"unsupported" features in IOS ("mpls mtu") (was: Re: [c-nsp] High Availability PoP Design)

[Rodney Dunn]

Setting the mtu for MPLS is supposed to just reflect
the additional encap bytes needed for the MPLS labels.

The underlying packet size does not change.

ie: We only support 1500 byte packets with 6 labels
+ ether headers (plus dot1q).

Setting it to 1516 would be valid and it's supposed to
mean: allow a packet through without fragmentation that
has less than 4 labels on it (4 labels at 4 bytes each
is 16 bytes).

What confused people was they tried to set the mtu to something
higher and then add gre headers in it not wanting the ip 
to be fragmented with say one label. It didn't work under all
scenarios and it never should have. That's where folks were
confused.

Rodney

On Fri, Nov 04, 2005 at 04:01:19PM -0200, Everton da Silva Marques wrote:
> Rodney,
> 
> On Fri, Nov 04, 2005 at 12:31:21PM -0500, Rodney Dunn wrote:
> > 
> > We support 1500 byte ip headers on those PA's +
> > 6 labels (24) bytes.
> > 
> > That is minus the ethernet/dot1 headers.
> > 
> > The problem was we originally didn't do per bounds checking
> > on each filed (it's too performance impacting).
> > 
> > It's an extremely complicated set of things on how and
> > when fragmentation is done.
> > 
> > People thought the tag mtu command was doing something it wasn't
> > in the first place so that's why we changed it.
> 
> We run "mpls mtu 1516" on some 7500 routers
> with IOS 12.0S, like this:
> 
>   7513-1#sh run int FastEthernet5/0/0 | i tag
>    tag-switching mtu 1516
>    tag-switching ip
>   7513-1#
> 
> If memory serves me correctly, we were advised by
> TAC to use the command to circumvent fragmentation
> difficulties.
> 
> Is that command being removed from 12.0S as well?
> If so, what is going to be the new proper setting?
> Just to remove the "mpls mtu" command from the
> interface?
> 
> Thanks,
> Everton