| View previous topic :: View next topic |
| Author |
Message |
H-Dragon
Guru
Joined: 22 Oct 2002
Posts: 547
|
 Posted: Thu May 24, 2007 8:13 pm Post subject: |
 |
|
powertop 1.3 is out with more tips.
i highly recommend the skunk-sources... unfortunately those do not (yet?) include suspend2 and the thinkpad-extras.
however those (2.6.2.2-rc1) sources did lower my wake-ups from ~150 to ~50 !!!!!
thats quite something I say...
_________________
WEBSITE
Do not meddle with the affairs of Dragons, for you are crunchy and tasty with ketchup! |
|
| Back to top |
|
 |
jdmulloy
Tux's lil' helper
Joined: 24 Dec 2004
Posts: 139
Location: Massachusetts, USA
|
| Posted: Fri May 25, 2007 6:20 am Post subject: |
|
|
The hardmask was finally lifted from suspend2-sources-2.6.20 and 2.6.21 so now I'm happily using 2.6.21 and powertop. I now have a tickless kernel and the results from power top are decent. When idle I get C3 80%-95% of the time.[/code]
_________________
Joe Mulloy | http://twitter.com/jdmulloy | Ron Paul in 2012! | 5-1-07 | Unban Playfool | Fire your "Too big to fail" bank http://moveyourmoney.info |
|
| Back to top |
|
|
Tlaloc
Guru
Joined: 21 Feb 2005
Posts: 451
Location: Europe - Alps - Tyrol
|
| Posted: Sun May 27, 2007 8:59 am Post subject: |
|
|
Yay, found the main perpetrator for firing interrupts, it was bluetooth connected via USB after all what showed up as hci_usb with 100 wakeups. Powertop 1.3 lead the way with its hints and the suggestion to execute
| Code: | | hciconfig hci0 down |
/etc/init.d/bluetooth stop wasn't enough...
Now I have to find a way to execute that command the best way... I wish qingy had some means to execute arbitrary commands since I start my different xorg-configurations (single/dual head with single head normally on battery) with it.
Bye, Tlaloc. |
|
| Back to top |
|
|
niskel
Guru
Joined: 17 Nov 2004
Posts: 457
Location: Fredericton, NB, Canada
|
| Posted: Sun May 27, 2007 11:59 pm Post subject: |
|
|
I must be doing something wrong. I don't run tickless yet because I am running x86_64. Even considering I am not running tickless, my results look absolutely terrible.
| Code: |
Cn Avg residency (5s) Long term residency avg
C0 (cpu running) ( 3.7%)
C1 0.0ms ( 0.0%) 0.0ms
C2 0.5ms (96.3%) 0.5ms
Wakeups-from-idle per second : 3894.4
Top causes for wakeups:
68.7% (422.0) <interrupt> : acpi
13.3% (81.6) <interrupt> : ipw3945, nvidia
8.1% (50.0) runscript.sh : queue_delayed_work_on (delayed_work_timer_
2.3% (14.2) kded : schedule_timeout (process_timeout)
1.5% ( 9.2) <interrupt> : libata
1.3% ( 8.2) kicker : schedule_timeout (process_timeout)
1.3% ( 8.0) <kernel core> : usb_hcd_poll_rh_status (rh_timer_func)
1.1% ( 7.0) : do_setitimer (it_real_fn)
0.6% ( 3.4) konsole : schedule_timeout (process_timeout)
0.3% ( 1.6) wpa_supplicant : schedule_timeout (process_timeout)
Suggestion: Enable the CONFIG_NO_HZ kernel configuration option.
This option is required to get any kind of longer sleep times in the CPU.
