Re: [mpls-linux-general] Label stacking example

[James R. L. <jl...@mi...>]

On Wed, Oct 20, 2004 at 03:34:57PM +0200, Birger T?dtmann wrote:
> 
> Hello James,
> 
> Am Mittwoch, den 20.10.2004, 08:26 -0500 schrieb James R. Leu:
> > I haven't implemented all of the command line args in the new "mpls" program
> > yet.  So for now you will have to use mplsadm2 to do label stacking:
> [...]
> 
> Many thanks for the elaborate example.
> 
> > 
> > Any questions? ;-)
> 
> Sure ;-)
> 
> -> Why do I need IP *at all* on the intermediate nodes?  
>    ("maplsadm2 .... :eth1:A.B.C.D")

Your using ethernet interfaces right?  Whenever you send a packet
over ethernet you need to know the MAC address of the peer on that ethernet
segment.  So that means the nexthop definition in the NHLFE needs to
somehow tell MPLS what the MAC address of the nexthop LSR/LER is.

Would you rather type an IP address or a MAC address? :-)  The IP address
is used to ARP for the MAC of the nexthop LSR/LER.

(BTW the nexthop definition in mplsadm2 is the 'set:eth0:ipv4:1.1.1.1'
instruction)

If you were using interfaces which do not need ARP (ie PPP interfaces) then
your nexthop defintion would be 'set:ppp0'.

> This bothered my as well for the simple example (without label
> stacking).  When thinking of smart LERs and dumb LSRs, IP-based FECs
> should be needed at the borders only (e.g. R1 will have an entry towards
> the IP adress of R4, the right border node, not towards R2).  I know
> many MPLS networks use IP in the core to find best paths, i.e. using
> OSPF-TE and the like, but can we as well build MPLS-only, non-IP cores
> with mpls-linux? 

The nexthop definition is not a FEC.  A FEC refers to the ultimate destination,
not an intermediate hop along the way.

If you wanted to type MAC addresses everywhere, then yes you could have
a MPLS core without using IP addresses.  mpls-linux would need some minor
modifications for that to work.

> 
> 
> Regards,
> -- 
> Birger T?dtmann
>  Technik der Rechnernetze, Institut f?r Experimentelle Mathematik und Institut 
>  f?r Informatik und Wirtschaftsinformatik, Universit?t Duisburg-Essen, Campus Essen
>  email:bto...@ex... skype:birger.toedtmann pgp:0x6FB166C9

-- 
James R. Leu
jl...@mi...