Fair enough. It's just that when I read "we have to do an additional
decoding", I thought "there's a way you don't need to decode". Anyway, your
reasons seem more than reasonable :)

2012/2/9 Ruediger Rolf <[email protected]>

>  Hi Ruben,
>
> we don't have performance problems currently, so why should I consider
> this. And we write the MPEG2-stream directly to file, so that this can not
> have framedrops or whatever, because it is harware encoded. We would only
> encounter problems in the live-streaming, but who cares about a few dropped
> frames there.
>
> Rüdiger
>
> Am 09.02.2012 11:01, schrieb Rubén Pérez:
>
> Hey Rüdiger,
>
>  so we have to do an additional decoding and re-encoding to h.264 for the
>> streaming.
>
>
>  Doesn't your Hauppauge model provide a "raw" video output? Have you
> considered using it instead the mpeg2-encoded one? We have tested two or
> three different models and all of them "created" three devices in the OS:
> one corresponding to the raw audio, other to the raw video, and other to
> the mpeg2-encoded combination of the former. I know that decoding mpeg2 is
> not very cpu-consuming, but you can give it a try anyway.
>
>  Best regards
>
> 2012/2/9 Ruediger Rolf <[email protected]>
>
>> Hi Kevin,
>>
>> we only use two i7 for capturing. One of these is a Windows PC using the
>> Blackmagic H.264 Pro Recorder box (that's why windows) for HD-recording.
>> But as this box is still missing the APIs that Blackmagic promissed the
>> recording is still manual. The machine is quite over-powered (20-30% CPU
>> load while capturing 1280x800 vga and 1080i video simultanous).
>> The second i7 is in use for our live streaming test system, that we use
>> for over a semester now. If we record, enable confidence monitoring and do
>> h.264-live-streaming for 2 streams (SD-video only) with the Matterhorn
>> Capture Agent we are getting to 40-50% CPU load. So again this machine
>> could be slower.  I production we used only single stream live-streaming,
>> which ran at about 30% CPU load. We are using the Hauppauge PVR-150 in all
>> our machines, so we have to do an additional decoding and re-encoding to
>> h.264 for the streaming.
>> Both machines are Dell Optiplex MT 980 with the cheapest quad-core i7 and
>> 4 GB RAM.
>>
>> In general we use Dell Optiplex 780 as recorders with a Pentium 4 Dual
>> Core with 2.8 Ghz and 4GB RAM. They run at 70-80% CPU load while recording
>> 2 streams. With the optimisations in Matterhorn 1.3 these machines can do
>> recording and live streaming at 90-100% CPU load, what may cause some frame
>> drops if we are unfortunate. For only recordings the general CPU load will
>> go down significantly too with 1.3. Again we are using Hauppauge PVR-150
>> and epiphan VGA2USB LR on all of these machines.
>>
>> Currently if I would need to buy new computers as recorders I would
>> recommend a Dell Optiplex MT 790 with a i5-2400. As this would probably
>> have enough power for reliable live-streaming at least.
>>
>> If you are interested I can try to check how much CPU load it will take
>> on the i7 to capture from the epiphan device and a Blackmagic Intensity Pro
>> HD-capture-card. This is already on my to do list, but I had no time to
>> evaluate this yet. The capture card is lying on my desk for a a while
>> already...
>>
>> Rüdiger
>>
>> Am 08.02.2012 18:14, schrieb Kevin Chan:
>>
>>  Hi Nils,
>>>
>>> Thanks for your info. While I have heard of other universities running
>>> different hardware, this is the first documented case (as far as I know) of
>>> someone specifically stating that someone is running capture agents using
>>> i5/i7 cores.
>>>
>>> At UC Berkeley, we are likely to need some higher powered capture agents
>>> to do some HD camera captures (and possibly for SD camera captures) and
>>> that's why I have been trying to get the Opencast Community to share their
>>> hardware/software/config setup on this list (and even better would be on
>>> the Opencast Wiki).
>>>
>>> As such, can you share some basic information about your setup (perhaps
>>> with a note on the path that led you to your particular setup)?
>>>
>>> Even though we are still in our very early stages of testing/silent
>>> piloting, if I have a few minutes today, I will try to get a "capture agent
>>> setup page" for UC Berkeley started on the Opencast wiki so that others can
>>> benefit from this information.
