Announcement

Collapse
No announcement yet.

Nailing Down The Linux Kernel Power Regressions

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #16
    if you want to do ocr on lcd 7 segments there is ssocr

    http://www.unix-ag.uni-kl.de/~auerswal/ssocr/index.html

    Comment


    • #17
      Originally posted by Michael View Post
      Too much work with too little ROI.
      Well, if I was not in a remote country without reliable post-office I would set up the system and send it to you. But I disagree on the low ROI. Once you have the system up and running you can use it not only to troubleshot this particular problem, but also make measurement on the other systems and test power efficiency in your reports.
      Using DMM has the added advantage you can also use it to test DC power and isolate sub-systems (Processor power consumption from the 4pins 12V, or graphic card with external power) It would be more accurate than simply measure the total power consumption of the system.
      But thank you for the good work you are making by giving us reliable benchmark; you can only optimize what you can measure!

      Comment


      • #18
        Originally posted by frederir View Post
        A bit over your budget ($140 to $150) but a better solution would be to use two VA18B DMM with usb interface, one to measure voltage, the other to measure current and then combine the two to have power.
        If you go down that path it could be worth looking for multimeters which claim to measure True RMS. Note, multiplying the readings of both multimeters will give you apparent power not real power. Might be good enough though.

        Comment


        • #19
          Originally posted by frief View Post
          If you go down that path it could be worth looking for multimeters which claim to measure True RMS. Note, multiplying the readings of both multimeters will give you apparent power not real power. Might be good enough though.
          Hum, you are totally right, without true RMS the power factor of the power supply under test will taint the results. And another consideration is the dephasage between measure of voltage and current. May be this DMM setup is not a so good idea after all ....

          Comment


          • #20
            with still needing to monitor them for manually toggling the AC power / battery
            With ThinkPads and the "tp_smapi" kernel module you can discharge the battery even with AC connected by writing a "1" to "/sys/devices/platform/smapi/BAT0/force_discharge" (switching back to AC can be done by writing a 0 or automatically when the battery is empty). So your laptop won't power off when you got the AC adapter inserted and the discharging forced.

            This way you could at least automate the "charge/build new kernel/wait until battery has reached >50%/reboot - switch to battery - measure - switch to AC" cycle with accurate (and easy!) measuring.

            More info (with hardware compatibility matrix) is here: http://www.thinkwiki.org/wiki/Tp_smapi
            Last edited by jannis; 06-03-2011, 08:30 AM.

            Comment


            • #21
              Originally posted by glasen View Post
              Just my idea:

              Point a web-cam to this device, do a little OCR magic and you have your numbers.

              BTW Michael:

              I just stumbled across a patch for (soon to arrive) version 2.6.39.1:

              http://thread.gmane.org/gmane.linux....02/focus=12505

              Maybe this small patch solves the power regression.

              I'm currently building a vanilla 2.6.39 kernel with this patch applied. Tomorrow i will compare it with a completely vanilla kernel. Maybe the power consumption is significantly lower with this patch.
              THIS!

              10letters.

              Comment


              • #22
                Originally posted by glasen View Post
                I just stumbled across a patch for (soon to arrive) version 2.6.39.1:

                http://thread.gmane.org/gmane.linux....02/focus=12505

                Maybe this small patch solves the power regression.

                I'm currently building a vanilla 2.6.39 kernel with this patch applied. Tomorrow i will compare it with a completely vanilla kernel. Maybe the power consumption is significantly lower with this patch.
                Nice find. Did it help?

                Comment


                • #23
                  Just a layman observation: the first power regression coincides with the start of ATI KMS power management (2.6.35), so could it be that the default power settings in KMS are even less aggressive than when PM was handled by the xorg driver? This would also explain the fact that not everybody has been able to replicate the regression.

                  On a side note, if linux (kernel/distros) would like to be more "environmental friendly", they would set all default power management settings (CPU/GPU) to the most economic ones. I have had more than enough computing power for my needs for the last decade, if I need more I can always change the settings to more powerful ones. As a laptop owner, I have to change now all settings in order to get more battery life and a cooler computer.

                  Comment


                  • #24
                    Originally posted by not.sure View Post
                    Nice find. Did it help?

                    I had the power issue big time with my battery and this ABSOLUTELY solved my problems. I installed the 2.6.39.1 kernel about 3 hours ago and if my fan has come on I haven't heard it. My temps have fallen drastically. I simply couldn't be more pleased. I really thank the one who posted this link.


                    Robin

                    Comment


                    • #25
                      Originally posted by z06gal View Post
                      I had the power issue big time with my battery and this ABSOLUTELY solved my problems. I installed the 2.6.39.1 kernel about 3 hours ago and if my fan has come on I haven't heard it. My temps have fallen drastically. I simply couldn't be more pleased. I really thank the one who posted this link.


                      Robin
                      Mhm, 2.6.39.1 is slightly better for me as well. But I don't think I'm back at the level of 2.6.37.

                      Comment


                      • #26
                        Originally posted by not.sure View Post
                        Mhm, 2.6.39.1 is slightly better for me as well. But I don't think I'm back at the level of 2.6.37.

                        One thing I have noticed with this kernel is the huge drop in wake ups from idle. It was in the 400 to 500 range on the 2.6.38 kernel and now it is in the 50-100 range. Temps are still low and my battery life is what it should be. I'm just grateful to be able to use my battery again. Lol

                        Comment


                        • #27
                          Measuring Power Usage tools

                          Hi Michael,

                          No one has mentioned this yet, but why not get a Tweet-A-Watt. This is where you take a bog-standard Kill-A-Watt meter and hack it to have a twitter interface. Actually, you hack it to have a wireless XBee interface, then you have a reader on your computer which gets the values.

                          Look at http://www.adafruit.com/category/32, they sell full kits you can use. I haven't had a major need for this myself, but it might be just what you want to use.

                          John

                          Comment

                          Working...
                          X