[c-nsp] MTU on MPLS TE Tunnels w/PA-FE-TX?

[Vinny Abello]

Rodney Dunn wrote:
> On Tue, Mar 20, 2007 at 11:21:02AM -0400, Vinny Abello wrote:
>> Rodney Dunn wrote:
>>> On Sun, Mar 18, 2007 at 11:08:49PM -0400, Vinny Abello wrote:
>>>> Nope, I think what I have wrong is is I'm not running an S train or that Cisco has not put those fixes in any mainline code yet. I checked a few routers that do have SB trains and it seems to work properly although I didn't do a complete test on them yet. I need to get some time to test more in a lab scenario.
>>>>
>>> I've never seen a report of MPLS mtu set to 1508 to get 2 bytes of
>>> labels (1 IGP + 1 VPN) not working on those PA's. You may be having a TE
>>> specific problem.
>> Possibly... Like I said, I'm new to it, but I'm just following guides in several Cisco Press books on MPLS. Anyone feel free to make any suggestions to something obvious I'm missing. Just as an FYI, I'm currently using TDP instead of LDP due to 12.2 apparently not supporting LDP. I can switch that if this has any bearing on this problem. I just thought I'd mention it. I just want to rap my head around traffic-engineering and that's the code I'm working with currently.
>>
> 
> I'm not sure exactly how that works given the tunnel could route over
> any interface with various mtu sizes (say you reopt to a serial path that
> can handle say a 4470 mtu).

Yes. I was under the impression from what I read that you need to make sure there is enough headroom in all your paths for the bytes needed for the tags, and that it was usually necessary to use the mpls mtu or tag-switching mtu command on FastEthernet interfaces to achieve this. I don't know how the Tunnel chooses the MTU size when, like you said, it can change paths with different MTU sizes.

> This sounds more like an ip2tag mtu check.
> 
> In the ip2tag path I think we are supposed to fragment before mpls
> encapsulation.

Does that cause any performance issues due to the fragmentation and reassembly?

> This sounds like a bug if it isn't working.
> 
> Can you try 12.4 mainline latest and see if it works?

I'll have to take a look and try it. I don't think I can do so currently on the router this is setup on because it's being used for other things. I'll have another one shortly that I could try this on.

It seems the easy answer though is bumping the physical MTU size on the FastEthernet interface which appears possible in 12.2(28)SB and 12.2(31)SB as you previously stated.


Thanks for your time. :) I'll see what I can find in 12.4. I'll probably end up using 12.2(28)SB or 12.2(31)SB if needed.