Re: Clear performance regression on reaim7 in 2.6.15-git6

[Andrew Morton <akpm@osdl.org>]

"Martin J. Bligh" <mbligh@google.com> wrote:
>
> OK, looking back through the perf results, these two graphs clearly show
> a perf regression in reaim7 from 2.6.15 to 2.6.16-rc1. We're loosing 
> over 50% of the performance.
> 
> http://test.kernel.org/abat/perf/reaim.moe.png
> http://test.kernel.org/abat/perf/reaim.elm3b67.png

Something very bad has allegedly been happening in -mm.  It would have been
nice to have heard about this problem in that multi-month wwindow there. 
Does no human look at these numbers?

> Drilling down (there's not enough detail on the graphs for releases that
> far back), I see it's actually between -git5 and -git6
> 
> These are both ia-32 NUMA machines, one is an x440, the other is NUMA-Q.

The 2.6.15-git5 -> 2.6.15-git6 diff is enormous.  I'd assume some bad thing
got transferred into mainline then.

How does one reproduce this?  Which type of filesystem, which command
line/config file?

Thanks.