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

[Andy Dills]

On Wed, 15 Sep 2004, Andy Dills wrote:

> On Wed, 15 Sep 2004, Rodney Dunn wrote:
>
> > 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.
>
> Wow, you are the cisco master :)
>
> Why is it that doing this pushes the WFQ to the VIP instead of the RSP?
>
> Since WFQ is the default queueing on T1s, you have to wonder why that's,
> by default, handled on the RSP. I'll go enable this on all the T1
> interfaces.
>
> Thanks, the show interface <int> stat is helpful for determining whether
> or not packets are being dCEFed or not.

Just wanted to be clearer for the archives...

The solution was indeed re-enabling WFQ.

Once dCEF is enabled, you must specifically re-enable WFQ in order to
initiate VIP-based WFQ.

Even if you don't use per-packet load balancing, once you enable dCEF on a
7500, you should still go and push the queueing to the VIP. Otherwise,
you'll notice a ton of "output buffers swapped out" in your interface
stats. No need to be passing those packets back to the RSP for queueing.

So, how do you know if you're using RSP-based or VIP-based fair queuing?

Do a show interface, and look for one of these lines, between the Input
and Output Queue lines:

Queueing strategy: weighted fair

Queueing strategy: VIP-based fair queuing

The first is RSP-based, and the second is VIP-based, obviously. If you're
using dCEF, you definitely want the second.

Thanks Rodney!

Andy

---
Andy Dills
Xecunet, Inc.
www.xecu.net
301-682-9972
---