MPLS AToM Fragmentation

[m.kenfacktabakem]

Guys,

I have read a few docs on it but seem more confused than I was at the beginning. This site (http://e-articles.info/e/a/title/Multiprotocol-Label-Switching-Virtual-Private-Network/) for instance says and I quote "If the payload is IP traffic, the ingress PE router can fragment the IP packet before it enters the pseudowire. In that case, the destination host reassembles the packet. If the frame payload is not IP, the ingress PE router can perform the fragmentation on the frame before it enters the pseudowire and the egress PE router reassembles the frame".

Given the network below built on GNS3:

  7200                           7200          mpls         7200           mpls         7200                                     7200

CE1-----------------------------PE1----------------------------P--------------------------------PE2-------------------------------------CE2

      mtu=4470   mtu=1500      mtu=4470                   mtu=4470                mtu=1500             mtu=4470

port based mpls xconnect between PE1 and PE2

Ping between CE1 and CE2 with mtu=2000 does not work. This tells me fragmentation is not hapenning at teh ingress PE as stated by the quote above. This was also observed on our production network (we use 6500/7600) when a customer could not pass traffic with mtu=1508 (interface mtu was 1500). Also mpls  mtu command made no difference. I am trying to find an authoritative document that would shed some light into AToM fragmentation as it happens on Cisco routers. Any pointers is appreaciated.

For now I am taking the view that it is not possible for  6500/7600 platform to fragment any MPLS frame. Also customer traffic at the ingress PE will be dropped unless their size is <= PE ingress interface mtu.

Am I right in my assessment?

Thanks,

Mathias

[Mohamed Sobair]

Hi,

Your assesment is correct.

Cisco recommends to set the Core Facing interfaces MTU value to a higher value than the Edge Interfaces to accomodate extra labels. as well as , the customer facing interface CE MTU value MUST match the PE interface MTU value, if its not the case, packets might not be forwarded, or Dropped.

The MPLS MTU should match the MTU value of the interfaces on most IOS releases to take effect, if you set it higher than the interface MTU, it would also lead to packet drops. Some IOS's doesnt allow the MPLS MTU to be higher than the interface MTU.

Please refer to the bellow document for more information,

http://www.cisco.com/en/US/docs/ios/12_2sb/feature/guide/newmtu.html

Regards,

Mohamed

[Mohamed Sobair]

Hi,

Your assesment is correct.

Cisco recommends to set the Core Facing interfaces MTU value to a higher value than the Edge Interfaces to accomodate extra labels. as well as , the customer facing interface CE MTU value MUST match the PE interface MTU value, if its not the case, packets might not be forwarded, or Dropped.

The MPLS MTU should match the MTU value of the interfaces on most IOS releases to take effect, if you set it higher than the interface MTU, it would also lead to packet drops. Some IOS's doesnt allow the MPLS MTU to be higher than the interface MTU.

Please refer to the bellow document for more information,

http://www.cisco.com/en/US/docs/ios/12_2sb/feature/guide/newmtu.html

Regards,

Mohamed

[Mohamed Sobair]

Hi,

Your assesment is correct.

Cisco recommends settig the Core facing interfaces to an MTU higher than the Edge to accomodate extra labels. besides the CE interface and the PE facing CE interfaces to be lower MTU value than the core and MUST match.

If the above doesnt take place, packet might not be forwarded and could be dropped , the MPLS MTU command must also be set equal to the interface MTU, if its also not the case , packet might be dropped. On some recent IOS releases , the MPLS MTU must be set equal or lesser to the interface MTU.

please refer to the below document;

http://www.cisco.com/en/US/docs/ios/12_2sb/feature/guide/newmtu.html

Regards,

Mohamed

[Laurent Aubert]

Hi Mathias,

EoMPLS fragmentation is not supported so what you see is expected. You must be sure your core MTU are high enough to transport maximum customer MTU plus the label stack.

HTH

Laurent.

[m.kenfacktabakem]

Many thanks guys for your answers - really appreciated.

Was on holiday so my late reply.

Mathias,