Jason Thorpe <thorpej@nas.nasa.gov> writes:

On Wed, 8 Apr 1998 15:41:07 -0700 (PDT) 
 Jonathan Stone <jonathan@DSG.Stanford.EDU> wrote:

> > But, as I understand it, only for multi-homed hosts where people want
> > to do PMTU, or to optimize better for TCP peers which are doing PMTU.

>So, which of your radio/ethernet laptops don't do Path MTU Discovery?

Right now? Today? All of them, AFAIK :)

But, again, the problem *isn't* the laptops themselves. 

The problem is all the third-party hosts on the local (class A!?)  net
which don't do PMTU, and which are run by different parts of Stanford,
and which are effectively completely separate organizations.

I have no input on, or control over, what these machines do, but I do
need to use them. Yes, this stuff is often ancient, like the IBM RT my
room-mate's group uses as their printserver, or like the registry
system on an IBM mainframe. (there's now a separate web frontend to
that, but I dont' think the host for _that_ does PMTU, either).

This isn't uncommon for academic sites.  Or,from what I've heard, from
a lot of corporate sites, either: people need to interopreate with
what we would think of as legacy systems.

That's why I keep saying ``Be conservative in what you send''.


If that's longwinded, or condescending, apologies; I'm just trying to
be clearer about why I think this issue is worth fixing.


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