Need clarity on MPLS and the various MTUs

[rhogh]

Hi All,

I'd like some clarification on MTUs and how they are affected by MPLS. This is my current understanding of a standard IP over Ethernet frame;

Ethernet MTU = 1514

IP MTU = 1500

TCP MSS = 1460

1. If this is correct what happens to the above sizes when MPLS adds one label?

2. Next what happens when MPLS adds a second label for VPN/VFR?

3. Do you have to re-configure switches along the path to accomodate larger Ethernet frames?

4. Also how do the above sizes relate to the MPLS MTU and how will it change depending on one or two labels ?

Next what happens when you try to run MPLS over GRE which I understand has the following characteristics;

Ethernet MTU = 1490 (inside the tunnel)

IP MTU = 1476

TCP MSS = 1436

5. Since you cannot adjust the Ethernet MTU of a Tunnel, as it's dependant on it's carrier interface, what adjustments will be needed in the above to make it work with one or two labels.

6. Last question, the ip tcp adjust-mss can be used to reduce fragmentation of tcp packets in the above environments. Will it work if applied to the PE (Ethernet)interface linking to a P. In other words will the MSS adjustment occur before label switching.

7. In the question above will ip tcp adjust-mss work in both directions e.g. ingress & egress of the MPLS?

If anyone can clarify the above related concepts I would greatly appreciaate it.

Thanks,

Roman.

[Harold Ritter]

1- Correct. Assuming you configure "mpls mtu xx", where xx is greater than 1500 (1524 would allow for 6 labels), the ethernet frame would be 4 bytes more.

2- The ethernet frame would be 8 more bytes (4 bytes per label).

3- It depends. Some switches support baby giants (frame slightly larger than 1500 bytes) by default, some others have to be explicitly configured for it and some other don't support it at all.

4- As mentioned above, each mpls label requires 4 additional bytes so "mpls mtu 1524" allows for 6 labels.

5- Why do you want to do MPLS over GRE?

6- If you configure "mpls mtu" and that all switches in your MPLS core support baby giants, you will not have fragmentation. Fragmentation is not good and you probably not want to see in your core.

7- see 6.

Hope this helps,

[Harold Ritter]

1- Correct. Assuming you configure "mpls mtu xx", where xx is greater than 1500 (1524 would allow for 6 labels), the ethernet frame would be 4 bytes more.

2- The ethernet frame would be 8 more bytes (4 bytes per label).

3- It depends. Some switches support baby giants (frame slightly larger than 1500 bytes) by default, some others have to be explicitly configured for it and some other don't support it at all.

4- As mentioned above, each mpls label requires 4 additional bytes so "mpls mtu 1524" allows for 6 labels.

5- Why do you want to do MPLS over GRE?

6- If you configure "mpls mtu" and that all switches in your MPLS core support baby giants, you will not have fragmentation. Fragmentation is not good and you probably not want to see in your core.

7- see 6.

Hope this helps,

[rhogh]

Hi,

Thanks for the info. Just want to clarify that MPLS MTU = IP MTU + LABELS

If ip mtu =1500 and 2 labels are added, mpls mtu = 1508 and therefore total frame size (on ether) is 1508 + 14 = 1522

Is this correct?

Thanks

[Harold Ritter]

That is correct.

[fmaisano]

As for item 5, I have a situation that fits. We are in the process of turning up an MPLS network. The customer's core site is being migrated to MPLS now while the remote office in the WAN will not be able to migrate for some time. Presently they have a GRE tunnel configured from the remote office to the core site. When we migrated the core to the MPLS network, their GRE tunnel broke.

Assuming this is an MTU isue, how is this best resolved?

FM

[attrgautam]

Did you try changing the MTU on the GRE tunnel ?

ip mtu 1500

[irodi]

Hello,

I have a situation like the one described and I use ip mtu 1550 on the GRE interface. We have two 7609 and one of this is a ce with VRF-lite. So on the CE router I have on the tunnel interface not only ip mtu 1500 beu ip vrf xxx forwarding too. On the othe side we have just only a tunnel interface with ip mtu 1500.

when a nig packet arrive on the tunnel int of the second router (non CE), this make two packet smaller and everything goes ok.

When a large packet arrive on the tunnel int of the first router (CE with ip vrf forwarding) this is not able to make smaller packet and drop this one?

What is the problem? Can I configure something special for avoid this behavoir? I don't want to use a smaller mtu because for us is important to haven't icmp unreachable.

Thanks

Ira