|
The items that concern me the most are:
- 68.7% (422.0) <interrupt> : acpi
- 8.1% (50.0) runscript.sh : queue_delayed_work_on (delayed_work_timer_
- 2.3% (14.2) kded : schedule_timeout (process_timeout)
- 1.5% ( 9.2) <interrupt> : libata
acpi alone has a higher WPS than most peoples entire system. And I have no idea what the deal is with runscript.sh, it doesn't even show up in the running process list for me. I don't know what kded is other than it is associated with KDE. Last, I don't know why libata is always listed; I am never doing much disc access or writing, is there a laptop mode or something I can put my disc in? |
|
| Back to top |
|
|
mariux2
Apprentice
Joined: 16 Sep 2003
Posts: 242
|
| Posted: Tue May 29, 2007 6:10 am Post subject: |
|
|
| Im unable to get anything else than C0, what gives? Centrino, ondemand.... |
|
| Back to top |
|
|
depontius
Veteran
Joined: 05 May 2004
Posts: 1925
|
| Posted: Tue May 29, 2007 11:51 am Post subject: |
|
|
| niskel wrote: | | acpi alone has a higher WPS than most peoples entire system. |
Is it possible that "acpi" is what gets the credit for the timer tick? After all, you're not running a tickless kernel, and the timer tick wake-ups ought to show somewhere.
_________________
.sigs waste space and bandwidth |
|
| Back to top |
|
|
zidour
Tux's lil' helper
Joined: 29 Apr 2004
Posts: 101
|
| Posted: Tue May 29, 2007 10:35 pm Post subject: |
|
|
| niskel wrote: |
Top causes for wakeups:
68.7% (422.0) <interrupt> : acpi
13.3% (81.6) <interrupt> : ipw3945, nvidia
8.1% (50.0) runscript.sh : queue_delayed_work_on (delayed_work_timer_
2.3% (14.2) kded : schedule_timeout (process_timeout)
1.5% ( 9.2) <interrupt> : libata
1.3% ( 8.2) kicker : schedule_timeout (process_timeout)
1.3% ( 8.0) <kernel core> : usb_hcd_poll_rh_status (rh_timer_func)
1.1% ( 7.0) : do_setitimer (it_real_fn)
0.6% ( 3.4) konsole : schedule_timeout (process_timeout)
0.3% ( 1.6) wpa_supplicant : schedule_timeout (process_timeout)
Suggestion: Enable the CONFIG_NO_HZ kernel configuration option.
This option is required to get any kind of longer sleep times in the CPU.
[/code]
The items that concern me the most are:
- 68.7% (422.0) <interrupt> : acpi
- 8.1% (50.0) runscript.sh : queue_delayed_work_on (delayed_work_timer_
- 2.3% (14.2) kded : schedule_timeout (process_timeout)
- 1.5% ( 9.2) <interrupt> : libata
acpi alone has a higher WPS than most peoples entire system. And I have no idea what the deal is with runscript.sh, it doesn't even show up in the running process list for me. I don't know what kded is other than it is associated with KDE. Last, I don't know why libata is always listed; I am never doing much disc access or writing, is there a laptop mode or something I can put my disc in? |
As for runscript.sh, have a look at my post here |
|
| Back to top |
|
|
Ibn al-Hazardous
Tux's lil' helper
Joined: 02 Sep 2004
Posts: 133
Location: Somewhere deep in the desert.
|
| Posted: Wed Aug 01, 2007 9:17 am Post subject: |
|
|
| depontius wrote: | | niskel wrote: | | acpi alone has a higher WPS than most peoples entire system. |
Is it possible that "acpi" is what gets the credit for the timer tick? After all, you're not running a tickless kernel, and the timer tick wake-ups ought to show somewhere. |
I'm having a similar problem, only the acpi wakeups are more than the timerticks should be:
| Code: | Cn Avg residency (10s) P-states (frequencies)
C0 (cpu running) (18.1%)
C1 0.0ms ( 0.0%) 2.00 Ghz 0.2%
C2 0.5ms (30.6%) 1.84 Ghz 0.0%
C3 0.3ms (51.4%) 1.67 Ghz 0.0%
1000 Mhz 99.8%
Wakeups-from-idle per second : 2208.9
no ACPI power usage estimate available
Top causes for wakeups:
58.2% (1567.4) <interrupt> : acpi
32.9% (886.7) <interrupt> : extra timer interrupt
1.9% ( 50.8) pbbuttonsd : schedule_timeout (process_timeout)
1.9% ( 50.0) runscript.sh : queue_delayed_work_on (delayed_work_timer_fn
1.1% ( 28.5) firefox-bin : futex_wait (hrtimer_wakeup)
0.7% ( 19.8) pbbuttonsd : input_open_polled_device (delayed_work_timer_fn)
Suggestion: Enable the CONFIG_NO_HZ kernel configuration option.
