Non directly connected bgp peers

[carl_townshend]

Hi all

question about non directly connected ibgp peers,

If I have router A B C all connected in a line and B is not running BGP, I know I can still form a relationship as long as the router is reachable, so the next hop from A to C will go via B.

A sends the packet to B, if B does not know the final destination, does it drop the packet ? so in the routing since knowing next hop will only form the relationship and not actually get the packet there?

for this to work would B need to run BGP also or have a copy of all the routes it’s trying to reach via some other IGP etc?

in large networks where there are lots of non directly connected ibgp peers such as isp’s, how do they get around this?

cheers 

[Harold Ritter]

Hi @carl_townshend ,

> A sends the packet to B, if B does not know the final destination, does it drop the packet ?

Yes, unless you run MPLS in your network, in which case the packet will go from A to C as labeled traffic. This scenario is often referred to as "BGP free core".

> for this to work would B need to run BGP also or have a copy of all the routes it’s trying to reach via some other IGP etc?

Yes, B would need to run BGP as well, unless you implement a BGP free core (MPLS core).

> in large networks where there are lots of non directly connected ibgp peers such as isp’s, how do they get around this?

They run MPLS in the core.

Regards,

[carl_townshend]

Hi Harold

How to large scale providers get around this?

Are most ISPs now using MPLS to get around this fully meshed requirement? and do ISPs also use it for internet routing and not just MPLS VPNs for customers?

If so, would they effectively just have all the PE routers peering to each other?

Would you have a diagram for a typical providers network design and how they achieve this?

Many thanks

Carl

[Harold Ritter]

Hi @carl_townshend ,

> Are most ISPs now using MPLS to get around this fully meshed requirement?

Most service providers use route reflectors to get around the full mesh requirement. MPLS is used to deliver additional services (L2VPN, L3VPN, etc) and/or to implement a BGP free core.

> If so, would they effectively just have all the PE routers peering to each other?

The PEs normally only run a BGP session with the route reflectors. 

> Would you have a diagram for a typical providers network design and how they achieve this?

I do not specifically have a diagram, but as I mentioned, you can picture a network where the PEs peer with the route reflectors and the core do not need to run BGP.

Regards,

[paul driver]

Hello
You are correct -  when you have ibgp routers as above, they will require a full mesh adjacency with each other to allow them to have full routing knowledge of each other’s networks, otherwise as you so rightly stated "B" rtr in this instance would not be able to route traffic from either A/C rtrs 

An alternative to having a large ibgp full mesh topology you could use route reflectors and for very large ibgp topology’s bgp confederations, both of which would drastically reduce the number of ibgp peering’s required.

[MHM Cisco World]

from site A to Site B, the ISP router not need all routes inside both site, ISP router will use as tunnel. 
Now ISP router will use MPLS if the Core is support that 
or ISP router use GRE or any other tunnel protocol is the Core is IP.

friend MPLS or GRE only add other headed to original packet this headed is known by ISP router.