em forwarding performance (was Proposed 6.2 em RELEASE patch

[Mike Tancsa]

At 03:06 AM 11/28/2006, Massimo Lusetti wrote:
>On Mon, 27 Nov 2006 16:36:34 -0500
>Mike Tancsa <mike at sentex.net> wrote:
>
> > OK, I added OpenBSD to the mix as well.  Results are pretty crappy
> > with the base default install. With one stream, the box essentially
> > live locks. This was just with the stock kernels from the CD.  The
> > PCIe bge nics dont work, so I cant test those.  I had a look at their
> > errata page and there seems to be some updates to those 2 nics so if
> > there is interest I can try compiling in those fixes and re-testing
>
>FWIW I would definitively like to see it. But thanks for going so far..

I will give it another try tomorrow as I dont have much time today. 
However, I did one test with a new CPU. I changed the cpu from an AMD 
3800 to an AMD4600. So going from a CPU with a clock speed of 2Ghz to 
2.4Gz.  The difference is sort of what one would expect.  In the ipfw 
tests, (with a sample of 40 seconds) opposite path forwarding with 10 
ipfw rules went from 208 Kpps to 235 Kpps or about a 10% increase as 
opposed to the ~ 15% increase in clock speed.  Perhaps bus is more of 
a limiting factor that CPU but then again some the difference could 
be explained with normal variance.

         ---Mike