[c-nsp] Multihomed network -> iBGP questions

[Thierry]

Dears,

We will go with route reflectors. The two routers A and B will be RR. Some
questions:

- Could we still use peers group with RR? 

- Is it needed to create a cluster for router A and B? The issue is that A
will not accept routes advertised by B because it's the same cluster. We
would like to choose the best routes to Internet (we will have two different
providers) Do I have to create two different clusters? 

Thanks.

Thierry


-----Original Message-----
From: Mark Tinka [mailto:mtinka at globaltransit.net] 
Sent: vendredi 21 mai 2010 5:26
To: cisco-nsp at puck.nether.net
Cc: Thierry
Subject: Re: [c-nsp] Multihomed network -> iBGP questions

On Thursday 20 May 2010 05:42:36 pm Thierry wrote:

> As internal routing protocols, I would like to use iBGP
>  for all customers' routes and OSPF for all backbone
>  routes (loopbacks and PtP between the routers).

Good.

> I would like to have the full internet table only on the
>  two routers connected to the providers (A and B). All
>  the others (C to J) will only have the internal routes +
>  a default route coming from the two border routers (A
>  and B) using OSPF.

Okay.

Have you also considered the possibility of originating 
default via iBGP?

> I would like also that all outgoing traffic goes to the
>  same router (for example A) and this router will choose
>  to send the traffic to the provider connected to himself
>  or to send the traffic to router B (in case it has a
>  better route), which will send to his provider.

You can do this through BGP policy on the iBGP sessions 
between your border and edge routers.

> I would like also that the network is scalable, for
>  example if we have in the future a new customer
>  connected to the router F and would like a full internet
>  table. In this case, we must advertise the full table to
>  router F, which can advertise to the customer.

If you anticipate that this is where you're headed, why not 
do this from Day One? Gives you more experience too, in case 
you get customers asking for it.

If your edge routers can take a full table, why not? If they 
can't, then that's another story :-).

> My idea was:

> -          OSPF for backbone routes.

Yes.

> -          iBGP for customer routes.

Yes.

> -          eBGP with the providers.

Of course.

> -          iBGP between A and B with no restrictions.

Consider running these as route reflectors for the rest of 
your iBGP speakers. Of course, BGP policy configuration can 
start to get complicated since they're also handling 
upstream peering. But it's not impossible.

Alternatively, if you can consider different boxes as route 
reflectors in the network, that's another option.

> -          Full mesh iBGP between all the routers (except
>  between router A and B) with a filter-list applied out
>  -> ip as-path seq 1 permit ^$ this should only advertise
>  prefixes originated inside the AS and where a network
>  statement is configured into BGP.

As mentioned, since scaling is on your mind, consider route 
reflectors.

> -          Default originate on router A and B inside
>  OSFP configuration with a different metric. Ex: A=
>  "default-information originate metric 1 metric-type 1"
>  and B= "default-information originate metric 100
>  metric-type 1".

Originating default in iBGP should also work. Either way, 
you should be fine.

> -          For the example explained above, we change on
>  router A and B the configuration with router F -> move
>  to the iBGP with no restrictions.

Again, consider doing this from Day One if you don't have 
any hardware restrictions.

Hope this helps.

Cheers,

Mark.