[LUAU] Re: Fedora & mythtv cpu load

[Eric Hattemer]

Whoever Whatever wrote:

>
>No, it wasn't KDE, I recompiled the kernel with athlon, then ended up
>recompling ivtv, lirc and other needed modules, reinstalled nvidia
>driver, system came up fast but with no sound.  It was pretty good
>with very low load average, video not jerky while compiling. I have to
>recompile alsa 1.09 to get the sound working again on via82xx, looks
>like kernel/OSS driver not good for my board.  I was back to that
>heavy load average with sound working again, I guess I nailed down the
>cause, most likely the sound modules, I will try the bleeding version
>later see if it's better.
>
>  
>
It sounds like it has a lot to do with the kernel and ivtv versions 
then.  The alsa that comes with 2.6.11.11 will probably work, but may 
require a little bit of .asoundrc configuration (ie. black magic). 

>I already have frontend running nice -15, do you mean options "nice
>-15" or nice level of "-15" ?
>  
>
let's call it "nice -n -15" to be perfectly clear.  But honestly, I 
don't know which one is best.  Its kind of a question of whether X, 
mythfrontend, or mythbackend has too high of a priority, the others 
might lag and skip.  If you record, then watch later, certainly the 
mythbackend should have a very important priority.  If you watch in 
realtime, the mythbackend might need a bad priority and the frontend and 
X an important priority.  The should probably all have a negative nice 
value so that default processes don't mess them up. 

>Do I need to bleed for ionice? I remember back in years ago, I was
>chasing kernel release on 1.3.x tree to get ip masq working on my
>33.6k dial up, stop doing that since the lap over of glib, remember
>new lib on kernel 2.x.x ?
>
>  
>
Like I said, I never got around to ionice, but it looks like a great 
(and well overdue) concept.  It looked like 2.6.11.11 had an option for 
ionice, and I compiled it in, but I don't see it now, so maybe not.  I 
don't have an ionice binary, so I assume that this version is built into 
the nice syscall.  Maybe I just forgot what I saw.  It might be related 
to cfq.  Anyway, look at http://kerneltrap.org/node/4406/print .  You 
might consider changing the io scheduler http://kerneltrap.org/node/3851 . 

>
>
>I was compiling since .11 running on the good old Mandrake, I am just
>tired of compiling and try to find a easy way out, well.. there's
>another reason, .18 starting to use qt3.3 on some plugins, so I am too
>lazy to upgrade from qt3.1 to qt3.3.  I will recompile it again to
>compare with the atrpms, but I will try to solve the high load average
>cause by sound modules first.
>
>  
>
Should be able to upgrade qt using yum, right?  Should only take a few 
minutes, I'd figure.  Then you can configure the mythtv sound the way 
you want it and feel comfortable that nothing has to do with some weird 
way that the atrpms stuff was compiled. 

>>What bitrate are your recordings?
>>
>>    
>>
>
>720x480 bitrate 4.5k max 6k, I did try 3k.. same problem, but
>shouldn't matter much to the cpu on hardware encoder right?
>
>  
>
That's pretty hardcore.  You're right that it won't matter on the cpu, 
but it will matter on the disk.  All my normal recordings are 2200, and 
my cartoons and talk shows are 800 or so.  If you check top, or gkrellm, 
is your cpu usage iowait or user?  You might just be overloading the 
hard drive.  The main problem I always had with the compiling was that 
even at nice -19, it still brutalizes the hard drive, and things start 
to skip.  That's why I was always excited about ionice.  However, he 
specifically says that its a "proof of concept", so I suspect its not 
production ready yet. 

-Eric Hattemer