[c-nsp] MPLS TE to use 2 default routes?

Pshem Kowalczyk pshem.k at gmail.com
Fri Oct 21 19:31:55 EDT 2011


Hi,

On 22 October 2011 07:35, Phil Mayers <p.mayers at imperial.ac.uk> wrote:
> On 10/21/2011 11:26 AM, Mark Tinka wrote:
>
>> We do this within our core network to support load balancing
>> for non-equal-cost distances (towards peers or customers),
>> even though bandwidth within the core is the same; just like
>> your case.
>
> That's promising. Having tried it on a test router, it seems a config like:
>
> int Tun1xx
>  ip unnumbered Loopback1
>  tunnel mode mpls traffic-eng
>  tunnel destination Rxx
>  tunnel mpls traffic-eng autoroute announce
>  tunnel mpls traffic-eng autoroute metric absolute 1
>  ! Just for example, obviously
>  tunnel mpls traffic-eng path-option 1 dynamic
>
> ...in particular, the "autoroute metric absolute" is needed to fix the
> metrics so that the tunnels are equal-cost, yes?

Yes, otherwise the tunnel will assume the IGP cost to the destination.

> Do I want "autoroute announce"? Since the routes I care about are BGP, the
> only thing I need to tunnels for is to force the IGP cost to the iBGP
> loopbacks to equal.

Yes you want that - the next hop for BGP prefix is taken from IGP, so
IGP has to know how to get there. Without autoroute annouce you'll
have to manually send traffic down the tunnel.

> In terms of my original ascii diagram, will the presence of these tunnels on
> R4 induce R5 to send traffic upwards to R4 (and via the tunnels) when it
> might previously have sent it directly? Or are the tunnels "local" to each
> router and not advertised into IGP?

Short answer - no it won't, long one - forwarding adjacency will
advertise tunnel into IGP, so it's visible from other routers (I
generally find that feature more dangerous then useful), autoroute
announce is only local to the router in which it's configured.

kind regards
Pshem



More information about the cisco-nsp mailing list