[c-nsp] Sup720-3BXL load balancing

[Kristofer Sigurdsson]

Hi list,

I was wondering how a 7600 series router with a Sup720-3BXL handles
load balancing between equal BGP paths.

I'm experiencing undesired behaviour in this area from at least two
routers, both running IOS 12.2(18)SXF.  They seem to load balance
extremely unevenly.  Same behaviour for the global routing table and
within a VRF.  I've got an example from within a VRF:


router#sh ip bgp vpnv4 vrf FOO x.x.9.0
BGP routing table entry for <ASN>:<RD>:x.x.9.0/24, version 1002918
Paths: (2 available, best #1, table FOO, not advertised to EBGP peer)
Multipath: eBGP iBGP
  Advertised to update-groups:
     3
  <Remote ASN>
    x.x.x.142 from x.x.x.142 (y.y.14.2)
      Origin incomplete, localpref 100, valid, external, multipath, best
      Community: no-export
      Extended Community: RT:<ASN>:<RD>
  <Remote ASN>
    x.x.x.146 from x.x.x.146 (y.y.14.100)
      Origin incomplete, localpref 100, valid, external, multipath
      Community: no-export
      Extended Community: RT:<ASN>:<RD>

As you can see, the two paths are equal, but still, one of them is
considered "best", but I've got BGP multipath enabled.

>From checking the routing table:

router#show ip route vrf FOO x.x.9.0
Routing entry for x.x.9.0/24
  Known via "bgp <ASN>", distance 20, metric 0
  Tag <Remote ASN>, type external
  Redistributing via ospf 104, ospf 103
  Advertised by ospf 104 metric 30 metric-type 1 subnets route-map <BAR>
               ospf 103 metric 400 metric-type 1 subnets route-map <BAR>
  Last update from x.x.x.142 02:56:45 ago
  Routing Descriptor Blocks:
  * x.x.x.146, from x.x.x.146, 1d07h ago
      Route metric is 0, traffic share count is 1
      AS Hops 1
      Route tag <Remote ASN>
    x.x.x.142, from x.x.x.142, 02:56:45 ago
      Route metric is 0, traffic share count is 1
      AS Hops 1
      Route tag <Remote ASN>

Still, a vast majority of the outgoing traffic (from us to them) goes
over the link with the x.x.x.142 IP address.

I've been experiementing a little with this, it seems the majority of
the traffic will go over the link which has the latest routing
information (i.e. I reset the BGP session on one link, all the traffic
goes to the other one while the session comes back up, then it all
moves to the one I reset).

Any pointers?

Thanks,
Kristo