]> www.infradead.org Git - users/jedix/linux-maple.git/commitdiff
sched: Disable default sched_autogroup to avoid the DBA performance regression
authorSantosh Shilimkar <santosh.shilimkar@oracle.com>
Mon, 9 Feb 2015 23:08:30 +0000 (15:08 -0800)
committerSantosh Shilimkar <santosh.shilimkar@oracle.com>
Mon, 29 Jun 2015 15:36:55 +0000 (08:36 -0700)
SCHED_AUTOGROUP optimizes the scheduler for common desktop workloads by
automatically creating and populating task groups. Though it helps desktop CPU
hungry workloads(linke build jobs), we found that it crteates 10% regerssion
on DBA perfromance.

Swingbench benchmark run for OLTP shows below:

@ UEK4-with-schedauto 3.18.4-5 7073
@ UEK4-without-schedauto 3.18.4-5 7873

So to have best of both words, we make the SCHED_AUTOGROUP feature
available on UEK kernels but the default state is disabled.

One can enable it using the sysctrl (kernel.sched_autogroup_enabled)

Orabug: 20476603

Tested-by: Thomas Tanaka <thomas.tanaka@oracle.com>
Signed-off-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
kernel/sched/auto_group.c

index eae160dd669d9d8d58bb911595c6391b2732edb4..6072b8c8eef02d24999ac7336ff1ee43f40894bf 100644 (file)
@@ -9,7 +9,7 @@
 #include <linux/security.h>
 #include <linux/export.h>
 
-unsigned int __read_mostly sysctl_sched_autogroup_enabled = 1;
+unsigned int __read_mostly sysctl_sched_autogroup_enabled = 0;
 static struct autogroup autogroup_default;
 static atomic_t autogroup_seq_nr;