Re: [OSPF] draft-ietf-ccamp-automesh-01.txt

["Igor Bryskin" <ibryskin@movaz.com>]

Kireeti,

Good to hear from you.

Please, see in-line.

Igor

----- Original Message ----- 
From: "Kireeti Kompella" <kireeti@juniper.net>
To: "Adrian Farrel" <adrian@olddog.co.uk>
Cc: <ospf@ietf.org>; <ccamp@ops.ietf.org>
Sent: Monday, May 15, 2006 10:12 AM
Subject: Re: [OSPF] draft-ietf-ccamp-automesh-01.txt


> Hi Adrian,
>
> On Sat, 13 May 2006, Adrian Farrel wrote:
>
> > There is a draft in CCAMP that I want to bounce off the OSPF working
group.
> ...
> > My questions to you:
> > 1. Is it a concern that P-routers are being used to store and forward
> >  opaque information only needed by a small subset of the routers
> >  in the network?
>
> Necessary evil (i.e., No; see below)
>
> (I would reconsider the use of the word "small" -- in most networks I
> have seen, the number of PE routers vastly outnumber the number of P
> routers.)

IB>> I think it is even more true in the context of L1 networks which in
most of the cases are bunch of interconnected rings of PEs with zero or very
few Ps.

>
> > 2. Is there a scaling concern that there is no control on the number of
> >  mesh groups that may exist, nor the number of mesh groups to
> >  which any router can belong?
>
> I guess an implementation could go berserk and advertise 65536 bytes
> worth of mesh groups, but no, this doesn't concern me too much.
>
> > Context:
> > This question arises in the context of
> > draft-bryskin-l1vpn-ospf-auto-discovery-01.txt that is being discussed
in the
> > L1VPN working group. This I-D proposes to use the IGPs (specifically
OSPF) to
> > distribute information about which VPNs can be accessed through the PEs
(not
> > general VPN membership or reachability information, but just a list of
VPN
> > IDs and the link I-Ds that are used to reach them). Loud voices have
been
> > raised in L1VPN about the scalability and appropriateness of such an
idea,
> > and since it seems to be very similar to automesh, I want to see whether
you
> > all think there is a problem with automesh.
>
> I have the same issues with using ISIS/OSPF for auto-mesh as I do for
> autodiscovery in L1VPNs -- OSPF and ISIS are not ideal vehicles for
> such information.  However, there are two very important differences
> in these two cases:
>
> 1) BGP is often not present on "interior" routers (consider the case
>     of "P" routers fully meshed with TE LSPs, and PEs running LDP; and
>     BGP running only on PEs -- "BGP-free core")
>
> 2) It is vital for VPNs that a good policy mechanism be available to
>     control the distribution of information -- otherwise, there could
>     be serious breaches of privacy.

IB>> Note, that what we are suggesting to advertise in L1VPN opaque LSAs are
locally configured CE-PE links and their association with VPNs. I am sure
that I am missing something, so could you help me understand:
a) what scenarios of "serious breaches of privacy" you have in mind?
b) how the BGP approach can help in this regard?

Thanks,
Igor

>
> That said, I would like to see automesh information carried in BGP,
> to be used in preference to ISIS/OSPF whenever possible.
>
> Kireeti.
> -------
>