[reSIProcate] Awful performance on Leopard

[Adam Roach]

On 07/12/2008 12:45 AM, Alan Hawrylyshen wrote:
>
> On Jul 11, 2008, at 21:52 , Adam Roach wrote:
>
>> So.... for <10k transactions, we would run *faster* with the firewall 
>> turned on some non-trivial fraction of the time? This is getting more 
>> bizarre...
>
> I'm not so sure that's valid. I mean the variability  under 10K 
> transactions (total) is so large that I think we have to assume that 
> we lose the precision in the range...

I agree in general, but the grouping for the green points for sub-10k is 
in a very predictable range. The red points are all over the place, 
including far more than I think sampling error can account for in the 
"faster than expected" range. Even the moving average graph shows this 
"faster than expected" trend on the low end. Is the firewall messing 
with our timer resolution somehow?

/a