Now we have high res timers there is less of a reason for a high HZ value.
Furthermore I think there a few reasons we should reduce HZ to 100:
- Timer interrupt overhead. While this overhead is small, there are
applications that are very sensitive to jitter (eg some HPC apps).
- Issues with the timer wheel code. When coming out of NO_HZ idle we work our
way through the timer code one tick at a time. If we have been idle a long
time, this adds up - I sometimes see milliseconds of time spent in that
loop.
Long term we should fix the timer wheel algorithm, but for now if we reduce
HZ then we reduce the amount of work the timer code has to do when coming
out of idle.
Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
CONFIG_NO_HZ=y
CONFIG_HIGH_RES_TIMERS=y
CONFIG_GENERIC_CLOCKEVENTS_BUILD=y
-# CONFIG_HZ_100 is not set
-CONFIG_HZ_250=y
+CONFIG_HZ_100=y
+# CONFIG_HZ_250 is not set
# CONFIG_HZ_300 is not set
# CONFIG_HZ_1000 is not set
-CONFIG_HZ=250
+CONFIG_HZ=100
CONFIG_SCHED_HRTICK=y
CONFIG_PREEMPT_NONE=y
# CONFIG_PREEMPT_VOLUNTARY is not set
CONFIG_NO_HZ=y
CONFIG_HIGH_RES_TIMERS=y
CONFIG_GENERIC_CLOCKEVENTS_BUILD=y
-# CONFIG_HZ_100 is not set
-CONFIG_HZ_250=y
+CONFIG_HZ_100=y
+# CONFIG_HZ_250 is not set
# CONFIG_HZ_300 is not set
# CONFIG_HZ_1000 is not set
-CONFIG_HZ=250
+CONFIG_HZ=100
CONFIG_SCHED_HRTICK=y
CONFIG_PREEMPT_NONE=y
# CONFIG_PREEMPT_VOLUNTARY is not set