Re: [Bug #14383] hackbench regression with kernel 2.6.32-rc1

["Zhang, Yanmin" <yanmin_zhang-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>]

On Wed, 2009-11-18 at 12:09 +0100, Peter Zijlstra wrote:
> On Mon, 2009-11-16 at 23:37 +0100, Rafael J. Wysocki wrote:

Sorry for replying late. There was a severe power failure in my Lab.

Below are updates against 2.6.32-rc7 kernel.

> 
> > Bug-Entry	: http://bugzilla.kernel.org/show_bug.cgi?id=14383
> > Subject		: hackbench regression with kernel 2.6.32-rc1
> > Submitter	: Zhang, Yanmin <yanmin_zhang-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>
> > Date		: 2009-10-09 9:19 (39 days old)
> > First-Bad-Commit: http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=29cd8bae396583a2ee9a3340db8c5102acf9f6fd
> > References	: http://marc.info/?l=linux-kernel&m=125508007510274&w=4
> > Handled-By	: Peter Zijlstra <a.p.zijlstra-/NLkJaSkS4VmR6Xm/wNWPw@public.gmane.org>
On core2 arch machines, hackbench regression disappears and there is much
improvement instead of regression. 
On Nehalem machine, no big change, comparing with 2.6.31.

On Itanium machines (2 sockets or 4 sockets), the regression become
about 20%. Originally it's 70%.


> 
> 
> > Bug-Entry       : http://bugzilla.kernel.org/show_bug.cgi?id=14384
> > Subject         : tbench regression with 2.6.32-rc1
> > Submitter       : Zhang, Yanmin <yanmin_zhang-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>
> > Date            : 2009-10-09 9:51 (39 days old)
> > First-Bad-Commit: http://git.kernel.org/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commit;h=59abf02644c45f1591e1374ee7bb45dc757fcb88
> > References      : http://marc.info/?l=linux-kernel&m=125508216713138&w=4
> > Handled-By      : Peter Zijlstra <a.p.zijlstra-/NLkJaSkS4VmR6Xm/wNWPw@public.gmane.org>

On core2 arch machines, tbench regression becomes about 4%. Originally, the
regression is about 33%.

On Nehalem, tbench regression is about 4%. Original is 7%.

On Itanium, tbench regression is about 16%. Original is 26%


> 
> 
> 
> > Bug-Entry       : http://bugzilla.kernel.org/show_bug.cgi?id=14621
> > Subject         : specjbb2005 and aim7 regression with 2.6.32-rc kernels
> > Submitter       : Zhang, Yanmin <yanmin_zhang-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>
> > Date            : 2009-11-06 7:38 (11 days old)
> > References      : http://marc.info/?l=linux-kernel&m=125749310413174&w=4
specjbb2005 and aim7 results almost have no variation.

> 
> 
> Yanmin, could you please update me on the status of these regressions?
> 
> Mike seems to have done a lot to address issues while I was out, and
> while I (hopefully) did read all resulting email, I must admit to
> loosing track of where we stand.

Mike's patch 1b9508f6831e10 could improve netperf loopback testing.
The latest upstream doesn't merge it yet.