>>>
>>>  Kevin Chan
>>>
>>>  Operations Team
>>>  Educational Technology Services
>>>  UC Berkeley
>>>
>>>
>>> On 2/7/12 10:46 PM, Nils Birnbaum wrote:
>>>
>>>> Hi Kevin,
>>>> besides Gallicaster, this is a general problem. Most Universities run i5
>>>> or i7 processor to capture sufficient framerates for 2-stream-recording.
>>>> So you have a trade off between CPU-Power and framerate (and some RAM)
>>>> to
>>>> get a result that match with the needs of your institution.
>>>>
>>>> Regards
>>>> Nils
>>>>
>>>>
>>>>  Hi Ruben,
>>>>>
>>>>> Thanks for the info, your approach sounds reasonable to me.
>>>>>
>>>>> I am still unclear on what hardware setup you are using. I found this
>>>>> page -
>>>>> http://wiki.teltek.es/display/Galicaster/Hardware+recommendations
>>>>> - so I am guessing you are probably running your captures against an
>>>>> Intel i3 processor with some sort of Hauppauge card.
>>>>>
>>>>> So in summary:
>>>>>
>>>>> a) using the above hardware setup, you had some video sync/dropped
>>>>> frame
>>>>> rate issues
>>>>> b) these issues were fixed with "a different pipeline structure with
>>>>> the
>>>>> element 'videorate' to guarantee the video stream synchronization"
>>>>>
>>>>> In theory, adding "videorate" to the MH capture agent pipeline should
>>>>> also fix this issue, but the placement of this element is probably key
>>>>> to the whole thing!
>>>>>
>>>>> Let me know if I got any of the above wrong.
>>>>>
>>>>>    Kevin Chan
>>>>>
>>>>>    Operations Team
>>>>>    Educational Technology Services
>>>>>    UC Berkeley
>>>>>
>>>>>
>>>>> On 2/6/12 4:51 AM, Rubén Pérez wrote:
>>>>>
>>>>>> Hi all,
>>>>>>
>>>>>> The pipeline we are using in Galicaster is rather different than the
>>>>>> one used in the standard capture agent in many ways, the first one
>>>>>> coming to my mind being the fact that Galicaster needs to provide
>>>>>> video feedback and the CA doesn't.
>>>>>>
>>>>>> Still, I will take a look and see if we can commit some changes to
>>>>>> improve the synchronization quality. I cannot guarantee though.
>>>>>>
>>>>>> The pipeline used is in the code of Galicaster, available at
>>>>>> www.galicaster.org<http://www.galicaster.org>  .
>>>>>>
>>>>>> Sorry I cannot be more specific right now. When I have some spare time
>>>>>> I'll try to document myself on this topic and try to apply a patch
>>>>>> based on our research in Galicaster, if that's feasible without
>>>>>> altering the current CA pipeline too much. Does this sound reasonable?
>>>>>>
>>>>>> Regards
>>>>>>
>>>>>> 2012/2/2 Kevin Chan<[email protected]
>>>>>> <mailto:[email protected]>>
>>>>>>
>>>>>>     Hi Ruben,
>>>>>>
>>>>>>     Was this "fix" done against the referenced MH capture hardware or
>>>>>>     some other hardware setup?
>>>>>>
>>>>>>     In any case, thanks for the info, it would be great if you can
>>>>>>     provide this info for the MH developers in:
>>>>>>     http://opencast.jira.com/browse/MH-8505
>>>>>>
>>>>>>     Also, if you can provide the full pipeline structure, that would
>>>>>>     be great (for novice pipeline constructors like me).
>>>>>>
>>>>>>        Kevin Chan
>>>>>>
>>>>>>        Operations Team
>>>>>>        Educational Technology Services
>>>>>>        UC Berkeley
>>>>>>
>>>>>>
>>>>>>     On 1/30/12 5:38 AM, Tobias Wunden wrote:
>>>>>>
>>>>>>>     Hi Ruben,
>>>>>>>
>>>>>>>     any chance that fix will make it back into trunk anytime soon? It
>>>>>>>     would have been great to have this in for 1.3 as well, given that
>>>>>>>     you seem to have sorted out that issue a while ago.
