[email protected]
[Top] [All Lists]

Re: the shen-mpls-nnhop Was:(Re: thoughts on draft-bryant-shand-ipfrr-no

Subject: Re: the shen-mpls-nnhop Was:Re: thoughts on draft-bryant-shand-ipfrr-notvia-addresses-00.txt
From: Alia Atlas
Date: Wed, 18 May 2005 10:05:23 -0400

In this case, I agree with Stewart. (Much less usual than Stewart agreeing with Mike :-)
To my mind, the IP FRR work that has been going on in rtgwg also applies to
LDP. LDP, after all, learns its next-hops from the IGP. I have been
referring to this work as IP/LDP FRR since the beginning and certainly
advocating for its applicability to LDP as well.
There are two points that I can see where there could be additional work
for the mpls wg. First, LDP FRR requires knowledge of labels from
neighbors who are not the primary neighbors. This means that the
loop-checking which can be enabled in LDP should not be (at least for the
non-primary neighbors)- because the label bindings do not all represent
paths that would be taken. This doesn't require changes to LDP; it is a
detail of the required mode. There could be more text in the LFA draft to
reflect the acceptable LDP modes/configurations that would work.
Second, an advanced method may require extensions to support label
discovery. For U-turn alternates, the requirement is merely that a router
advertise its label bindings to all neighbors - not merely those that are
primary. This is already clearly within the LDP spec - and a good idea for
convergence reasons. Thus, there's no work required for that aspect. For
U-turn alternates, for the explicitly marked U-turn packet, I do use a
specific label that all implementations need to use in common; ideally
that'd be a reserved label (13) but I haven't tried to start that
discussion, until the future of U-turn alternates becomes more clearly
defined. For a tunnel approach that has the tunnel egressing at the
next-next-hop, Naiming's idea of learning label bindings via a neighbor
might be helpful. For the tunnel approaches, there's also the
potential/probable need for targeted sessions to support multi-homed
prefixes, but that requires no extensions. For a tunnel approach, it is
highly likely that one tunnel technology used would be LDP, since that uses
a hardware mechanism clearly supported today.
My expectation was that the rtgwg would need to last-call the IP/LDP FRR
work in the MPLS WG, as well as the ISIS and OSPF WGs, because of the
applicability there.
What do you think?


At 09:26 AM 5/18/2005, Stewart Bryant wrote:

Loa Andersson wrote:
Maiming and Stewart,
you wrote - and I guess this is valid:
Naiming Shen wrote:

I think the needs start emerging, any important services
riding on top of IP/MPLS transport infrastructure needs
fast convergence services. It's not reasonable to assume
only RSVP-TE LSPs need fast reroute, and other
network transport does not. This nnhop-ldp draft is to
facilitate the LDP based MPLS network for FRR with
node protection. I have been talking to some providers
in the past year, there are certainly interests in
this service.
Stewart Bryant wrote:
 > I think the interest that we are seeing in IPFRR for LDP
 > networks demonstrates the usefulness of this work. Sure
 > RSVP-TE FRR exists, but some customers have expressed
 > an interest in a solution that does not use RSVP.
However the MPLS working group need to take a decision if,
when and how this futopicctionality should be developed.

The way of doing this would be
- write down the requirements
- send them to the mpls wg
- we go through the moces to establish the mpls wg consensus
- I don't think FRR for LDP based MPLS enabled IP networks
  should not go into the rtgwg, the charter of the rtgwg seems
  to say the rtgwg takes of everything that does not fit into
  other routing are working groups (Alex and Bill comments on this?)
I disagree.

Whilst the MPLS WG must be engaged in this work, to do it
exclusively in MPLS WG would be a mistake.

The work that we have done to date has achieved a high degree
of commonality between the IP solution and the MPLS solution.
Maintenance of that commonality would be at risk if the
MPLS WG struck out in their own direction. This work
requires a significant knowledge of the working of routing
protocols and therefore RTGWG (or perhaps a new WG in routing)
would seem a more natural home than MPLSWG.

- my take is that for this purpose you could either document
  the requirements in a separate document or put them into one
  the existing documents. I've no preferences, but this should
  be rather short.
I think that we need an FRR requirements draft.

BTW the scope is further widened by the applicability of elements
of the solution (loop free convergence) to network management.

- my preference would be to speed this up so we could have
  a decision going into the Paris meeting.
Also I think it will be right to take this discussion to the
mpls wg list. I will write a mail to mpls list to that effect.
Certainly the MPLS WG need to be an aware and active participant
in this work, but please do not precipitate a schism.

- Stewart


Rtgwg mailing list
[email protected]

Rtgwg mailing list
[email protected]

<Prev in Thread] Current Thread [Next in Thread>