[c-nsp] NPE-G2: mixed interrupt/polling packet processing on POS interface

[Rodney Dunn]

Philippe,

I need to check back with the BU on it. I thought they were putting
a document out on this G2 processing reporting issue.

Basically due to the way the G2 does CPU accounting it looks
like it's higher at lower loads based on the CPU measurements.

If you put it in the lab and measure NDR's the CPU can forward
much more than the G1 even though the G2 "shows" higher CPU
in the output.

That has confused a lot of folks and understandably so.

Rodney

 

On Wed, Feb 20, 2008 at 03:12:27PM +0100, Philippe Strauss wrote:
> Hello c-nsp,
> 
> We are running c7200 vxr with NPE-G2, but are seeing way to high CPU
> usage (85%) at only 225kPPS in+out.
> 
> It _seems_ it's due to the fact we're using a POS interface.
> I _guess_, the G2, when using GigE interface, is doing a mixed
> interrupt/polling mode for processing each packets.
> (like linux NAPI: above a certain PPS value, the driver switch
> from interrupt to polling mode). It's a guess.
> 
> But the CPU usage is for sure much lower when using only the 3 GigE
> embeded on the mainboard.
> 
> Keeping interrupt only processing for POS, which may be used
> to switch voice traffic, has some rational behind it.
> 
> But we're using it only to switch data/ip, and I'm wondering
> if there's a way to lower the CPU usage despite using POS?
> 
> like a switch to enable mixed interrupt/polling driver on the POS?
> 
> would be highly helpfull.
> 
> regards.
> 
> -- 
> Philippe Strauss
> av. de Beaulieu 25
> 1004 Lausanne
> http://philou.ch
> _______________________________________________
> 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/