Hi Eric,
Just one comment inlined.
Thanks,
Pranjal
From: Eric W Gray [mailto:ewgray2k@netscape.net]
Sent: Thursday, 13 April, 2006
7:33 PM
To: DECRAENE Bruno RD-CORE-ISS
Cc: mpls@ietf.org
Subject: Re: [mpls] One LDP
Implementation specific question ofreceive labelmapping for prefix FECs
Bruno,
No, if anything specific had to be said, it would be that Ru
MUST NOT advertise a more
specific FEC to an upstream peer (Ruu in this case) than it has advertised in
routing.
I cannot speak - obviously - for every implementation out
there, but here's how I believe
they _should_ work when dealing with an ABR (or other sumarization point):
o The ABR-LSR advertises summarized routes into
an adjacent Area.
o The ABR-LSR distributes labels corresponding
to those routes to LDP
peers adjacent to it in that same
Area.
o The ABR-LSR is then the egress for the
corresponding LSPs, in those
cases where it has downstream
routing entries corresponding to more
specific routes.
[Pranjal] This is the very critical point.
In this case Rd has more specific entries, but it has decided that its not the
egress for those
FECs (specific entries), rather the egress
for the summary FECs (Prefixes) that it had advertised to Ru, If that is so
than LDP has to
be more tightly coupled with Routing
(specific IGP and needs to track which Route got summarized across boundaries
or so),right?
o The ABR-LSR MAY - at that point - inject
arriving packets on downstream
LSPs - after it performs a
routing look-up and discovers an FTN (thus
determing an appropriate label and
outgoing interface).
As I mentioned to Pranjal, this is defined this way for
scalability reasons. It is not the
intention of MPLS generally to eliminate the need for routing capability in the
network at
a cost of introducing unscalably large numbers of label distributions
throughout the same
network.
Just as an opinion, a device that is incapable of
making this kind of routing decision,
has no business being in a position to summarize routes.
--
Eric