panel plugins that cause too many wakeups/s

[Kok, Auke]

Brian J. Tarricone wrote:
> Tino Keitel wrote:
>> Hi,
>>
>> I noticed at some point that the Xfce panel produces a fair number of
>> wakeups/s. So I always kill it when running on battery, and I can see
>> that this gains some battery runtime.
>>
>> I tried to find out what specific panel component is responsible for
>> this. Conclusion: the most wakeups/s are caused by the system load
>> plugin. Here are some numbers:
>>
>> system load plugin: 11

probably hits /proc/ too much

> I tried to fix the system load plugin a few months ago by decreasing the 
> update interval, but even with the interval set to several seconds, I 
> was still seeing several wakeups per second, and could never figure out 
> why.  I had the same issue with the netload plugin.
> 
>> battery monitor plugin: 3
> 
> I believe the current released battery plugin polls for battery status, 
> so there's not much you can do except increase the interval.  The 
> version in Nick's "hal_based" branch in SVN works very well, and doesn't 
> poll, but instead uses HAL which does async notification of changes. 
> I've been using it for many months now without problems, and I never see 
> it show up in powertop at all.

one of the design issues is that the battery plugin hits ACPI 4 or 5 times for
each battery level poll:

1) to check if the system is on battery or AC power
2) to see what batteries are connected
3) to read a batteries charge

the code could remember (1) and (2) and attempt to just do (3) on each poll
eliminating most of the wakeups. On error it could fall back to re-reading (1) and
(2)....

> 
>> Without the above plugins, an idle desktop causes less than 7
>> wakeups/s. This is with a running panel with the weather, clock, and
>> sensors plugin running (and a few other plugins with static content,
>> like the pager, task list, menu, dict, quit button and some launchers).
>> The sensors plugin update interval is 60 seconds.
> 
> Heh, I can't get my laptop under 200 wakeups per second or so, but 
> that's probably just ppc kernel issues and too many (hopefully) useless 
> interrupts...  And firefox.

wakeups counts, but average sleep time also. You can have 1000 wakeups per second
and still get 99% sleep time. So please post all the powertop numbers, not just
wakeups.


Auke