[PLUG] Hot computer

[Daniel Hedlund]

On Thu, Dec 9, 2010 at 17:24, John Jason Jordan <johnxj at comcast.net> wrote:
> They all reported nothing, except "unhide sys" found a long list of
> hidden processes. "Aha!" I thought. So I checked them with top (maximum
> 20 at a time). All looked perfectly normal and only two or three were
> consuming any CPU% at all, and they were taking only 1-3%. The only
> oddity was that several had more than one instance of the process
> running, and one (console-kit-dae) had around 60 instances running. But
> none of them were taking any CPU or RAM.


I believe in an earlier email you suggested that you experimented a
little with powertop but did not specify whether you understood any of
its output; it might be time to revisit the program.

Beyond a 'top'-like interface, powertop can show a summary of data
collected over a period of time with the '-d' option.  You should be
able to get a list of processes that are waking up the CPU most often,
what percentage of the time the CPU is at each frequency and the
percentage the CPU is in a given state.  While a program might not be
using 100% of the CPU, it might be interrupting the CPU often enough
to prevent it from sleeping.  Try running powertop for a full 60
seconds with the following command:
$ powertop -d -t 60

Information about processor states can be found at:
http://en.wikipedia.org/wiki/Advanced_Configuration_and_Power_Interface#Processor_states

Here's a partial dump from my laptop:

Collecting data for 15 seconds

Cn	          Avg residency
C0 (cpu running)        ( 3.7%)
polling		  0.0ms ( 0.0%)
C1 mwait	  0.0ms ( 0.0%)
C2 mwait	  0.3ms ( 0.9%)
C4 mwait	  3.2ms (95.4%)
P-states (frequencies)
  2.51 Ghz     0.1%
  2.50 Ghz     0.0%
  2.00 Ghz     0.0%
  1.60 Ghz     0.0%
   800 Mhz    99.9%
Wakeups-from-idle per second : 330.7	interval: 15.0s
no ACPI power usage estimate available
Top causes for wakeups:
  32.0% (127.8)     <kernel core> : hrtimer_start_range_ns (tick_sched_timer)
  26.6% (106.2)      npviewer.bin : hrtimer_start_range_ns (hrtimer_wakeup)
  13.2% ( 52.7)       <interrupt> : extra timer interrupt
...


In the above example, the npviewer.bin (a wrapper around the flash
plugin) is a major offender; It's only consuming 2-3% of CPU but it's
responsible for 26% of wakeups.  What's nice is that it also break out
which parts of the kernel are waking up the CPU which top/'gnome
system monitor' doesn't seem to do.  If you're not sure what your
results mean, feel free to paste it back to the list.  If powertop
really has nothing useful then, as others have started to question, it
might be hardware and/or fan control.


Cheers,

Daniel Hedlund
daniel at digitree.org