[c-nsp] MPLS down to the CPE

[Adam Vitkovsky]

> We distribute the larger aggregation nodes to various sites and then
generally have access rings with access nodes hung off of those.  
Are the access rings in a separate area/level or running a separate igp, or
how do you scale your backbone IGP please?

adam
-----Original Message-----
From: Phil Bedard [mailto:philxor at gmail.com] 
Sent: Monday, July 08, 2013 8:23 PM
To: mark.tinka at seacom.mu
Cc: Adam Vitkovsky; Andrew Miehs; cisco-nsp at puck.nether.net
Subject: Re: [c-nsp] MPLS down to the CPE



On 7/8/13 11:14 AM, "Mark Tinka" <mark.tinka at seacom.mu> wrote:

>On Monday, July 08, 2013 12:33:36 PM Phil Bedard wrote:
>
>> XR supports it in the latest revision, didn't know about the 3600 
>> support. I guess this is the C-NSP list. We have thousands of 
>> non-Cisco nodes deployed using RSVP-TE in the access layer but it 
>> requires stitching at the service layer to scale. It has shown to be 
>> scalable at least for us.
>
>You're brave.
>
>We, at the time, opted to wait for IP LFA since RSVP-TE in the Access 
>(even just to the adjacent PE routers) just didn't look 
>administratively feasible, let alone scale :-).
>
>I'd be glad to hear more about your experiences.

It really depends on how things are deployed and how distributed they are.
 We distribute the larger aggregation nodes to various sites and then
generally have access rings with access nodes hung off of those.  We
typically do not have too many nodes on a single 1GE ring, they are balanced
using passive CWDM/DWDM.  Each access node has a single TE LSP to
each aggregation node over which all services ride.   With access rings
with less than 10 nodes you have a low number of actual LSPs.  The agg nodes
might terminate 20-30 access rings resulting in most cases well under a
thousand LSPs, which isn't really a big deal for modern equipment.
 The services are generally either terminated or stitched at the agg node,
so it does result in quite a few unique LDP sessions, but the scale of those
are in the thousands today.  It's also not terribly difficult to add
more agg nodes but to date we haven't had to do that.   I can think of
high density situations where this model probably wouldn't work out due to
control plane scale issues.


>
>> Unified/Seamless MPLS has been around for years and would be ideal 
>> but the end nodes have traditionally had poor BGP support. The agg 
>> nodes will do BGP-LU to LDP translation but the LFIB capacity is 
>> fairly small.
>> Requires using something like LDP DOD to really work.
>> The nodes support using an aggregate or default prefix for the LDP 
>> IGP route.
>
>MPLS-in-the-Access has been around for a long time, but only if you 
>were prepared to spend a lot on big boxes in order to have good 
>support.
>
>Some folk decided to put in the 7604's or MX240's, but those were all 
>simply still too big.
>
>Luckily, we waited long enough until the Brocade CER/CES, the Cisco 
>ME3600X/3800X and the Juniper MX80 started to materialize.
>
>Each of these has great all-round support (including QoS in the 
>ME3600X/3800X, which has always been a turn-off in Cisco switches), and 
>we went for the ME3600X.
>
>So, now at least, operators have a real shot at extending MPLS into the 
>access for the price, without compromising features.
>
>Mark.

We have been using ALU gear in that capacity for going on 4 years now, but
they have only had nodes capapable of doing a 10G ring and drop for about a
year now, but that was a very small part of our overall business.
Cisco and Juniper didn't even make it past the early RFP stages because at
the time they had no comparable equipment.  I'm not going to say the gear is
perfect, and has been lacking things like H-QoS until very recently, but all
of their boxes have had feature sets for years that Cisco/Juniper are only
now getting to.  Now their boxes support things like RFC3107/LDPoRSVP/PBB
even on the access nodes.

Phil