[j-nsp] internal BGP necessary ?

[Piotr]

Hi

I attached diagram. I have to connect together two different companies. 
They have ethernet leased lines, so i can use bgp on private AS numbers. 
Each have own OSPF. There is max 1k routes in igp. In this moment i use 
only ebgp, i removed ibgp. I made redistrubutions ospf-(e)bgp and 
(e)bgp-ospf on A1 A2 B1 B2,  i have also tags against mutual 
redistributions. Ebgp has 200 ad on cisco, For now all works, but first 
time without ibgp. I'm affraid about some problems like loops..

thanks for any advice or notice
regards,
Piotr


W dniu 2013-03-23 22:29, Patrick Okui pisze:
> Hi Piotr,
>
> On  23-Mar-2013 16:13:18 (+0200), Piotr wrote:
>> I have to connect two networks via bgp, both have own ospf area0, there
>> are 2 points od redistribution between BGP and OSPF. We use med to have
>> symmetry, because Junipers must work in flow based (ipsec).
>
> Probably a small ascii diagram would help in this case. If you're using
> MED then these two networks are two separate ASes correct? Juniper in
> one, Cisco in another?
>
>> When we set on cisco  default AD, on juniper we change bgp ad to 20 (
>
> The default AD of eBGP in Cisco is 20, but of iBGP is 200. However,
> Cisco by default will wait to see a route in the IGP before installing
> it into BGP. I usually set both to 200 (and turn off synchronisation)
> thusly:
>
> router bgp NN
>    distance bgp 200 200 200
>    no synchronisation
>
> You may also want to set bgp deterministic-med on the cisco side to
> match how most other vendors treat MED.
>
>> like in Cisco) juniper prefer routes from ibgp and there is a problem
>> with redistribution between ospf and ebgp because there is no prefix
>> from ospf in rib.
>
> one of BGP's first rules is reachability to the next hop. If BGP can't
> reach the next hop it will drop the prefix advertisment. Your IGP has to
> guarantee this.
>
>>
>> I think that when i remove internal bgp session there should be ok, in
>> rib there will be prefixes only from ospf so there will be no problem
>> wirh redistribution? There are some disadvantages when there is no ibgp ?
>
> I think my first question is ... why are you redistributing in the first
> place? iBGP will handle more prefixes than any IGP - so usually you just
> want to carry your infrastructure/interface/link prefixes in the IGP and
> everything else in BGP. If you go down this route, then a full iBGP
> mesh[*] between BGP speakers is required to avoid loops.
>
> Again, it's not easy to guess what you're trying to achieve without a
> diagram.
>
> Will OBrien's comments above give hints on getting redistribution
> working but then again, we're all shooting in the dark.
>
> --
> patrick
>
> [*] or a route reflector or ...
>