Re: Two sessions BGP on same AS operator

sebastien3 · ‎10-08-2021

Hello,

I would like your opinion on the BGP configuration of two full-views from the same operator.

R1 has a full BGP session with AS174
R2 has a full BGP session with AS174
R3 has a full BGP session with AS6939
All routers (R1 + R2 + R3 + R4) are full-mesh.

By default BGP uses load-sharing, traffic will flow 50/50 on R1 and R2 for traffic coming from AS174, we agree on that ?

Before adding the second full view on R2, I had another operator that I deleted because it was too expensive...

Each router had a weight in order to always favor the exit by a router.

Example :
If I am located in the R3 area then R3 will be my next-hop (then BGP on R3 decides to get me out by R4/R2/R1)
If I am located in the R2 area then R2 will be my next-hop (then BGP on R2 decides to get me out by R4/R3/R1)
If I am located in the R1 area then R1 will be my next-hop (then BGP on R1 decides to get me out by R4/R3/R2)

Now on R2 I have the same transit as R1.

My fear is to have traffic coming from AS174>R2>Switch> area of R1 when we should do AS174>R2>area of R2 !

What would you do in my case for the two BGP AS174 sessions ?

Thanks !

Georg Pauwen · ‎10-08-2021

Hello,

--> By default BGP uses load-sharing, traffic will flow 50/50 on R1 and R2 for traffic coming from AS174, we agree on that ?

The opposite is the case: by default, BGP selects only a single best path and does not perform load balancing.

Giuseppe Larosa · ‎10-08-2021

Hello @sebastien3 ,

as already noted by @Georg Pauwen BGP by default does not perform load balancing but it selects one best path for each prefix.

>> By default BGP uses load-sharing, traffic will flow 50/50 on R1 and R2 for traffic coming from AS174, we agree on that ?

We do niot agree on this AS174 router would need maximum-paths 2 under router bgp to perform load balancing towards R1 and R2 and only for prefixes advertised by both R1 and R2.

You have four routers R1,R2, R3, R4 in your AS, are you using an IGP like OSPF between them ?

Each router has an eBGP session R4 with IX and R3 with AS6939 .

Are you receiving a BGP full table also from AS 6939 ?

is R4 connected to an Internet Exchange Point and it has peerings relationships with some other AS there?

You should increase local-preference on all prefixes learned by R4 on the IXP.

So that a received prefix arriving from an iXP eBGP peer will be preferred over the same prefix learned from AS174 and/or AS6939.

Finally, if you are running an IGP the exit point towards AS174 depends on the point of view :

Assuming the second EBGP session with AS174 towards R2 is using a different VLAN

R4 will pick up R1 for the lowest IGP metric to iBGP next-hop assuming here that R4 is only receving partial views from peers in the IXP.

R3 will pick up R2 for the lowest IGP metric to iBGP next-hop.if the path via R2 from AS174 for a prefix has better attributes then those of the path received from AS6939 ( assuming here that also AS6939 is providing a full Internet table)

To ne ntoed with a different topology with some internal iBGP routers you could achieve load balancing outbond with

eibgp multlpath 2 if supported on device.

Hope to help

Giuseppe

sebastien3 · ‎10-09-2021

as already noted by @Georg Pauwen BGP by default does not perform load balancing but it selects one best path for each prefix

I am not sure... Since AS174, to join an IP in my AS, if I do not use preprend or weight it seems to me that the traffic will be able to come as well via R1 as R2 randomly ? If AS174 does not force traffic to one of the two BGP sessions (R1 or R2). I'm stuck on thinking here...

You have four routers R1,R2, R3, R4 in your AS, are you using an IGP like OSPF between them ?

Yes I have 4 routers on my AS. I use an IGP with EIGRP.

Each router has an eBGP session R4 with IX and R3 with AS6939 .

Yes !

Are you receiving a BGP full table also from AS 6939 ?

Yes, I receive the full table from AS2639.

is R4 connected to an Internet Exchange Point and it has peerings relationships with some other AS there?

Yes !

You should increase local-preference on all prefixes learned by R4 on the IXP.
So that a received prefix arriving from an iXP eBGP peer will be preferred over the same prefix learned from AS174 and/or AS6939.

This is already the case on R4.

To ne ntoed with a different topology with some internal iBGP routers you could achieve load balancing outbond with
eibgp multlpath 2 if supported on device.

I didn't understand, can you explain to me ?

Many thanks @Giuseppe Larosa !

Giuseppe Larosa · ‎10-10-2021

Hello @sebastien3 ,

>>

To ne ntoed with a different topology with some internal iBGP routers you could achieve load balancing outbond with

eibgp multlpath 2 if supported on device.

>> I didn't understand, can you explain to me ?

I just wanted to point out that with another toplogy you could try configure iBGP multipath towards R1 and R2 for best prefixes learned via AS 174 in the upstream ( to the Internet direction) .

The command should be given on a router that sees R1 and R2 with equal IGP metric to their respective loopbacks the command depending on platform and OS type and version can be supported

router bgp XXX

address-faimly ipv4 unicast

eibgp multipath 2

Hope to help

Giuseppe