Re: pktgen

[Lennert Buytenhek <buytenh@xxxxxxxxxxxxxx>]

On Sat, Nov 27, 2004 at 01:04:53PM +0100, Robert Olsson wrote:

> 60    825789
> 64    748975
> 68    729149
> 72    719721
> 76    720204
> 80    720127
> 84    702722
> 88    702799
> 92    705107
> 96    701711
> 100   703858
> 104   692120
> 108   696139
> 112   708936
> 116   697780
> 120   677887
> 124   678158
> 128   739290
> 132   737070
> 136   736894
> 140   737645
> 144   737816
> 148   682755
> 152   648547
> 156   646464

This part is strange.  In my case the pps rate for 132-byte packets is
100kpps lower than for 60-byte packets, in your case the curve is rather
flat.  Perhaps you're CPU-bound here (or hitting some NIC limit.)

Your data is a bit noisy -- can you rerun the test for the 60-200 byte
packet range but using 10M packets per run instead of 1M?


> 160   679270

From this point onwards your HW can saturate the pipe, this is the
boring part of the graph.


thanks,
Lennert