[mythtv-users] tests: CPU load and capture size

[Ray Olszewski]

At 01:16 AM 6/6/2003 -0400, Isaac Richards wrote:
>On Friday 06 June 2003 12:59 am, Ray Olszewski wrote:
> > I haven't
> > tested this one yet, but I'm wondering if using an aspect ratio that is not
> > 4:3 increases the load on the frontend (decoding and displaying) process
> > appreciably ... I've always been puzzled as to why mythfrontend uses so
> > much more CPU than other decoding applications like xine.
>
>Hmm -- quite possibly.  It _may_ be that the Xv stuff is doing at least some
>of the scaling in software on your machine, but you've said you're using a
>nvidia card, so I don't think it should be doing that.  Still, might be on to
>something, there.
>
>Isaac

OK, Isaac. here are some actual numbers from this morning's tests. I'm a 
bit puzzled about what they mean ... some speculation is at the end ... but 
I can vouch that they are reported accurately.

System: Cel 1.7 GHz, 256 MB PC133 SDRAM, AverTV bt878 vidcap card, nVidia 
GeForce4 MX440-SE AGP Video Card with TV-out (using nVidia's X driver). 
Debian-Sid. Combined MythTV frontend/backend using pre-packaged .debs for 
Sid. X display is set at 640x480x16bpp. Xvideo and TV out work.

Idle: 0-1% CPU usage. (Idle means the usual background processes, plus an 
ssh connection running "top").

Tests run in order reported. Usage is approximate based on visual 
observation of changing values reported by "top". All tests with the same 
channel (Food TV, the cleanest channel I receive). All caps are NTSC and 
use MPEG4 buffering; other settings are Myth defaults.

"live" TV at 320x240:   7-12% CPU "user" usage
         backend: 2-8%
         frontend: 2-4%
         system: 2-5%

"live" TV at 640x480:   99-100% CPU "user" usage
         backend: 75%
         frontend: 19%
         system: 4%
         capture is noticably erratic and jumpy

"live" TV at 480x480:   75-80% CPU "user" usage
         backend: 70%
         frontend: 5%
         system: 4%

"live" TV at 640x240:   35-50% CPU "user" usage
         backend: 35-45%
         frontend:  1-3%
         system: 3%

"live" TV at 480x240:   25-35% CPU "user" usage
         backend: 22-30%
         frontend:  1-4%
         system: 1-3% (occasional transients at 7%)

This suggests to me that my earlier guess was wrong: the high frontend 
usage in the biggest case is merely a side effect of hitting the CPU limit 
(causing increased process contention), rather than anything specific about 
the HxV proportions used. The frontend is behaving just as it should. The 
one odd thing I see here is that relative to any other choice, 320x240 is a 
disproportionately low user of CPU cycles (I mean low relative to the pixel 
count).