[30146] in North American Network Operators' Group

home help back first fref pref prev next nref lref last post

Re: Path-MTU-discovery

daemon@ATHENA.MIT.EDU (Patrick W. Gilmore)
Mon Jul 17 02:39:25 2000

Message-Id: <4.2.2.20000717023136.0292aea0@ianai.net>
Date: Mon, 17 Jul 2000 02:37:08 -0400
To: nanog@merit.edu
From: "Patrick W. Gilmore" <patrick@ianai.net>
In-Reply-To: <Pine.LNX.4.10.10007170821120.30477-100000@uplift.swm.pp.se
 >
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Errors-To: owner-nanog-outgoing@merit.edu


At 08:26 AM 7/17/00 +0200, Mikael Abrahamsson wrote:

 >On the other hand, at work we're doing some tunneling using ciscos. Due to
 >routing etc the ICMP "need-to-frag"-messages get lost and the people
 >behind those tunnels cannot use 90% of the www sites (so they have to
 >resort to proxies). Seems to me that PMTUd works better than most people
 >think.

Wow, why would the ICMPs get lost?

Also, cisco has a feature on tunnels now where the routers will frag & 
de-frag making the tunnel MTU effectively 1500 bytes (or whatever you set).


 >I do believe that NT and Win2k comes default with a registry setting that
 >makes it send all TCP traffic with the DF flag set (which I can see no
 >reason for unless M$ IP stack cannot do refragmentation properly). This
 >setting is changable as far as I know but I cannot seem to find the
 >information at this time. Anyone?

I have no clue if that is really a setting.  (Do not run any MS web servers.)

However, end stations do not do fragmentation.  They do re-assembly, but 
the receiving station has no control over whether something gets fragmented 
in transit to it.

If the MTU of a path is less than, say, 1500, the end station just sends 
out smaller packets, not fragments.

Since it is likely that you have visited an MS-power site, and you say you 
can reach all sites, then the MS IP-stack can probably send out packets < 
1500 bytes long.


If the setting you describe does exist (and it may very well considering 
MS' history), it is probably just another screw up from the world's 
black-hole for bad programmers.


 >Mikael Abrahamsson    email: swmike@swm.pp.se

TTFN,
patrick



home help back first fref pref prev next nref lref last post