Re: Preparing UTFPR for the video team
On Fri, Aug 17, 2018 at 12:51 PM Louis-Philippe Véronneau
<pollo@debian.org> wrote:
>
> >>> * minipc
> >>> Specs?
> >>> i3 2gig of ram is acceptable. 10gig of disk per hour of recording.
> >>> pci-e slot for the capture card.
> >>
> >> In our recent tests, a 2015 i5 was not sufficient to run voctomix
> >> properly. I would highly recommend you get a machine with a recent i7 CPU.
> >
> > What tests?
> >
> > I just used a 2009 i3 to record 2 days of PyOhio.
> >
> > I have it running now with 2 gst videotestsrc/audiotestsrc which are
> > more cpu bound than reading from the BM or network card.
> >
> > core, gui, 2 sources, 2 file sync.
> > htop shows all 4 cores around 60%
>
> We used a 2015 i5 at DebConf18 and we had a lot of trouble.
> We had to
> reduce the preview quality a lot since we were loosing frame and were
> experiencing audio desync because of a high CPU usage.
>
2015 - there have been some significant performance improvements in
Voctomix. I think this invalidates your tests.
> I think the main difference between our setups is we have a 2nd camera
> and thus 3 sources.
voctocore - 60%
voctogui - 10 processes at 5% = 50%
ffmpeg file sink 30%
each source = 20%
3 sources - no cores maxed:
juser@cnt1:~$ sudo mpstat 5 1
Linux 4.9.0-7-amd64 (cnt1) 08/17/2018 _x86_64_ (4 CPU)
01:40:05 PM CPU %usr %nice %sys %iowait %irq %soft
%steal %guest %gnice %idle
01:40:10 PM all 53.14 0.00 13.78 0.16 0.00 0.69
0.00 0.00 0.00 32.23
juser@cnt1:~$ uptime
13:56:19 up 16 days, 55 min, 11 users, load average: 5.04, 7.09, 8.67
load swings between 4 and 8 about every minute. seems odd.
4 sources seems doable on this box - 10% idle.
5 sources ... seems to be working, but no:
Average: CPU %usr %nice %sys %iowait %irq %soft
%steal %guest %gnice %idle
Average: all 76.48 0.00 21.56 0.05 0.00 0.95
0.00 0.00 0.00 0.95
Average: 0 80.56 0.00 18.44 0.00 0.00 0.60
0.00 0.00 0.00 0.40
Average: 1 74.10 0.00 23.49 0.00 0.00 1.20
0.00 0.00 0.00 1.20
Average: 2 72.80 0.00 24.80 0.20 0.00 1.40
0.00 0.00 0.00 0.80
Average: 3 78.31 0.00 19.48 0.00 0.00 1.00
0.00 0.00 0.00 1.20
juser@cnt1:~$ uptime
14:45:02 up 16 days, 1:44, 11 users, load average: 31.60, 25.00, 16.68
>
> From our tests the linux kernel patches for Specter (KPTI) also has a
> large impact.
I was thinking video stuff might be impacted. current kernels are patch, right?
juser@cnt1:~$ uname -a
Linux cnt1 4.9.0-7-amd64 #1 SMP Debian 4.9.110-1 (2018-07-05) x86_64 GNU/Linux
>
> >>> * Cameras (borrowed from the university
> >>> What model?
> >>> they should have balanced audio in, and SDI or HDMI output.
> >>
> >> SDI is preferable to HDMI, since you'll have to use HDMI extenders if
> >> you go with HDMI.
> >
> > I use HDMI, and I don't use extenders.
>
> But you don't have an audience camera that is 50+ meters away from the
> mixing PC :P
Ah right.
If no one has done video before, I would start small. 1 camera, 1
Opsis, no streaming, no tally lights, no bgloop, basically what I
did for PyOhio. this will be about as educational as trying to run
the full stack of features.
My crystal ball predicts:
If someone can get possession of all the equipment 2 weeks before and
spend 4-8 hours setting it up with help from IRC, 80% chance of
success at the event.
If all that has to be done a day or two before, it drops to 40%.
If they try and do everything DC does... 20%.
I would much rather the crew have done an event successfully than
failed. There are pros n cons to both, but understanding how much
work it is to achieve results is more valuable than collecting
problems that need to be fixed next time. Because there are never
enough next times.
>
> --
> ,''`.
> : :' : Louis-Philippe Véronneau
> `. `'` pollo@debian.org / veronneau.org
> `
>
--
Carl K
Reply to: