[c-nsp] tag-switching advertise-tags for?

[Rodney Dunn]

On Thu, Oct 27, 2005 at 08:42:17AM -0400, Jon Lewis wrote:
> On Wed, 26 Oct 2005, Oliver Boehmer (oboehmer) wrote:
> 
> >> routes in question are not in our iBGP.  They're being learned via
> >> OSPF.
> >
> > Well, then your issue can be addressed by restricting label
> > advertisements to BGP next-hops (i.e. PE loopbacks)? Then your
> > OSPF-learnt routes won't be label-switched, voila.
> 
> Using the no tag-switching advertise tags, and tag-switching advertise 
> tags for commands on just two of our routers, I got this working such that 
> plain IP packets are routed (between the two routers) now instead of 
> tag-switched, which allows me to staticly route this customer's VOIP 
> server down a dedicated path.
> 
> BTW, upgrading from 12.2S to 12.3M did get tagged packets over MLPPP cef 
> switched instead of process switched, but it seems to have introduced a 
> new problem.  Under either 12.2S or 12.3M, the Mu1 interface has had 
> issues with output drops even though we rarely got above 50% utilization. 
> Though I know cisco recommends against it, I was able to nearly get rid of 
> them under 12.2S by increasing the output hold queue.  It just seemed kind 
> of wrong that as individual T1's there should be output queues of 40 per 
> interface, but as a MLPPP interface of 5-6 T1s, there should be a single 
> output queue of 40.

That output queue doesn't do anything unless there isn't enough space
on the tx-rings of the member links to transmit packets. Which means
that most likely you are bursting traffic. Try enabling a service policy
with policing to transmit for conform and exceed and see if you see
exceed matches. That's the #1 hardest thing to prove when people say
they see output drops. Are they bursting or not.


  Under 12.3(16), increasing the Mu1 output hold queue 
> doesn't seem to help as much.  Without any increase, I was seeing around 
> 0.5% packet drops while there was only about 1mbps going out the Mu1. 
> With a hold queue of 300, the output drops are down to under 0.05% with 
> about 3.5mbps going out the Mu1 at the moment.  I played around with 
> fair-queue/no fair-queue and tried enabling fragmentation.  None of these 
> made a difference.  I may end up just going back to cef per-packet now 
> that we can keep the voice traffic on its own circuit.
> 
> Apparently as a result of the tag-switching advertise changes, I'm now 
> seeing messages like this from time to time on the two changed routers.
> 
> Oct 27 08:16:14: %TIB-5-WDRAWTAG: 209.208.109.128/255.255.255.128, tag 
> 814; Withdrawn tag record has timed out.
> 
> Nobody's complaining yet, so I don't think its a problem.
> 
> ----------------------------------------------------------------------
>   Jon Lewis                   |  I route
>   Senior Network Engineer     |  therefore you are
>   Atlantic Net                | 
> _________ http://www.lewis.org/~jlewis/pgp for PGP public key_________
> _______________________________________________
> 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/