Re: What's your favorite MTU?

mogul (Jeffrey Mogul) Fri, 13 April 1990 18:26 UTC

Received: by acetes.pa.dec.com (5.54.5/4.7.34) id AA07289; Fri, 13 Apr 90 11:26:48 PDT
From: mogul (Jeffrey Mogul)
Message-Id: <9004131826.AA07289@acetes.pa.dec.com>
Date: 13 Apr 1990 1026-PST (Friday)
To: art@opal.acc.com (Art Berggreen)
Cc: mtudwg
Subject: Re: What's your favorite MTU?
In-Reply-To: art@opal.acc.com (Art Berggreen) / Tue, 10 Apr 90 09:58:06 PDT. <9004101658.AA27286@opal.acc.com>

    I'm a bit concerned about building into a protocol the MTUs that happen
    to be used on today's network technologies.

It is VERY important to understand that I am not building this list into
the protocol specification.  I am perhaps building it into the
"implementation suggestion" section of the spec, although I expect that
as a guideline rather than a spec, it will be plastic enough to be
modifiable as times change.

Note that this part of the implementation is meant to support the
use of routers that haven't been upgraded to meet the new requirements
outlined in the spec.  I assume that router implementations made to
handle data-link layers defined in the future will also be upgraded
to meet this spec, so the issue of my little list will be moot in those
cases.

Remember also that the alternative to having a list of likely MTUs
is to have some algorithm (such as "reduce your segment size by 25%")
which, although much simpler and more abstract, has nothing at all
to do with reality.
    
    >PMTU-step	MTU	Protocol   Comments		Specification
    >		 2002	IP-IEEE    IEEE 802.5			RFC 1042
    
    Don't 16Mbit TRs support much larger MTUs?

RFC1042 mentions that 802.5 systems can potentially support these MTUs:
	8188, 4092, 2044, 1020, 508
It isn't clear from the RFC just which of these actually get used.
I have plateaus at or just below all of these values, except for
4092.  Putting one at 4092 would probably cause some problems as
FDDI becomes ubiquitous; either there would be an extra RTT spent
getting from an FDDI MTU to an Ethernet MTU, or I would have to
remove the FDDI plateau (now changed to 4352, in the lastest
draft RFC) and potentially waste 6% of the FDDI MTU.  If you
expect that 802.5 networks with MTU=4092 are going to be common,
then I will have to take that into account.
    
    >		  576	IP-X25     X.25 Networks		RFC 877
    
    There is no inherent MTU limit in X25 (using M-bit sequences).  In fact
    if large X.25 packet sizes can be negotiated, there are real advantages
    in using a larger MTU.
    
That suggests that there should be a few more "arbitrary" plateaus
at higher MTUs (such as 16k, 32k, and perhaps a few in between).  It
would help to know what values are likely; i.e., if I put a plateau
at 16384, but due to header length considerations the actually MTU
used is 16382, then the algorithm will drop the segment size to
around 8K and waste some potential bandwidth.

-Jeff