The point here, though, is that in_maxmtu advertisement are orthogonal
to whether either host is acutally *doing* PMTU. *OR* blackhole
discovery. And here too, in_maxmtu can cause lossage which didn't
happen before.

``Be conservative in what you send'' means being _conservative_ even
when the RFCS say you _may_ do something, in order to avoid breaking
interoperability with some poor bugger who isn't fully in conformance
with the RFCs.

That's why existing practice and ``be conservative in what you send''
*have precedence* over a literalist interpretation of the RFCs.

If such firewalls do exist, and if NetBSD boxes fail to work through
them while other, non-in_maxmtu systems work through them just fine---
who do you think the sysadmin who's got a misconfigured firewall is
going to point the finger at?

Same thing with the lossage in_maxmtu can cause when talking to hosts
that don't do PMTU.

It's much, much easier to avoid lossage in the first place, by
conservative engineering, than it is to educate people who start
saying ``Netbsd is broken, it doesn't work in situation X''.  OK?