Re: [nsp] CEF stability

From: Tony Tauber (ttauber@bbnplanet.com)
Date: Mon Jan 24 2000 - 10:06:01 EST


On Mon, 24 Jan 2000, Steve Meuse wrote:

> At 03:09 PM 01/23/2000 -0800, Jim Warner wrote:
> >Brian said:
> >
> > I'm trying to find a 12.x ios that's CEF-friendly. We've disabled it on
> > our border boxes, because it doesn't seem to play nice w/bgp at all,...
> >
> >Maybe I'm missing something. We're running a 7507 with dCEF and netflow
> >switching and a full bgp view on 12.0(8)S. What terrible things should
> >I watch out for?
> >
> >-jim warner
>
> Watch out for table mis-matches, that is, the CEF table on the line card or
> VIP is different than the CEF table on the RSP.
>
> -Steve

Note that we've not seen this as an especially big problem,
but there is the chance that some of the IPC messages will
get lost as they're transmitted across the same bus as other
traffic, so when things get tight....

The other problem we've seen is input ignores when SRAM on
the VIPs isn't that high (2MB) and there's lots of traffic
bound for one or two busy output pipes (like the normal aggregation
box) with high numbers of input interfaces (like a box loaded with
CT3s). Due to the way the buffers are carved, any given channel on
the CT3 gets not too much memory.
Watch out for VIP2-40s (can't carry more than 2MB) or cards
based on them like the CT3IP (vs. PA-MC-T3 with a VIP2-50).
Looks like upping the SRAM to 4 or 8MB works much better.

Tony



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:09 EDT