The MPLS WG Archive

[Jim Guichard <jguichar@cisco.com>]

Vijay,

I do not think that anyone is saying that this is the only option, it is
just one of several. If there are a few customers that wish to receive full
routes, and in my experience this is not a great number, then you can of
course offload the eBGP session from your PE router and let the customer
run directly with the internet exit point (or router that can establish the
best exit point). It is clear that not all PE routers will have VPN
customers attached that require internet access, or if they do, they do not
require full routing. So maybe the question should be what percentage of
VPN customers will want full routes, and where will these customer sites be
located ? if the answer to this is a small number with certain sites
requiring full routes then the answer may be dedicated Internet boxes,
multi-hop sessions or even restriction of VPN attachment to the PE holding
the full routes (with non-full route VPN customers perhaps located on other
PE routers within that POP) .. Jim

At 14:04 26/10/2000 -0400, Vijay Gill wrote:
>
>[ folks, it doesn't hurt to trim headers ]
>
>On Thu, 26 Oct 2000, Eric Rosen wrote:
>
>> Randy> if an enterprise wanting to deploy it is not an isp, then they'll
>> Randy> need their own layer 1 or2 connectivity between all endpoints. 
>> 
>> Just to be  clear, you are saying  that, to the best of  your knowledge,
all
>> networks fall into one of the following two categories:
>> 
>> 1. needs to carry all Internet routes to every edge
>
>Having worked at one or two promising local ISPs, I assure you that the
>plan of deploying two routers where one will do (and where one is just to
>aggregate VPN customers) is something that will not fly very well.
>
>This of course leads to a realistic scenario where you have the minimum
>number of routers necessary to satisfy the business need at the edge and
>since a couple of customers insist on having full routes, well....
>
>I suspect this is where Randy is coming from. Given that routers _barely_
>work as it is, we'd much rather wish this additional problem onto someone
>else than have it sit in the ISP core, where routers crashing lead to SLA
>violations and asosciated pain.
>
>> 2. needs complete mesh of layer 2 connections between all endpoints. 
>> 
>> I had no  idea that all networks  fell into one of these  two
categories!  I
>> don't know how I missed that ;-)
>
>Just most _realistic_ ISP networks fall into case 1.
>
>/vijay
> 


Jim Guichard CCIE #2069
Network Design Consultant EMEA
Global Solutions Engineering 

+44 208 756 8806
Mobile: +44 7802 809763