[nsp] C7200 Bandwidth Points

Stephen J. Wilcox steve at telecomplete.co.uk
Mon Jul 14 09:39:16 EDT 2003


I did a comparison between the two ports on one PA-2FE and also between one port 
each on two PA-2FE..

I have to add tho this really wasnt very scientific or a well controlled 
experiment! 

One thing I wasnt too sure about was the supposed limit in terms of packets per 
second that the different hardware has a published by cisco.. seems that in 
terms of performance, PA-FE > PA-2FE > PA-2FE-ISL

I suspected there are some difference between the different chipsets that are in 
use on those items..

Do you know, in the case of a PA-2FE switching packets coming in one port to the 
other port on CEF, where does the data go? FE to PA to BUS to CPU and back or 
are there any shortcuts?

Steve

On Mon, 14 Jul 2003, Dmitri Kalintsev wrote:

> Did you try to comare the performance figures you've got with what you can
> get out of two PA-FEs sitting on the same PCI bus? (I.e. both FE PAs in
> either even or odd numbered slots). This might hold the key to the
> performance difference you saw.
> 
> On Sat, Jul 12, 2003 at 11:23:05AM +0100, Stephen J. Wilcox wrote:
> > On Sat, 12 Jul 2003, Gert Doering wrote:
> > 
> > > On Sat, Jul 12, 2003 at 10:36:06AM +0100, Stephen J. Wilcox wrote:
> > > > subtleties too such as if u use a multiport adaptor eg PA-2FE you can load them 
> > > > higher by switching between the two ports as your not using the backplane!
> > > 
> > > You're sure of that?  As the 7200 is a centralized architecture, I'd
> > > assume that everything has to go to the CPU and back...
> > 
> > no not really! but i did some load testing and managed to get a significantly 
> > higher throughput once the bus was loaded by adding traffic in and out of the 
> > ports on the same PA
> > 
> > there could be other explanations, something to do with the limit of the PA-2FE 
> > rather than the bus.. 
> > 
> > Steve
> ---end quoted text---
> 
> SY,
> 



More information about the cisco-nsp mailing list