[PATCH] sched/numa: recovered SD_PERFER_SIBLING on CPU domain

[PATCH] sched/numa: recovered SD_PERFER_SIBLING on CPU domain

[Alex Shi]

commit 8e7fbcbc22c(sched: Remove stale power aware scheduling remnants
and dysfunctional knobs) removed SD_PERFER_SIBLING on CPU domain
On numa machine, that causes load_balance didn't perfer LCPU in same
physical CPU package.

It causes some regression on our numa machines from core2 to NHM and SNB.
Recover this domain flag can recover the performance drop.

This recovery doesn't has any bad impact on my all benchmarks specjbb,
kbuild, fio, hackbench .. etc, on all my machines.
Since it is a recovery, I assume it also has no bad impact for other
platforms.

Signed-off-by: Alex Shi <alex.shi@intel.com>
---
 include/linux/topology.h |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/include/linux/topology.h b/include/linux/topology.h
index e91cd43..fec12d6 100644
--- a/include/linux/topology.h
+++ b/include/linux/topology.h
@@ -164,6 +164,7 @@ int arch_update_cpu_topology(void);
 				| 0*SD_SHARE_CPUPOWER			\
 				| 0*SD_SHARE_PKG_RESOURCES		\
 				| 0*SD_SERIALIZE			\
+				| 1*SD_PREFER_SIBLING			\
 				,					\
 	.last_balance		= jiffies,				\
 	.balance_interval	= 1,					\
-- 
1.7.5.4

Re: [PATCH] sched/numa: recovered SD_PERFER_SIBLING on CPU domain

[Peter Zijlstra]

On Fri, 2012-07-20 at 14:19 +0800, Alex Shi wrote:
> commit 8e7fbcbc22c(sched: Remove stale power aware scheduling remnants
> and dysfunctional knobs) removed SD_PERFER_SIBLING on CPU domain
> On numa machine, that causes load_balance didn't perfer LCPU in same
> physical CPU package.
> 
> It causes some regression on our numa machines from core2 to NHM and SNB.
> Recover this domain flag can recover the performance drop.
> 
> This recovery doesn't has any bad impact on my all benchmarks specjbb,
> kbuild, fio, hackbench .. etc, on all my machines.
> Since it is a recovery, I assume it also has no bad impact for other
> platforms.
> 
> Signed-off-by: Alex Shi <alex.shi@intel.com>

Thanks!

[tip:sched/urgent] sched/numa: Add SD_PERFER_SIBLING to CPU domain

[tip-bot for Alex Shi]

Commit-ID:  6956dc568f34107f1d02b24f87efe7250803fc87
Gitweb:     http://git.kernel.org/tip/6956dc568f34107f1d02b24f87efe7250803fc87
Author:     Alex Shi <alex.shi@intel.com>
AuthorDate: Fri, 20 Jul 2012 14:19:50 +0800
Committer:  Ingo Molnar <mingo@kernel.org>
CommitDate: Thu, 26 Jul 2012 11:46:58 +0200

sched/numa: Add SD_PERFER_SIBLING to CPU domain

Commit 8e7fbcbc22c ("sched: Remove stale power aware scheduling remnants
and dysfunctional knobs") removed SD_PERFER_SIBLING from the CPU domain.

On NUMA machines this causes that load_balance() doesn't perfer LCPU in
 same physical CPU package.

It causes some actual performance regressions on our NUMA machines from
Core2 to NHM and SNB.

Adding this domain flag again recovers the performance drop.

This change doesn't have any bad impact on any of my benchmarks:
 specjbb, kbuild, fio, hackbench .. etc, on all my machines.

Signed-off-by: Alex Shi <alex.shi@intel.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Link: http://lkml.kernel.org/r/1342765190-21540-1-git-send-email-alex.shi@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
---
 include/linux/topology.h |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/include/linux/topology.h b/include/linux/topology.h
index e91cd43..fec12d6 100644
--- a/include/linux/topology.h
+++ b/include/linux/topology.h
@@ -164,6 +164,7 @@ int arch_update_cpu_topology(void);
 				| 0*SD_SHARE_CPUPOWER			\
 				| 0*SD_SHARE_PKG_RESOURCES		\
 				| 0*SD_SERIALIZE			\
+				| 1*SD_PREFER_SIBLING			\
 				,					\
 	.last_balance		= jiffies,				\
 	.balance_interval	= 1,					\