[c-nsp] C76 and ES+ LC - show mpls l2transport bug

Holger L cisco at entrap.de
Fri Jul 26 08:22:03 EDT 2013


Hi all,

This may be interesting for everybody using MPLS-TE and Loadbalancing with
ES+ LC.

We use Cisco 7600 with IOS 15.3.1-S2 and the following LCs:
Module   Part number               Series      CEF mode
1        RSP720-3C-10GE        supervisor           CEF
2        76-ES+T-4TG              dCEF720          dCEF
3        76-ES+T-4TG              dCEF720          dCEF
4        76-ES+T-40G              dCEF720          dCEF

We set up a MPLS core and because of multiple links between core routers,
we have multiple MPLS-TE tunnels and load balancing between each two of
them.
E.g. Tunnel10301, Outgoing Label 39 and Tunnel10302, Outgoing Label 53

Some days ago I discovered that the imposed label stack on a VC does not
show the truth as I compared "sh mpls l2transport vc 2105 detail" which
said "Output interface: Tu10301, imposed label stack {39 0 18}" to my
packet capture which showed label stack {53 0 18}. Thus the EoMPLS gets
transferred on a not supposed tunnel which results in traffic on a
different interface.

I opened a TAC case and they raised a bug for this, CSCui14755.
Now I got the following response:

> Here is the situation. Whenever dual path exists to carry the EoMPLS VC,
> when it is established, the software will take one of the CEF path to
> build the label stack that is displayed under 'show mpls l2 vc ...
> detail' but it is possible the hardware will forward the traffic on a
> different path based on the load-balancing hashing algorithm.
> The problem here is that there is no way with the current code and
> hardware for the software to retrieve the hardware information which
> could be used to forward the traffic. The developers looked to the code
> architecture and it is not possible to modify it.
> I even asked if this could be implemented as a new feature (then having
> the whole code rewritten) but that still does not seem to be an option.

First, I would like to warn everyone of you trusting the "imposed label
stack" together with load balancing.
Second, I would to know how many of you are affected by this bug which
will not be fixed..

Holger



More information about the cisco-nsp mailing list