mpls MTU

Unanswered Question
May 17th, 2007

Hi folks,

I've to use a 3745 as PE on my MPLS infrastructure. But change the fastethernet interface MTU is not permitted. Well, the command 'mpls mtu', or 'tag-switching mtu', is enough, or is not possible to use a 3745 as PE?

thanks for your support



  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (4 ratings)
ariela Thu, 05/17/2007 - 12:07

mmm, I don't understand what the command does, in effect ...

Is it possible to have an interface mtu 1500 and a "mpls mtu" 1508?

Thanks for your support


Harold Ritter Thu, 05/17/2007 - 12:25


It is indeed possible to have an MPLS mtu greater than the interface MTU. It is the case most of the time for MPLS based application.

Hope this helps,

rajju Thu, 05/17/2007 - 13:41

Interesting !!

Harold: would you be able to elaborate on this a bit further for better understanding?


Harold Ritter Thu, 05/17/2007 - 14:41


Sorry for the brief explanation. Let me try to a bit clearer. Traffic comes in the edge LSR and could be up to 1500 bytes. For L3VPN, you generally need to add 2 labels (IGP label and VPN label), therefore increasing the original frame size by 8 bytes (in most cases).

For L2VPN it gets worst, as the original L2 frame is encapsulated into yet another L2 frame and then extra overhead, such as control word and labels are added.

It is therefore common for the MPLS MTU to require to be larger than the interface MTU.

That being said, in some new IOS releases, such as latest 12.0S and 12.2S, it is not allowed anymore to set the MPLS MTU more than 24 bytes beyond the interface MTU. Although 24 bytes is more than enough for L3VPN, it is not likely sufficient for L2VPN. The way to address this issue is to set the interface MTU higher. Most interfaces on core equipment will allow you to increase the interface MTU without an issue.

Let me know if I have addressed your question,

ariela Thu, 05/17/2007 - 15:00


mpls mtu is independent from int mtu, correct? in some cases I could have a PE with ospf in global, int mtu 1500 and mpls mtu 1508 so labels work, but ospf neighborship too without mtu-ignore ... is it?

The problem, in effect, is that, expecially on ME switches, there are a lot of command about mtu (system mtu, routing mtu, etc...).

Thanks for your support


Harold Ritter Thu, 05/17/2007 - 16:50

That is correct. Changing the MPLS MTU doesn't affect the ip MTU. On the other hand changing the interface MTU has a direct effect on all other MTU (MPLS, CLNS, IP).

Hope this helps,

ariela Thu, 05/17/2007 - 22:22

Thank you very much, Harold, for your explanation.

Last question: I need an IOS with Jumbo Frame support to use correctly the 'mpls mtu' command? What that feature does?

Thanks for your support


Harold Ritter Fri, 05/18/2007 - 03:24


Jumbo Frame support would be required on a L2 (or L2/L3) device. The notion of jumbo frame support on a device such as a 3745 is only at either the interface or MPLS MTU level.

If your 3745 Ethernet interface connects to a L2 device, make sure this device supports jumbo frames or at least baby giants (slighltly larger than 1500 bytes).

Hope this helps,

mazlan.alatif Fri, 05/18/2007 - 18:33

Hi Harold,

Most of new IOS now support mpls mtu overide, and it can configre +80 byes of its interface mtu. FOr exmp ethernet int can be configured with mpls mtu 1580. Can we use this features to support l2vpn?

Do you have the exact mtu size for l2vpn ie eompls, vpls and etc. I have tried l2vpn in my lab and using mpls mtu 1524 and this not working. Its ok for l3vpn since its only stack of 2 label (igp and vpn).

Can you explain.


Harold Ritter Sat, 05/19/2007 - 17:06


As far as I know, this command is only available in a few IOS trains (12.2SBC, 12.2SB and 12.2SRA/SRB). Note that with this command, there is a chance that packets will not be handle properly depending on the HW/drivers used. See following document for more information.

As explained in this document, the best way to cope with the oversized packets in an MPLS network is to increased the interface MTU on all core interfaces whenever it is possible to do so.

As for your second question, here's a breakdown of the l2vpn overhead.

2 labels = 8 bytes.

1 control word = 4 bytes.

original ethernet frame including dot1q = 18 bytes (14 without dot1q)

total = 1530 bytes (1526 without dot1q)

Hope this helps,


This Discussion