This option is required to get any kind of longer sleep times in the CPU. |
There should be 1000 ticks per sec, as I too have x86_64 with no tickless available. I believe that is the second line. But acpi is waking up the cpu 1.5 times as much - in addition to the timer interrupt.
Small wonder that linux drains the batteries in half the time macos needs (this is a macbook). Any ideas?
Edit:
I configured out the serial port and PS2 mouse (which I don't have on this hardware), and I applied the hpet_patches_45.patch that is linked on the intel site. Now acpi only comes up when I type (ie right now) - so it seems to be serial controllers other than usb.
Edit 2:
Applying the dyntick-patches was even better (when I got it working - dyntick and hpet are not possible to combine). I've now gotten it down to ~200 wakeups/sec. So, now I'm in the ballpark for more ordinary efficiency hunting. 
_________________
/Ibn |
|
| Back to top |
|
|
Clete2
Guru
Joined: 09 Aug 2003
Posts: 527
Location: Bloomington, Illinois
|
| Posted: Wed Aug 01, 2007 3:00 pm Post subject: |
|
|
How do I set NO_HZ? There is no option for it in the kernel.
_________________
My Blog |
|
| Back to top |
|
|
Antti Alien
n00b
Joined: 28 Nov 2003
Posts: 74
Location: Tampere, Finland
|
| Posted: Wed Aug 01, 2007 3:23 pm Post subject: |
|
|
| Clete2 wrote: | | How do I set NO_HZ? There is no option for it in the kernel. |
Dynticks or NO_HZ came with 2.6.21. So update your sources.
In case you're just not seeing it:
| Code: |
Processor type and features --->
[*] Tickless System (Dynamic Ticks)
|
|
|
| Back to top |
|
|
barophobia
Apprentice
Joined: 27 Apr 2004
Posts: 224
Location: somewhere
|
| Posted: Wed Aug 01, 2007 4:31 pm Post subject: |
|
|
| Antti Alien wrote: |
Dynticks or NO_HZ came with 2.6.21. So update your sources.
|
Unless you are using 64bit then wait until 2.6.23. |
|
| Back to top |
|
|
Clete2
Guru
Joined: 09 Aug 2003
Posts: 527
Location: Bloomington, Illinois
|
| Posted: Wed Aug 01, 2007 6:52 pm Post subject: |
|
|
| barophobia wrote: | | Antti Alien wrote: |
Dynticks or NO_HZ came with 2.6.21. So update your sources.
|
Unless you are using 64bit then wait until 2.6.23. |
Yeah, I'm using 64bit.
What exactly does a tickless system do?
2.6.22 is the latest suspend2-sources in portage at the moment. Is suspend2 maintained often?
_________________
My Blog |
|
| Back to top |
|
|
Etal
Veteran
Joined: 15 Jul 2005
Posts: 1580
|
| Posted: Wed Aug 01, 2007 7:06 pm Post subject: |
|
|
| Clete2 wrote: |
What exactly does a tickless system do?
|
Tickless system prevents the kernel from waking up the processor at regular intervals for no reason, letting it stay in a lower C-state longer, saving power.
_________________
“And even in authoritarian countries, information networks are helping people discover new facts and making governments more accountable.”– Hillary Clinton, Jan. 21, 2010 |
|
| Back to top |
|
|
Evincar
Apprentice
Joined: 13 Feb 2007
Posts: 217
Location: Madrid
|
| Posted: Wed Aug 01, 2007 7:48 pm Post subject: |
|
|
| Ibn al-Hazardous wrote: | | depontius wrote: | | niskel wrote: | | acpi alone has a higher WPS than most peoples entire system. |
Is it possible that "acpi" is what gets the credit for the timer tick? After all, you're not running a tickless kernel, and the timer tick wake-ups ought to show somewhere. |
I'm having a similar problem, only the acpi wakeups are more than the timerticks should be:
| Code: | Cn Avg residency (10s) P-states (frequencies)
C0 (cpu running) (18.1%)
C1 0.0ms ( 0.0%) 2.00 Ghz 0.2%
C2 0.5ms (30.6%) 1.84 Ghz 0.0%
C3 0.3ms (51.4%) 1.67 Ghz 0.0%
1000 Mhz 99.8%
Wakeups-from-idle per second : 2208.9
no ACPI power usage estimate available
Top causes for wakeups:
58.2% (1567.4) <interrupt> : acpi
32.9% (886.7) <interrupt> : extra timer interrupt
1.9% ( 50.8) pbbuttonsd : schedule_timeout (process_timeout)
1.9% ( 50.0) runscript.sh : queue_delayed_work_on (delayed_work_timer_fn
1.1% ( 28.5) firefox-bin : futex_wait (hrtimer_wakeup)
0.7% ( 19.8) pbbuttonsd : input_open_polled_device (delayed_work_timer_fn)
Suggestion: Enable the CONFIG_NO_HZ kernel configuration option.
