From: Nick Piggin <nickpiggin@yahoo.com.au>
Don't balance on clone by default.
Balance on clone has a number of trivial performance failure cases, but it was
needed to get decent OpenMP performance on NUMA (Opteron) systems. Not doing
child-runs-first for new threads also solves this problem in a nicer way
(implemented in a previous patch).
Signed-off-by: Nick Piggin <nickpiggin@yahoo.com.au>
Signed-off-by: Andrew Morton <akpm@osdl.org>
---
25-akpm/include/linux/sched.h | 3 ---
1 files changed, 3 deletions(-)
diff -puN include/linux/sched.h~sched-no-balance-clone include/linux/sched.h
--- 25/include/linux/sched.h~sched-no-balance-clone 2004-06-30 00:06:58.258948864 -0700
+++ 25-akpm/include/linux/sched.h 2004-06-30 00:06:58.262948256 -0700
@@ -612,7 +612,6 @@ struct sched_domain {
.per_cpu_gain = 15, \
.flags = SD_BALANCE_NEWIDLE \
| SD_BALANCE_EXEC \
- | SD_BALANCE_CLONE \
| SD_WAKE_AFFINE \
| SD_WAKE_IDLE \
| SD_SHARE_CPUPOWER, \
@@ -635,7 +634,6 @@ struct sched_domain {
.per_cpu_gain = 100, \
.flags = SD_BALANCE_NEWIDLE \
| SD_BALANCE_EXEC \
- | SD_BALANCE_CLONE \
| SD_WAKE_AFFINE \
| SD_WAKE_BALANCE, \
.last_balance = jiffies, \
@@ -657,7 +655,6 @@ struct sched_domain {
.cache_nice_tries = 1, \
.per_cpu_gain = 100, \
.flags = SD_BALANCE_EXEC \
- | SD_BALANCE_CLONE \
| SD_WAKE_BALANCE, \
.last_balance = jiffies, \
.balance_interval = 1, \
_