[lkml]   [2011]   [Jun]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: REGRESSION: Performance regressions from switching anon_vma->lock to mutex
On Wed, 2011-06-15 at 23:37 +0200, Peter Zijlstra wrote:
> On Wed, 2011-06-15 at 14:12 -0700, Tim Chen wrote:
> > Thanks to Andi for providing the info. We've used this workaround in
> > our testing so it will not mask true kernel scaling bottlenecks.
> tip+sirq+linus is still slightly faster than .39 here, although removing
> that sysconf() wreckage closed the gap considerably (needing to know the
> number of cpus to optimize locking sounds like a trainwreck all of its
> own, needing it _that_ often instead of just once at startup is even
> worse).


Fengguang's readahead fixes for tmpfs removed another bottleneck before
anon_vma->lock become dominant.
We've found this issue when we were testing exim earlier.
It was merged in 3.0-rc2 but not in plain 2.6.39. So with this patch on
2.6.39 we should get better comparison with 3.0-rc2.



 \ /
  Last update: 2011-06-16 03:11    [from the cache]
©2003-2014 Jasper Spaans. hosted at Digital Ocean