[opensuse-kernel] nohz bad for powerconsumption ?
I'm using the new default kernel (from kerneltest) with nohz enabled and noticed the following with powertop on an idle system: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ PowerTOP version 1.7 (C) 2007 Intel Corporation Cn Avg residency (3s) P-states (frequencies) C0 (cpu running) ( 2.0%) C1 0.0ms ( 0.0%) 1400 Mhz 33.0% C2 7.4ms (48.8%) 1300 Mhz 0.0% C3 7.6ms (30.3%) 1200 Mhz 0.0% C4 4.7ms (18.9%) 600 Mhz 67.0% Wakeups-from-idle per second : 145.7 Power usage (ACPI estimate): 12.1W (0.9 hours) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ Booting the same kernel with nohz=off, I get: ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ PowerTOP version 1.7 (C) 2007 Intel Corporation Cn Avg residency (10s) P-states (frequencies) C0 (cpu running) ( 5.4%) C1 0.0ms ( 0.0%) 1400 Mhz 0.0% C2 2.9ms (42.0%) 1300 Mhz 0.0% C3 3.3ms (14.2%) 1200 Mhz 0.0% C4 3.2ms (38.5%) 600 Mhz 100.0% Wakeups-from-idle per second : 307.5 Power usage (ACPI estimate): 10.5W (1.0 hours) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ So, for me the nohz kernel uses more power. This seems to be a result From running on an idle system with 33% on high power. Some information about the system: cat /proc/cpuinfo processor : 0 vendor_id : GenuineIntel cpu family : 6 model : 13 model name : Intel(R) Pentium(R) M processor 1.40GHz stepping : 6 cpu MHz : 600.000 cache size : 2048 KB fdiv_bug : no hlt_bug : no f00f_bug : no coma_bug : no fpu : yes fpu_exception : yes cpuid level : 2 wp : yes flags : fpu vme de pse tsc msr mce cx8 apic sep mtrr pge mca cmov pat clflush dts acpi mmx fxsr sse sse2 ss tm pbe up est tm2 bogomips : 1197.25 clflush size : 64 $ cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_governor ondemand Is this worth a bugreport? Anything else I should supply? Andreas -- Andreas Jaeger, Director Platform/openSUSE, aj@suse.de SUSE LINUX Products GmbH, GF: Markus Rex, HRB 16746 (AG Nürnberg) Maxfeldstr. 5, 90409 Nürnberg, Germany GPG fingerprint = 93A3 365E CE47 B889 DF7F FED1 389A 563C C272 A126
[apologies for the duplicated mail; working around the brain dead opensuse list server]
Is this worth a bugreport? Anything else I should supply?
Are you sure the user land setup is identical? nohz tends to come with hr timers so in theory there might be more wakeups if you're unlucky because hr timers will wake up programs at the exact time they request if the time source can deliver it; not always rounded to jiffies (2.5ms) It might be interesting to try with a NOHZ kernel without CONFIG_HIGH_RES_TIMERS -Andi -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Andi Kleen wrote:
[apologies for the duplicated mail; working around the brain dead opensuse list server]
Is this worth a bugreport? Anything else I should supply?
Are you sure the user land setup is identical?
nohz tends to come with hr timers so in theory there might be more wakeups if you're unlucky because hr timers will wake up programs at the exact time they request if the time source can deliver it; not always rounded to jiffies (2.5ms)
It might be interesting to try with a NOHZ kernel without CONFIG_HIGH_RES_TIMERS
I only enabled nohz, not hrtimers in our kernel. - -Jeff - -- Jeff Mahoney SUSE Labs -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.5 (GNU/Linux) Comment: Using GnuPG with SUSE - http://enigmail.mozdev.org iD8DBQFGsNsRLPWxlyuTD7IRAgZmAKCFZ8q9GVPHsMZPhRLnLctpENLmzACdGTxC cIyoI7bYmm9udvCfqEPaIj8= =NbVn -----END PGP SIGNATURE----- -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
On Wednesday 01 August 2007 21:12:17 Jeff Mahoney wrote:
Andi Kleen wrote:
[apologies for the duplicated mail; working around the brain dead opensuse list server]
Is this worth a bugreport? Anything else I should supply?
Are you sure the user land setup is identical?
nohz tends to come with hr timers so in theory there might be more wakeups if you're unlucky because hr timers will wake up programs at the exact time they request if the time source can deliver it; not always rounded to jiffies (2.5ms)
It might be interesting to try with a NOHZ kernel without CONFIG_HIGH_RES_TIMERS
I only enabled nohz, not hrtimers in our kernel.
Then perhaps his userland setup was not identical. -Andi -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Andi Kleen
Is this worth a bugreport? Anything else I should supply?
Are you sure the user land setup is identical?
