[76803] in North American Network Operators' Group
Re: Smallest Transit MTU
daemon@ATHENA.MIT.EDU (Dan Hollis)
Wed Dec 29 18:44:20 2004
Date: Wed, 29 Dec 2004 15:39:28 -0800 (PST)
From: Dan Hollis <goemon@anime.net>
To: Florian Weimer <fw@deneb.enyo.de>
Cc: Jerry Pasker <info@n-connect.net>, <nanog@merit.edu>
In-Reply-To: <873bxopxa3.fsf@deneb.enyo.de>
Errors-To: owner-nanog-outgoing@merit.edu
On Wed, 29 Dec 2004, Florian Weimer wrote:
> * Dan Hollis:
> > On Wed, 29 Dec 2004, Jerry Pasker wrote:
> >> Is there an RFC that clearly states: "The internet needs to transit
> >> 1500 byte packets without fragmentation."??
> > Actually the bigger problem imo is the number of sites which block ECN
> > http://urchin.earth.li/ecn/
> Why is this a problem? ECN has to be deployed on routers, and it
> currently isn't.
Because tcp connection endpoints have to implement ECN in order to manage
the flow.
Many OSes (Linux/bsd/aix/solaris/etc) support ECN but due to the large
number of braindamaged firewalls out there (http://urchin.earth.li/ecn/),
it defaults to off.
Any host which tries to negotiate ECN in a tcp connection will run into
lots of problems as millions of idiotic firewalls drop the packets on the
floor. Quite often the same firewalls which drop 69/8 on the floor.
Its sad because ECN is quite useful. Though the damage by clueless
network admins blocking 69/8 is worse.
> Cisco seems to offer it on some platforms, but their implementation
> provides a strong incentive to constantly set the ECN flags in a
> certain way, to push the packets into a different QoS class.
> (This is from memory, and it might have been corrected.)
http://www.icir.org/floyd/ecn.html
-Dan
