The MPLS WG Archive

[Bora Akyol <akyol@pluris.com>]

Moreover, is this material discussed in RSVP-TE, if not why not?
A discussion of hierarchical tunnels in that draft or an accompanying draft would
have been nice.

Bora


Curtis Villamizar wrote:

> In message <14657.53536.422247.469955@redd235.procket.com>, Tony Li writes:
> >
> >
> >  | How does one establish hierarchical tunnels using RSVP-TE?
> >  |
> >  | I know that the label object can be deeper than one label, but you need mo
> > re
> >  | than this to establish hierarchical tunnels.
> >
> >
> > A hierarchical tunnel requires the participation of at least two LSRs.  The
> > operation of the LSR creating the innermost tunnel is straightforward and
> > is unchanged from what has been endlessly discussed here.
> >
> > For an LSR to create an outer tunnel, it simply creates another tunnel and
> > specifies MPLS as the L3PID.  It then forwards packets received on the
> > inner tunnel by pushing a label for the outer tunnel.  The RSVP control
> > traffic for the inner tunnel is also forwarded through the outer tunnel
> > after first being encapsulated with a null IP label.
> >
> > Recurse, ad nauseum.
> >
> > Simple, really.  ;-)
> >
> > Tony
>
> Tony,
>
> Two minor points.
>
> If the outer tunnel is constrained (by configuration) to accept only
> inner tunnels with one L3PID, then that L3PID can be placed on the
> outer tunnel.  If there are optimization possible for a particular
> L3PID (IP src/dst hash to split loading along the way comes to mind
> for IPv4) then putting the inner L3PID on the outer tunnel has
> advantages.
>
> No where have I seen a requirement to put a null label on traffic
> destined to the egress LSR.  I had been assuming that if the IP packet
> under the top label has the address of the router ID of the egress
> under the outer label, then the packet would be delivered to the
> egress.  If the egress required PHP, then the penultimate LSR would
> put a null label on the control traffic for the the inner label and
> the egress would POP the null label and see its own IP address.
> Otherwise, the egress would see two null labels.
>
> If I'm missing something and/or there will be any interoperability
> problem with the way I intended to encapsulate control traffic inside
> the outer label, please tell us about it.
>
> Curtis