BGP Pre-pending

[TRACY HARTMANN]

We have Two different ISP's and want to load balance across them.  Looking for some direction on how to complete that.  I does seem like we are doing that pretty well going outbound but inbound most of the traffic looks to only be going to one vendor.  He was suggested we use pre-pending.

We have a router that I manage with one of the circuits connected, then the other vendor connections their router into this same router.

router ospf 1
 router-id x.x.x.x
 area 0 authentication message-digest
 redistribute bgp 11111 subnets route-map BGP=>OSPF
 network x.x.x.x  0.0.0.255 area 0
 default-information originate

router bgp 11111
 bgp log-neighbor-changes
 bgp bestpath as-path multipath-relax
 network x.x.x.x  mask 255.255.255.0
 network x.x.x.x  mask 255.255.255.0
 neighbor x.x.x.x remote-as 7018
 neighbor x.x.x.x  ebgp-multihop 3
 neighbor x.x.x.x  update-source GigabitEthernet0/0/3
 neighbor x.x.x.x  soft-reconfiguration inbound
 neighbor x.x.x.x  remote-as 2222
 neighbor x.x.x.x  update-source GigabitEthernet0/0/0
 neighbor x.x.x.x  soft-reconfiguration inbound
 neighbor x.x.x.x  prefix-list WINDSTREAM-BGP-OUT out
 maximum-paths 2

Thanks

[Reza Sharifi]

You can use weight to achieve some load sharing.

Have a look at this document.

http://www.cisco.com/c/en/us/support/docs/ip/border-gateway-protocol-bgp/13762-40.html#anc22

HTH

[Richard Burts]

Reza makes a good suggestion about using weight to achieve load sharing. But using weight is effective in selecting routes for outbound traffic. And the original post says that outbound is already working fairly well and where they need suggestions is for inbound traffic. But weight will not have an effect on inbound traffic.

For inbound traffic with BGP there are a couple of alternatives to consider. One option is to split the address space and advertise one part of the address space to one ISP and to advertise the other part to the other ISP. The original post hides all of the information about addressing so it is difficult to know what to suggest. But it does suggest that they have 2 /24 networks. So it might be feasible to advertise one of the /24 to the first ISP and advertise the other /24 to the other ISP. The other option would be to use prepending. With prepending you use a route map to control advertisement of your networks to your BGP peers. What prepending does is to add extra instances of your AS number in the AS path advertised to the peer. Normally you advertise your AS number once in the AS path. But with prepending you could use the route map so that when you advertise to peer 1 you advertise the first /24 normally and advertise the second /24 with three or four copies of your AS number. Then in advertising to peer 2 you advertise the first /24 with three or four copies of your AS number and advertise the second /24 normally.

HTH

Rick

[paul driver]

Hello

Following on from richards post - And looking at your posted config, May I  possibly suggest to apply a filter towards both ebgp peers to negate your router becoming a transit path for either ISP.

ip as-path access-list 10 permit ^$

Router bgp 11111
neighbor x.x.x.x filter-list 10 out
neighbor x.x.x.x filter-list 10 out


res
Paul

[Richard Burts]

When I was making my response I was focused on the issue the post asked about. I agree that Paul makes a very excellent point (and +5 for that) looking at the big picture and identifying an aspect that would be very important in a live production BGP environment.

HTH

Rick

[Joseph W. Doherty]

Paul's point is excellent and Rick does well to highlight its importance.

However, the ISPs should also block transit traffic from a customer.