The MPLS WG Archive

[David Charlap <David.Charlap@marconi.com>]

Manoj Agiwal wrote:
> 
> draft-ietf-mpls-rsvp-lsp-tunnel-08.txt talks about sharing the common
> label across LSPs which are shared explicit.
> 
> What are the advantages one get on assigning a common label ?
> Are there any other advantages apart from conserving the label
> space.

Conserving label space is important.  On some hardware platforms, label
space may be a scarce resource.

> Why we not share common labels across the Explicitlt Routed LSPs?

I'm not sure what you're asking here.

> Why there is a requirement to make sure if Path messages have
> identical Explicit Route Objects ?

Because, once you merge labels, data-plane packets for the two LSPs will
have to follow the same path to the end of the LSP.  If the LSPs are
supposed to diverge later on, the shared label will prevent this.

Note that since ERO objects are shortened as they propagate across the
network, each LSR is really only comparing the portion of the ERO from
itself to the end of the LSP.  Upstream nodes are not represented in an
ERO and therefore don't come into consideration.

> I am including the lines from the draft which talks about assigning
> common label to LSPs .
> 
> "If a node receives a Resv message that has assigned the same label
> value to multiple senders, then that node MAY also assign a single
> value to those same senders or to any subset of those senders.  Note
> that if a node intends to police individual senders to a session, it
> MUST assign unique labels to those senders."

No issue here.  Since QoS is on a per-label basis, if two LSPs are
supposed to have different QoS levels, they must also have different
lables.  (DiffServ extensions not withstanding, of course.)

> "SE style reservations can be provided using multipoint-to-point
> label-switched-path or LSP per sender.  Multipoint-to-point LSPs may
> be used when path messages do not carry the EXPLICIT_ROUTE object, or
> when Path messages have identical EXPLICIT_ROUTE objects.  In either
> of these cases a common label may be assigned."

Again, no question here.

The big deal is that if two LSPs are to share a common label, then they
must follow identical routes to the egress LER.

If there is no ERO, then the routes to the LSP egress will be identical,
since both will be determined by the routing table (looking up the route
to the session address.)

If there is an ERO, then the ERO determines the route to the egress.  If
the EROs differ, then the LSPs will diverge somewhere downstream. 
Therefore, you should not assign a common label - if you do, then the
data packets won't be able to diverge when the LSPs do.

-- David