Re: -mm seems significanty slower than mainline on kernbench

[Peter Williams <pwil3058@bigpond.net.au>]

Peter Williams wrote:
> Martin Bligh wrote:
> 
>>
>>> Thanks and looks like the results are in from 2.6.14-rc2-mm1 with the 
>>> patch backed out.
>>>
>>> Drumroll....
>>>
>>> http://test.kernel.org/perf/kernbench.moe.png
>>>
>>> The performance goes back to a range similar to 2.6.14-rc1-mm1 (see 
>>> 2.6.14-rc2-mm1 + 20328 in blue). Unfortunately this does implicate 
>>> this patch. Can we put it back into -mm only and allow Peter's 
>>> tweaks/fixes to go on top and have it tested some more before going 
>>> upstream?
>>
>>
>>
>> Hmm. Looks like we didn't get this as fixed up as I thought. Moe seems 
>> to be fixed (16x NUMA-Q), but elm3b132 is not (it's 4x, flat SMP ia32).
>> Look at the latest graphs ....
>>
>> Is it possible it only got fixed for NUMA boxes?
> 
> 
> It should be totally independent.
> 
> But anyhow, I can't see the problem.  The only numbers that I can see 
> for 2.6.16-rc1-mm[1|2] (which are the ones with the latest fix) on this 
> graph are approx. 101 which is much the same as the best of the rest. Or 
> have I missed something?

Oops.  I was looking at the graphs for Moe but 
<http://test.kernel.org/perf/dbench.elm3b132.png> doesn't appear to be 
demonstrating a problem either.  Given the fluctuation in the 2.6.16-rc1 
results (235, 234, 211, 228.5 and 237.5), the results for 2.6.16-rc1-mm1 
(229) and 2.6.16-mm2 (219) aren't significantly different.

Peter
PS I have a modification for kernbench that calculates and displays the 
standard deviations for the various averages if you're interested.  This 
would enable you to display 95% (say) confidence bars on the graphed 
results which in turn makes it easier to spot significant differences.
-- 
Peter Williams                                   pwil3058@bigpond.net.au

"Learning, n. The kind of ignorance distinguishing the studious."
  -- Ambrose Bierce