[mythtv-users] Deadlock was Re: HD thin client?

[Bruce Taber]

Scott D. Davilla wrote:
>> I think we seriously need to re-evaluate the requirements for HD content
>> on Mythtv.
>>
>> I seen in many many documents that 3 Ghz or higher is needed to play
>> back HDTV.  And a year ago, I would definitely agree with that
>> statement. 
>>
>> However, I think the devs have made so many improvements to the code in
>> the last 6 months that this is simply no longer the case.
>>
>>     
>
> I see conflicting performance. With my mythfrontend only 1GHz 
> pentium-m, 256MB ram AppleTV running linux, I can do mpeg2 software 
> decode of 480p (less than 25 percent cpu) and 720p (60 percent cpu). 
> That's great for a less than $250 silent box that includes 40GB hard 
> drive, 10/100BaseT ethernet, 802.11 b/g and n wireless IR w/ 
> controller, HDMI video, component video, analog and optical audio.
>
> 1080i mpeg2 software decode saturates the cpu and I need to do xvmc 
> assisted mpeg2 decode. However, ever since the ffmpg sync about six 
> months ago, 1080i with xvmc has been creeping up in cpu usage. What 
> was once 45-55 percent is now 65-80 percent cpu usage. Have not tried 
> svn since before the multi-record sync. From the commits on the dev 
> list, the code is still settling down. Maybe in a few weeks, I'll try 
> svn again.
>
> There is also a race deadlock using 1080i xvmc that I have not found 
> yet. Random, sometimes minutes, sometimes hours before it will 
> deadlock and hang the mythfrontend.
>
>
> _______________________________________________
> mythtv-users mailing list
> mythtv-users at mythtv.org
> http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-users
>
>   
Check the audio settings. I found that on the 2.6.23.14-107.fc8 and the 
2.6.23.9-85.fc8 kernels using alsa was causing deadlocks and kernel 
oops. The systems are Epia M10Ks. Setting the audio to use /dev/dsp 
eliminated the problems and reduced cpu usage to less than half of what 
it was doing. This was with SD rather than HD. There may actually be 
additional variables (pulseaudio comes to mind) that contributed to the 
problem but I have not yet investigated completely.

Bruce