[nsp] GE aggregation for last mile Ethernet

[Pete Kruckenberg]

My bad for responding with a non-Cisco solution on the Cisco 
list.

I had breakfast with my local Foundry SE, who described how
he had deployed a GE solution in a 7000-home planned
neighborhood network, that I think is one of the most 
elegant I've seen.

Cisco may have a solution like this, I haven't looked yet.

I'll do my best to describe it. If you're interested in more 
detail, I'd be happy to get you in contact with him.

Foundry has "VLAN groups" which allow one to apply certain
policies and features on a multi-VLAN level. Basically, a
tagged/trunk link can be treated in certain wais as if it
was a single VLAN.

One feature that can be applied to a VLAN group is "ip
follow-me" or something like that (I was eating, didn't take
written notes). As far as I can tell, the "ip follow-me" me
feature is like a simple virtual router, based on proxy ARP.  
It can be enabled per-VLAN, within the VLAN group. This has
an added bonus of serious address space conservation (ie one
host address per CPE port, instead of one subnet per port).

So, the way this architecture works is this. Put Cisco
3550's (could be any Layer2 switch) in the neighborhoods,
with each last-mile connection on it's own VLAN.  
Trunk/tagged port from the 3550 to a Foundry core switch
(BigIron or so). Every 3550 has the *exact same*
configuration, except for the management IP address (this is
a big bonus for operational simplicity). The 3550's also use 
very basic features, so less likelihood of software 
failures.

The Foundry switches use VLAN groups and ip follow-me to
enable Layer3 features between the last-mile connections (on
a per-customer basis).  The Foundry switches also handle
rate-limiting, QoS, accounting, etc. (All at line-rate, he
says).

Scales to 32,000 VLANs.

I would guess this could probably be extended to support
redundancy (FSRP/VRRP-wise).

Sounds pretty cool, inexpensive, simple to manage/operate,
and scalable. Is there any way to do this with Cisco?

Pete.

On Wed, 25 Jun 2003 Jack.W.Parks at alltel.com wrote:

> Date: Wed, 25 Jun 2003 14:18:54 -0500
> From: Jack.W.Parks at alltel.com
> To: cisco-nsp at puck.nether.net
> Subject: [nsp] GE aggregation for last mile Ethernet
> 
> I've been doing research for deploying Catalyst for GE aggregation in
> last mile Ethernet scenarios.  After scouring through CCO, it seems
> there are two limits aside from physical port limitation.  This seems to
> be common across the 3550/4500/6500 platforms.
> 
> * Limit 1 - Active VLANs
> 
> Although you have 4096 VLAN id's, you can only have ~1000 active VLANs.
> In the extended VLAN range, it appears that each configured ext. VLAN is
> assigned to a lower (1-1006) VLAN number. Cisco's recommendation is to
> assign the >1000 VLAN ids starting at 1 and growing n+1.  For extended
> VLAN ids, assignments should be made starting at 4096 and decrementing
> by 1.  This is to prevent overlapping VLAN id's
> 
> * Limit 2 - PV+STP
> 
> The maximum number of Per VLAN Spanning Tree instances is 128. After you
> grow past 128 VLANs you will need to configure MSTP and begin grouping
> VLANs into common Spanning Tree instances.  If you haven't taken this
> into consideration on initial deployment you could be reconfiguring your
> network
> 
> Knowing these two limits you could potentially install a 6509 and only
> use a couple ports before reaching the maximum active VLANs or changing
> over to MSTP.  This would be a huge waste of capital.
> 
> My questions are: 
> 1) How do you size the switch?  Are other NSP's using Cat6506 w/ Sup1 or
> beefing up the chassis to 6509 w/ Sup720?  
> 2) Do you limit the amount of VLANs per customer facing port (i.e. 100
> per GE)? Potentially a customer could ask for 500 VLANs in a hub and
> spoke topology effectively taking 1/2 of the available VLANs.
> 3) Inter-connecting switches seems across POPs seems ludicrous.  Would
> there be a compelling reason to do this if you have an MPLS - L2VPN?
> 4) What is the average number of VLANs per GE port are others seeing out
> there?
> 
> Thanks for your time...
> 
> Jack
>  
> Jack W. Parks IV
> Sr. Network Engineer
> ALLTEL Communications
> jack.w.parks at alltel.com
> Work: 501-905-5961
> Cell: 501-680-3341
> 
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> http://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/
>