This option is required to get any kind of longer sleep times in the CPU. |
There should be 1000 ticks per sec, as I too have x86_64 with no tickless available. I believe that is the second line. But acpi is waking up the cpu 1.5 times as much - in addition to the timer interrupt.
Small wonder that linux drains the batteries in half the time macos needs (this is a macbook). Any ideas?
Edit:
I configured out the serial port and PS2 mouse (which I don't have on this hardware), and I applied the hpet_patches_45.patch that is linked on the intel site. Now acpi only comes up when I type (ie right now) - so it seems to be serial controllers other than usb.
Edit 2:
Applying the dyntick-patches was even better (when I got it working - dyntick and hpet are not possible to combine). I've now gotten it down to ~200 wakeups/sec. So, now I'm in the ballpark for more ordinary efficiency hunting. |
I KNOW acpi sometimes gets credit for some obscure interrupts, happens here as well. No matter what I am doing, PowerTOP points ~50% of the ticks to acpi.
I wonder: After all the patches, is your battery life still half of OS X?? The extra usage of the CPU doesn't seem to justify the change. What could be the problem, unused devices powered up, perhaps? It's certainly not happening here, but those "Linux eats my battery with popcorn" stories have always mistified me.
_________________
<@Chin^> My sister caught me jacking off the other week and calls me a pervert
<@Chin^> just the other day i walked into my room and caught my sister masturbating
<@Chin^> So she calls me a pervert again?!?
<@Chin^> there is no justice in the world... |
|
| Back to top |
|
|
kernelOfTruth
Watchman
Joined: 20 Dec 2005
Posts: 5345
Location: Vienna, Austria; Germany; hello world :)
|
|
| Back to top |
|
|
antonlacon
Apprentice
Joined: 27 Jun 2004
Posts: 223
|
| Posted: Wed Aug 01, 2007 8:54 pm Post subject: |
|
|
| barophobia wrote: | | Antti Alien wrote: |
Dynticks or NO_HZ came with 2.6.21. So update your sources.
|
Unless you are using 64bit then wait until 2.6.23. |
2.6.24. http://kerneltrap.org/node/11751 |
|
| Back to top |
|
|
Ibn al-Hazardous
Tux's lil' helper
Joined: 02 Sep 2004
Posts: 133
Location: Somewhere deep in the desert.
|
| Posted: Thu Aug 02, 2007 7:02 am Post subject: |
|
|
| Evincar wrote: | | Ibn al-Hazardous wrote: |
------8<-----
Edit:
I configured out the serial port and PS2 mouse (which I don't have on this hardware), and I applied the hpet_patches_45.patch that is linked on the intel site. Now acpi only comes up when I type (ie right now) - so it seems to be serial controllers other than usb.
Edit 2:
Applying the dyntick-patches was even better (when I got it working - dyntick and hpet are not possible to combine). I've now gotten it down to ~200 wakeups/sec. So, now I'm in the ballpark for more ordinary efficiency hunting. |
I KNOW acpi sometimes gets credit for some obscure interrupts, happens here as well. No matter what I am doing, PowerTOP points ~50% of the ticks to acpi.
I wonder: After all the patches, is your battery life still half of OS X?? The extra usage of the CPU doesn't seem to justify the change. What could be the problem, unused devices powered up, perhaps? It's certainly not happening here, but those "Linux eats my battery with popcorn" stories have always mistified me. |
With the hpet patch, I got a bit over two hours. With the dyntic patch I get 3.5 hours - so it's much better now (oh, and that was while keeping a connection open to the net with my 3g-phone).
_________________
/Ibn |
|
| Back to top |
|
|
|
Off the Wall
