[mythtv-users] Mythfrontend idle cpu consumption help

[Jim Stichnoth]

On Tue, Jan 26, 2010 at 10:13 AM, jansenj <jansenj+myth at gmail.com> wrote:
>
>> Would it be possible for you to try out the patch in ticket #7953?  I
>> know that QTimer should be responsible for up to 1ms of spin-waiting
>> per signal, but you (and everyone else in the 14-15% camp) are
>> spending 2ms of CPU time per signal.  I'd like to know if your idle
>> frontend CPU usage drops all the way to 1-2%, or only goes down to
>> 7-9%.
>
> The patch got mythfrontend down to spending most of the time at 0.7% usage,
> sometimes up to 1.3% while sitting idle on the main menu screen.

Great!  That matches what I see.

> If I read your patch correctly, you are just sleeping for a time equivalent
> to one period.  Its important to note that this makes your "real" animation
> pulse closer to 69Hz or so.  I don't know what was magical about 70Hz to
> begin with other than to have it be a faster sample rate than most screen
> refreshes.

The sleep is in a separate thread, so it probably is pretty close to
70Hz, which as you observe is a fairly arbitrary value.

> As some have said, the next step would be to connect it up to the screen
> refresh, or opengl vsync.  But your patch is a very good start, and would
> contend that something similar could be incorporated into the 0.22-fixes
> branch.  A guy can wish can't he?

Daniel committed a better version of the patch to trunk.  I highly
doubt it will go into 0.22-fixes given the current emphasis on getting
0.23 out, along with the fact that it's extremely hard to observe the
problem without "top" or "ps".

Jim