[c-nsp] ip cef optimize neighbor resolution

[Ross Vandegrift]

On Thu, Nov 06, 2008 at 08:49:27PM +1100, Lincoln Dale wrote:
> the feature is essentially an enhancement for how CEF Gleans and the 
> like are handled.
> if you're familiar with how those kinds of things work, they would 
> typically need to punt to software to resolve a (not yet available) 
> adjacency through ARP or similar.
> 
> the enhancement/optimization here is that it can do it with fewer CPU 
> cycles.

Well that always sounds like a good idea!

> as to whether this would help in your scenario, its certainly a 
> possibility.  but unless what you're seeing is a relatively short period 
> of higher CPU during ARP / IP to MAC discovery, me thinks you may want 
> to look into what your ARP aging timers are relative to your MAC aging 
> timers.

Hmmm, we do see ARP Input spikes, but the symptoms are a pretty
constantly elevated CPU utilization.  Sounds like a win but not a
panacea.  Do you know if a change is service impacting?

Our MAC aging timers are 300 seconds and ARP is 4 hours.

-- 
Ross Vandegrift
ross at kallisti.us

"If the fight gets hot, the songs get hotter.  If the going gets tough,
the songs get tougher."
	--Woody Guthrie