[c-nsp] cef load-sharing per-packet bizzare behavior

Rodney Dunn rodunn at cisco.com
Wed Sep 15 15:10:50 EDT 2004


Oh..and one thing I did just remember during lab testing.

Make sure if these are T1's you either do:
fair-queue
or 
no fair-queue 

so it's no doing RSP based WFQ and the packets are
dCEF switched.

Rodney

On Wed, Sep 15, 2004 at 03:01:31PM -0400, Rodney Dunn wrote:
> Are you doing this on the RSP or the VIP?
> 
> Do this:
> 
> 
> clear counters
> wait 30 seconds
> sh int se 5/1/0 stat
> sh int se 5/1/1 stat
> sh int (inbound traffic interface) stat
> 
> 
> if-con <inbound slot>
> sh ip cef <dst-prefix> int
> sh ip cef <dst-prefix>
> 
> 
> 
> Rodney
> 
> 
> On Wed, Sep 15, 2004 at 02:23:09PM -0400, Andy Dills wrote:
> > On Wed, 15 Sep 2004, Rodney Dunn wrote:
> > 
> > > Does the downstream routers show the same unequal
> > > load inbound?
> > 
> > Yes. The customer was the first to notice it. The same unequal load is
> > seen from the CPE as from this end.
> > 
> > > Is that the only prefix being forwarded over
> > > the links?
> > 
> > Yes, absolutely. One of the first things I checked. All of the routes are
> > announced to me via BGP, and I've got dual static routes to the address I
> > peer with. This has been verified 100% in the cef table. (sh ip cef | incl
> > <interface>)
> > 
> > One more tidbit to add. When I shut one interface, the bulk of the load
> > shifts to the other interface (and stays there), and vice versa.
> > 
> > So, for instance, if I shut s5/1/0, only s5/1/1 gets used. The moment I
> > 'no shut' it, s5/1/1 has about 2/3rds of the traffic and s5/1/0 gets about
> > 1/3rd of the traffic (this is measured 10 minutes after bringing back up
> > the interface). If I then go and shut s5/1/1, when I 'no shut' it, the
> > 2/3rds of the traffic moves to and stays on s5/1/0.
> > 
> > So, in essence, whichever interface has been up longer bares the bulk of
> > the traffic.
> > 
> > Thanks for any advice,
> > Andy
> > 
> > ---
> > Andy Dills
> > Xecunet, Inc.
> > www.xecu.net
> > 301-682-9972
> > ---


More information about the cisco-nsp mailing list