[mythtv] JobQueue Suggestion: Average Load of Backend

[osma ahvenlampi]

On 1/9/07, Blammo <blammo.doh at gmail.com> wrote:
> I'll make this quick and easy. My primary system at home has (1)
> dedicated backend, and two FE/SBE's. All machines have lots of
> horsepower, and help out with commercial flagging. However, given that
> my families watching habits vary from day to day, sometimes there will
> be a couple flagging jobs running, which, for example, someone is
> trying to watch a show. My only recourse, right now, is to block out
> the times during the day, for both frontends, on the CHANCE that
> someone might be watching something.

I think I've implemented what you're asking for. See
http://cvs.mythtv.org/trac/ticket/2782 and test it if it seems
appropriate. You'll need to apply three patches (split to parts to be
easier to manage, against 0.20 but I know of no reason why it
shouldn't apply against SVN). Earlier discussion at
http://www.gossamer-threads.com/lists/mythtv/dev/240442#240442

However, I'm not entirely certain I understand why you "block out ...
on the chance that someone may be watching". I don't see any problems
running up to three background commflag and transcode jobs on a
single-CPU, single-disk backend/frontend while recording two programs
(hence two real-time commflags) and watching something. For me, the
motivation of the patch was to maximise the number of jobs to run
without need of configuration while avoiding potential IO trashing,
and the patch has worked beautifully for that purpose. In fact, the
standard version also works just fine under any circumstance I can
think of for up to one simultaneous job - it just won't necessarily
utilise ALL resources efficiently.

I'm not using loadav in the decision, because it's way too coarse to
be useful. Linux 2.6's addition of "iowait time" to the
user+system+nice+idle scheduler stats however works great for this
purpose.

-- 
Osma Ahvenlampi   <oa at iki.fi>       http://www.fishpool.org