Bug 680041

Summary: High CPU usage regardless latest kernel versions
Product: [Fedora] Fedora Reporter: Krzysztof "Uosiu" Hajdamowicz <uosiumen>
Component: kernelAssignee: John Feeney <jfeeney>
Status: CLOSED NOTABUG QA Contact: Fedora Extras Quality Assurance <extras-qa>
Severity: high Docs Contact:
Priority: unspecified    
Version: 14CC: bugzilla, gansalmon, itamar, jfeeney, jonathan, kernel-maint, madhu.chinakonda
Target Milestone: ---   
Target Release: ---   
Hardware: x86_64   
OS: Linux   
Whiteboard:
Fixed In Version: Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: 635813 Environment:
Last Closed: 2011-02-28 23:11:20 UTC Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Attachments:
Description Flags
Powertop console output
none
/boot/grub/menu.lst
none
Powertop on idle Laptop
none
Archive screenshot from powertop - Dec 5, 2009 none

Description Krzysztof "Uosiu" Hajdamowicz 2011-02-24 07:39:40 UTC
Created attachment 480677 [details]
Powertop console output

Description of problem:
According to bug https://bugzilla.redhat.com/show_bug.cgi?id=635813
I'm creating new bug report to resolve problem with Load Balancing Tick.

Version-Release number of selected component (if applicable):
kernel-2.6.35.11-83.fc14.x86_64

How reproducible:
Always

Steps to Reproduce:
1. Boot machine
2. Watch the powertop
3.
  
Actual results:
35W power consumption on idle'ing Core2Duo T5250 (2*1,5GHz) with integrated graphics!

Expected results:
~13-15W when idle'ing (easy to acheive on fedora 8-9-10-11-12-13-early 14) with a perspective to acheive low 10W (My personal record is 10,2W with 7h@battery)

Additional info:
Lenovo Thinkpad R61i 8943-DKG with Core2Duo T5250, i965, 3GiB RAM and 640gb/5200RPM HDD and 1024x768 15" LCD/cold cathode backlight
http://www.smolts.org/client/show/pub_155ba98d-3576-4e0a-8940-e209a5675506

Comment 1 Krzysztof "Uosiu" Hajdamowicz 2011-02-24 07:44:48 UTC
Created attachment 480678 [details]
/boot/grub/menu.lst

Comment 2 Matthew Garrett 2011-02-28 19:14:16 UTC
Please repeat using an idle system. You've got ~300 wakeups a second from your web browser and java, which is going to be responsible for a lot of the rebalancing.

Comment 3 Matthew Garrett 2011-02-28 19:15:25 UTC
*** Bug 664027 has been marked as a duplicate of this bug. ***

Comment 4 Matthew Garrett 2011-02-28 19:16:09 UTC
Also, please make sure that there are no wakeups from i8042 (ie, don't touch the keyboard or touchpad)

Comment 5 Krzysztof "Uosiu" Hajdamowicz 2011-02-28 22:54:48 UTC
Created attachment 481465 [details]
Powertop on idle Laptop

23:54 [ root @ samhain ] /opt ROOTMODE> uname -a
Linux samhain 2.6.35.11-83.fc14.x86_64 #1 SMP Mon Feb 7 07:06:44 UTC 2011 x86_64 x86_64 x86_64 GNU/Linux


Completely idle

Comment 6 Krzysztof "Uosiu" Hajdamowicz 2011-02-28 23:00:03 UTC
Created attachment 481467 [details]
Archive screenshot from powertop - Dec 5, 2009

I can't remember Kernel version and Fedora version, but this screenshot was taken on December 5th 2009 with up-do-date system and packages. Unbeaten power consumption score.

Comment 7 Matthew Garrett 2011-02-28 23:11:20 UTC
You've got activity from evince, nautilus, something's hitting the disk and X is asking for interrupts. Otherwise you'd have pretty identical results, so I don't see any sign of a kernel bug here. You may want to file bugs against evince, nautilus, knotify and psi.

Comment 8 Krzysztof "Uosiu" Hajdamowicz 2011-02-28 23:24:16 UTC
Regardless of that activity, Load Balancing Tick causes most CPU wakeups. It's job is causing more job for itself.

Starting boinc-client with milestone RSA bruteforcing and WorldCommunityGrid - Fight AIDS@Home causes:

  86,7% (1721,6)   [kernel scheduler] Load balancing tick
   2,2% ( 43,0)   chromium-browse
   1,8% ( 35,8)   Xorg
   1,0% ( 20,1)   wcg_faah_autodo
   1,0% ( 20,0)   knotify4
   0,9% ( 18,2)   milestone_1.02_


Isn't it illogical to wake up CPU 1700 times/sec (with 2000/sec is a max value achieved on my CPU) to try to balance load caused by two threads, when both are designed to eat 100% of core running on?

In my opinion load balancer should be disabled od throttled way down when cores load is above specified treshold.

Comment 9 Matthew Garrett 2011-02-28 23:34:33 UTC
No, the timeslice is 1ms. You'll enter the scheduler 1000 times a second per CPU assuming a fully loaded system. If your CPU is busy then wakeups cost you nothing.

Comment 10 Krzysztof "Uosiu" Hajdamowicz 2011-02-28 23:50:14 UTC
Maybe costs me nothing, naybe not- a fact is that due to additional CPU work CPU is warmer and I can't pass through 5h on battery barrier whatever I do or I don't.
When kernel hadn't mentioned bug/feature this score was easy to achieve.

Comment 11 Matthew Garrett 2011-03-01 00:17:34 UTC
You're seeing far more userspace wakeups than you were before, which would explain a lot of that.