[c-nsp] Sup720-3BXL load balancing re-visited

[Chris Griffin]

bgp bestpath compare-routerid?

Kristofer Sigurdsson wrote:
> Hi,
> 
> This morning, I believe we got something solid to present about load
> balancing in the Sup720-3BXL, IOS 12.2(18)SXF being broken.
> 
> Scenerio.  All links are POS STM-1's.  Our network is behind the 7600,
> the 7500's are edge routers to our upstreams, located in London/New
> York.  Same AS on both sides, so equal load balancing (mostly).
> 
>                                                                       
> 7500, RSP8, VIP4-80 (7500 no 1)
>                                                                      /
> 7600, Sup720-3BXL w/ SIP-200, 2xPOS STM1
>                                                                     \
>                                                                     
> 7500, RSP16, VIP6-80 (7500 no 2)
> 
> Just before 1 o'clock last night (UTC), the link to 7500 no 2 went
> down for about 40 secs due to unknown failure (we're investigating
> that).  All outbound (from the 7600) traffic moved to the other link
> (to 7500 no 1) instantly, but it never came back once the other link's
> state was restored.  Inbound traffic behaved normally (as that's of
> course dictated by our provider's routers, on the other side of the
> 7500's).
> 
> Okay, now we've got all outbound traffic going to 7500 no. 1, but
> inbound traffic load balancing normally.  We tried clearing all BGP
> peers on 7500 no. 2.  Then all outbound traffic went to 7500 no. 2 and
> nothing to 7500 no. 1.  Still, inbound traffic was normal.  We cleared
> all BGP peers on 7500 no. 1, and all traffic shifted to no. 1, nothing
> to no. 2.
> 
> We cleared all BGP peers on the 7600, now we have almost equal load
> balancing (we've found that we really can't get equal load balancing
> on the 7600, it's more like 1/3 vs. 2/3, but it's better than all
> going one way and nothing the other!).
> 
> Now, my opinion is that this is a problem with load balancing in the
> 7600.  We already have "mls ip cef load-sharing simple", which was
> supposed to solve this problem, at least for MPLS tagged routes.  Any
> other ideas?
> 
> Thanks,
> Kristo
> 
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/

-- 
Chris Griffin                                cgriffin at ufl.edu
Sr. Network Engineer - CCNP                  Phone: (352) 392-2061
Network Services / Florida LambdaRail        Fax:   (352) 392-9440
University of Florida                        Gainesville, FL 32611