The MPLS-OPS Archive

[Robert Raszuk <raszuk@cisco.com>]

I assume you have "tag mtu 1500" configured on your mpls interfaces ?
Now one thing in your msg worries me that you say:

> According to FIB this packet should be send as labeled
&
> and send out as unlabelled packet.

So if the outgoing label is other then pop or untagged and the MRU
("MTU" on the "sh tag for <yr ICMP dst> det") is not 1504 you may indeed
have problems. 

To follow more on this pls send offline "sh tag for xxxx det", "sh cef
xxxx internal" as well as simple topology indicating the ICMP
destination.

R.

> Krzysztof Szarkowicz wrote:
> 
> Robert,
> 
> you are right, I though about the extension to LDP, I was not carefully reading original e-mail.
> 
> But anyway, some problems around MPLS MTU exist. MTU xyz should indicate that the router is able to send labeled packet with total size (label header + IP packet itself) as indicated by MTU printout. Otherwise the packet should be fragmented before sending it. However, on IOS 12.2(4)T1 following happens in my lab environment:
> 
> MPLS MTU is configured to 1500 on all MPLS enabled interfaces. On one of not MPLS enabled interfaces router receives ICMP packet (ping) with size 1500. DF is not set. According to FIB this packet should be send as labeled packet on one of MPLS enabled interfaces. Because is to big (1500 + 4 bytes label) the packet shouls be fragmented. However, for some reason which is unclear to me, the packet is not fragmented and send out as unlabelled packet. I could not figure out why this happens.
> 
> To mention is, if I place previously mentioned non-MPLS interface into VRF, and after that send 1500 bytes long packet over this interface to the router, than it works correctly. I.e. resulting size would be 1508 (two labels now: LDP + VPN), which is to big, and the packet is fragmented.
> 
> I still don't know it is an undocumented feature or bug.
> 
> Best regards,
> Krzysztof Szarkowicz
> 
> ----- Original Message -----
> From: "Robert Raszuk" <raszuk@cisco.com>
> To: "Krzysztof Szarkowicz" <kszarkowicz@freemail.hu>
> Cc: <mpls-ops@mplsrc.com>; "Donn Lasher" <dlasher@clearskynet.net>
> Sent: Friday, February 22, 2002 5:38 PM
> Subject: Re: MPLS & Path MTU problems
> 
> >
> > Krzysiek,
> >
> > No. The MTU in the output of "show mpls forwarding-table X.X.X.X.
> > detail" has nothing to do with MTU :). The term "MTU" there is a
> > misnomer, it should be actually a maximum receive unit.
> >
> > If the "MTU=" there is xyz then it implies that a labeled packet of size
> > xyz bytes can be *received* by the router and sent out on interface
> > indiacted without needing to fragment it. the MTU printed out depends on
> > each prefix and could have different values depending on how many labels
> > are pushed/popped.
> >
> > It also has nothing to do with LDP. LDP at least for now does not
> > advertise MTU :). I am assuming you are talking about extensions
> > proposed in: draft-black-ldp-mtu-extensions-00.txt right ?
> >
> > R.
> >
> >

-------
The MPLS-OPS Mailing List
Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
Archive: http://www.mplsrc.com/mpls-ops_archive.shtml