[j-nsp] Provider segregation and subscribed services

Erdem Sener erdems at gmail.com
Wed Apr 12 17:46:12 EDT 2006


Hi Paul,

 Very similar issues have been discussed on the list before. As far as
I can recall, the tendence is for what you call 'vrf-ish' solution
since you cannot control 'as-path selection' sort of alghoritms
otherwise.

HTH
Erdem


On 4/12/06, Paul Connally <paul.connally at gmail.com> wrote:
> I've got something I'm working on that's giving me a bit of a
> headache.  We have two upstream providers ("regular" Internet and
> Internet2).  We're taking routes via BGP from both, and we have set
> local preference so that routes learned from Internet2 are preferred
> over regular Internet routes.  All of our downstreams peer with us and
> their own Internet providers, and this tends to work pretty well; the
> downstreams that subscribe (aka pay money) for the regular Internet
> service get sent the full routing table, and those that don't only get
> sent the Internet2 routes.  Since our big downstreams peer with their
> regular providers, there's not really a problem.
>
> We're going to start having some smaller customers who will only be
> 'subscribing' to the Internet2 service.  It's probably safe to assume
> that some of them (due to limited hardware, expertise, or whatever)
> won't be peering with their regular Internet provider, but will be
> using a default route.  Here's a big problem I see happening:
>
> Small customer peers BGP to us.  We advertise to them the Internet2
> routes that are currently installed in our inet.0 table.  Say for
> example Joe's Barber College is advertising 192.168.0.0/16 to
> Internet2 and to the regular Internet.  Since we prefer the learned
> Internet2 routes in our network, we'll advertise the /16 downstream
> with an Internet2 AS path.
>
> However, Joe's Barber College is, for whatever reason, also
> advertising a smaller network (192.168.1.0/24) to the regular Internet
> ONLY and not to Internet2 (I've seen this in practice; there's several
> instances of this currently).  The problem comes with the customer
> downstream from us who's not peering BGP to their regular provider
> when they want to reach a host on that /24.
>
> Since we advertise the aggregate /16 via BGP to them, packets destined
> to that /24 are going to be preferred via their link to us using the
> Internet2 BGP path over their default route to the regular Internet.
> The packet comes into our network, looks at the inet.0 routing table,
> and sees the more specific /24 route via the regular Internet.  Since
> our downstream has not subscribed to our regular Internet service, our
> filter drops the packet.  That /24 is now basically black-holed to our
> Internet2-only downstream customer.  The easy solution is to tell them
> "peer with your regular Internet upstream" to learn that specific /24
> from them, but there's a possibility that might not be an option.
>
> I'm thinking the solution is going to be something VRF-ish; set up a
> 2547 VPN that holds the Internet2 routes from our upstream provider as
> well as the Internet2-only customer routes.  We're not currently using
> any routing-instance configurations, and I'd prefer not to have to
> reconfigure my Internet2 peer under a separate routing-instance ( and
> risk breaking connectivity to any current 'full' connectors).  I'd
> like to be able to just copy the Internet2 learned routes from inet.0
> into the separate Internet2 VRF, but haven't found a very intuitive
> method for doing this.
>
> There's also the chance that at some point in the future, we'll be
> peering with another "private" network that will also be offered as a
> subscribed service.  That means that basically I'll need to be able to
> say "Customer A,  you can use these three different upstreams",
> "Customer B, you can use these two upstreams", and "Customer C, you
> can use only this one specific upstream".
>
> Any advice?
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp at puck.nether.net
> http://puck.nether.net/mailman/listinfo/juniper-nsp
>



More information about the juniper-nsp mailing list