Subject: Re: perhaps time to check our TCP against spec?
To: None <thorpej@nas.nasa.gov>
From: Jonathan Stone <jonathan@DSG.Stanford.EDU>
List: tech-net
Date: 04/07/1998 20:11:43
On Tue, 7 Apr 1998 19:44:01 -0700 (PDT) 
 Jonathan Stone <jonathan@DSG.Stanford.EDU> wrote:

 > media (FDDI or HIPPI); and no old systems to support, want to turn on

>Hah!  I guess you've never seen our computer room!  It's pretty hard to
>not spot the Convexes :-)

Heck, Jason, for all I know you could still have the Amdahls and that
fluorocoarbon water fountain, uh, Cray-2;) You guys are more than
smart enough to support it all and keep everything working, whatever
changes you make.

But the world I'm in still has people using IBM RTs as group-level
servers, and support staff that (i'd guess) just isn't as good.

I can hack *my* laptop so's it will DTRT (the old behaviour) when i
switch to a Metricom radio, but I couldn't tell my brother, for
example, why he should frob the OS to set things up.  And it would get
broken everytime he upgraded.  I'd have trouble getting things to work
well even for him.

On an institutional level, I just don't think it's a supportable
option.  That's not a threat. I'm really sorry if it seemed that way.

I do get people asing me about what software to run on their machines,
and I could not in good conscience tell them to use NetBSD while
in_maxmtu is the default.   I *am* sorry about that, for two reasons: 

  1) Personally, I'd *like* to be able to turn on both in_maxmtu and
     PMTU and  have it Just Work, with no unecessary fragementation,
     everywhere.
  2) I'd like to be able to encourage people to use NetBSD with
     a clear conscience.

Where I'm coming from is that until (1) happens, then having in_maxmtu
on by default means I cant' do (2).  No threats, just facts.

I think you keep saying the problems caused by in_maxmtu are not real,
or are not significant, or are only a problem for obsolete machines,
and that  basically, you don't care about it. 

I keep telling you it is a problem, that people have built networks
for nearly a decade based on the assumptions of those rules.  And the
tcp-impl consensus is that it violates ``be conservative in what you send''.

Jason, which part of ``be conservative in what you send'' is it 
which you ``don't CARE'' about?