Yes, it should be the same - I just rebooted the machine.
nohz tends to come with hr timers so in theory there might be more wakeups if you're unlucky because hr timers will wake up programs at the exact time they request if the time source can deliver it; not always rounded to jiffies (2.5ms)
It might be interesting to try with a NOHZ kernel without CONFIG_HIGH_RES_TIMERS
The question is as well what is the cpu frequency scaling doing? Andreas -- Andreas Jaeger, Director Platform/openSUSE, aj@suse.de SUSE LINUX Products GmbH, GF: Markus Rex, HRB 16746 (AG Nürnberg) Maxfeldstr. 5, 90409 Nürnberg, Germany GPG fingerprint = 93A3 365E CE47 B889 DF7F FED1 389A 563C C272 A126
On Wed, 1 Aug 2007, Andreas Jaeger wrote:
Cn Avg residency (3s) P-states (frequencies) C0 (cpu running) ( 2.0%) C1 0.0ms ( 0.0%) 1400 Mhz 33.0% C2 7.4ms (48.8%) 1300 Mhz 0.0% C3 7.6ms (30.3%) 1200 Mhz 0.0% C4 4.7ms (18.9%) 600 Mhz 67.0% Wakeups-from-idle per second : 145.7 Power usage (ACPI estimate): 12.1W (0.9 hours)
vs.
Cn Avg residency (10s) P-states (frequencies) C0 (cpu running) ( 5.4%) C1 0.0ms ( 0.0%) 1400 Mhz 0.0% C2 2.9ms (42.0%) 1300 Mhz 0.0% C3 3.3ms (14.2%) 1200 Mhz 0.0% C4 3.2ms (38.5%) 600 Mhz 100.0% Wakeups-from-idle per second : 307.5 Power usage (ACPI estimate): 10.5W (1.0 hours) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
So, for me the nohz kernel uses more power. This seems to be a result From running on an idle system with 33% on high power.
How often does the ondemand governor sample for busyness? % cat /sys/devices/system/cpu/cpu*/cpufreq/ondemand/sampling_rate (and compare with sampling_rate_max, try playing with echoing different values int sampling_rate). But it's strange nevertheless, as you have less wakeups with nohz, as expected. Are you sure that the system at the time you measured with nohz really was as idle as with nohz=off? No ugly indexer running? (FWIW: my power consumption goes down with nohz) Ciao, Michael. -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org For additional commands, e-mail: opensuse-kernel+help@opensuse.org
Michael Matz
On Wed, 1 Aug 2007, Andreas Jaeger wrote:
Cn Avg residency (3s) P-states (frequencies) C0 (cpu running) ( 2.0%) C1 0.0ms ( 0.0%) 1400 Mhz 33.0% C2 7.4ms (48.8%) 1300 Mhz 0.0% C3 7.6ms (30.3%) 1200 Mhz 0.0% C4 4.7ms (18.9%) 600 Mhz 67.0% Wakeups-from-idle per second : 145.7 Power usage (ACPI estimate): 12.1W (0.9 hours)
vs.
Cn Avg residency (10s) P-states (frequencies) C0 (cpu running) ( 5.4%) C1 0.0ms ( 0.0%) 1400 Mhz 0.0% C2 2.9ms (42.0%) 1300 Mhz 0.0% C3 3.3ms (14.2%) 1200 Mhz 0.0% C4 3.2ms (38.5%) 600 Mhz 100.0% Wakeups-from-idle per second : 307.5 Power usage (ACPI estimate): 10.5W (1.0 hours) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
So, for me the nohz kernel uses more power. This seems to be a result From running on an idle system with 33% on high power.
How often does the ondemand governor sample for busyness? % cat /sys/devices/system/cpu/cpu*/cpufreq/ondemand/sampling_rate
/sys/devices/system/cpu/cpu0/cpufreq/ondemand/sampling_rate 500000 /sys/devices/system/cpu/cpu0/cpufreq/ondemand/sampling_rate_min 250000 cat /sys/devices/system/cpu/cpu0/cpufreq/ondemand/sampling_rate_max 250000000
(and compare with sampling_rate_max, try playing with echoing different values int sampling_rate). But it's strange nevertheless, as you have less wakeups with nohz, as expected. Are you sure that the system at the time you measured with nohz really was as idle as with nohz=off? No ugly indexer running?
(FWIW: my power consumption goes down with nohz)
Yes, it was idle, Andreas -- Andreas Jaeger, Director Platform/openSUSE, aj@suse.de SUSE LINUX Products GmbH, GF: Markus Rex, HRB 16746 (AG Nürnberg) Maxfeldstr. 5, 90409 Nürnberg, Germany GPG fingerprint = 93A3 365E CE47 B889 DF7F FED1 389A 563C C272 A126
participants (4)
-
Andi Kleen
-
Andreas Jaeger
-
Jeff Mahoney
-
Michael Matz