The MPLS-OPS Archive

[Robert Raszuk <raszuk@cisco.com>]

Just to add to what Alan said the good practice in the following case
would be to configure your P2-PE2 ethernet switch/router to support
giant frames. Most ethernet platform/drivers allow this today. 

Fragmentation is the last resort but is used in a lot of production
environments running MPLS as the easiest patch :). Of course the
recommended place for fragmentation in your case is PE1 or speaking more
generally the edge of the network.

R.

> Alan Hannan wrote:
> 
>   This is a known issue, operationally.
> 
>   You must set the MTU of your Layer 3 transmittal driver to be small
>   enough to allow needed Label Stacking.  I believe we set the MTU of
>   much ethernet to 1490 or so to allow this, which also wreaks
>   cosmetic habit with certain syslog-alarm-happy boxes...
> 
>   In practice, most POS MTUs are 4470, not 1500, FWIW.
> 
>   The router should fragment, and it will, but the switch will drop.
> 
>   -alan
> 
> Thus spake Loureiro, Rodrigo - (Bra) (rodrigo.loureiro@attla.com)
>  on or about Fri, Oct 19, 2001 at 02:32:59PM -0300:
> > Hi all,
> >
> > I am investigating a problem with packet forwarding/switching in a MPLS
> > scope regarding maximum transfer unit, and i would like to share my doubts
> > with you.
> >
> > Imagine a four routers setup: PE1, P1, P2 and PE2. PE1 is connected to P1
> > via POS, P1 is connected to P2 via ATM and P2 is connected to PE2 via a
> > FastEthernet cross-connection. Imagine also that POS interface is configured
> > with a 1500 bytes MTU, ATM with 4470 bytes and FastEthernet with 1500 bytes
> > (consider this hardware is capable to generate Ethernet frames up to 1510
> > bytes of payload for labeling purposes).
> >
> > Now, consider that PE1 and PE2 has a MP-BGP session directly between them.
> > Considering that BGP is capable to generate messages up to 4096 bytes (RFC
> > 1771), we should guarantee that packets greater than 1500 bytes sent from
> > PE1 toward PE2 are being fragmented. Instead of this, the packets will be
> > discarded in the FastEthernet connection between P2 and PE2, once MPLS will
> > not  fragment the packets.
> >
> > I made a test with ICMP from PE1 to PE2, generating packets greater than
> > 1500 bytes, and i could see that packets were not being fragmented by PE1.
> > So, one question arises: What should be the correct behavior for IP packets
> > originated inside the router toward another router in a MPLS environment:
> > should it follow the interface MTU and fragment the packet or should it
> > ignore it and label switch the packet regardless of its lengh ?
> >
> >
> > Regards,
> > --
> >
> > Rodrigo Loureiro
> > AT&T LA - Network Engineer

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