>>>>>>>
>>>>>>>     Tobias
>>>>>>>
>>>>>>>     On 30.01.2012, at 09:15, Rubén Pérez<[email protected]
>>>>>>> <mailto:[email protected]>>  wrote:
>>>>>>>
>>>>>>>      Hi Kevin,
>>>>>>>>
>>>>>>>>     I can also confirm that we experienced the same frame drop/out
>>>>>>>>     of sync problem in long recordings, both in the capture agent
>>>>>>>>     and also in Galicaster's development stages. Finally, we
>>>>>>>>     discovered that the key was a different pipeline structure with
>>>>>>>>     the element 'videorate' to guarantee the video stream
>>>>>>>>     synchronization.
>>>>>>>>
>>>>>>>>     Maybe you could try and add this element to the CA's standard
>>>>>>>>     pipeline, and see what happens. Of course, you can also try
>>>>>>>>     Galicaster :)
>>>>>>>>
>>>>>>>>     Good luck
>>>>>>>>     Rubén
>>>>>>>>
>>>>>>>>     On 25 Xan, 2012 07:18, "Kevin Chan"<[email protected]
>>>>>>>> <mailto:[email protected]>>  wrote:
>>>>>>>>
>>>>>>>>         Hi all,
>>>>>>>>
>>>>>>>>         During preliminary testing of a 2h 15m capture (which we
>>>>>>>>         intend to post produce for YouTube), UC Berkeley discovered
>>>>>>>>         that the capture files for video and screen were off by 15m
>>>>>>>>         (with the screen capture via Epiphan being 15m short), which
>>>>>>>>         is more or less unacceptable to our post production folks
>>>>>>>>         (as it is very difficult for them to stitch this back
>>>>>>>>         together). This is likely to be due to dropped frame rate,
>>>>>>>>         but it is a bit difficult to confirm.
>>>>>>>>
>>>>>>>>         I am wondering if anyone else (besides Saskatchewan folks,
>>>>>>>>         who confirmed this issue) has encounter this issue and how
>>>>>>>>         they are working around it. I imagine that lowering
>>>>>>>>         framerates and bitrates would work, though some initial
>>>>>>>>         testing using various capture agent configurations seems to
>>>>>>>>         suggest that the capture files are still off by a few
>>>>>>>>         minutes even with the CPUs being not fully taxed.
>>>>>>>>
>>>>>>>>         We are using the reference capture agent hardware and very
>>>>>>>>         basic capture agent settings. This issue has been filed in
>>>>>>>>         MH Jira if anyone is interested:
>>>>>>>>         http://opencast.jira.com/browse/MH-8505
>>>>>>>>
>>>>>>>>         Finally, we have heard that there are various hardware and
>>>>>>>>         configuration setups for Matterhorn Capture Agents in
>>>>>>>>         production. I think it would be great if that we, as a
>>>>>>>>         community, can share this information (via the MH wiki) to
>>>>>>>>         help each other understand and learn about the various use
>>>>>>>>         cases and setups that are employed.
>>>>>>>>
>>>>>>>>         Thanks,
>>>>>>>>
>>>>>>>>         --
>>>>>>>>            Kevin Chan
>>>>>>>>
>>>>>>>>            Operations Team
>>>>>>>>            Educational Technology Services
>>>>>>>>            UC Berkeley
>>>>>>>>
>>>>>>>>
>>>>>>>>         _______________________________________________
>>>>>>>>         Matterhorn-users mailing list
>>>>>>>>         [email protected]
>>>>>>>> <mailto:[email protected]>
>>>>>>>>
>>>>>>>> http://lists.opencastproject.org/mailman/listinfo/matterhorn-users
>>>>>>>>
>>>>>>>>     _______________________________________________
>>>>>>>>     Matterhorn-users mailing list
>>>>>>>>     [email protected]
>>>>>>>> <mailto:[email protected]>
>>>>>>>>
>>>>>>>> http://lists.opencastproject.org/mailman/listinfo/matterhorn-users
>>>>>>>>
>>>>>>>
>>>>>>>     _______________________________________________
>>>>>>>     Matterhorn-users mailing list
>>>>>>>     [email protected]
>>>>>>> <mailto:[email protected]>
>>>>>>>
>>>>>>> http://lists.opencastproject.org/mailman/listinfo/matterhorn-users
>>>>>>>
>>>>>>     _______________________________________________
>>>>>>     Matterhorn-users mailing list
>>>>>>     [email protected]
>>>>>> <mailto:[email protected]>
>>>>>>
>>>>>> http://lists.opencastproject.org/mailman/listinfo/matterhorn-users
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> Matterhorn-users mailing list
>>>>>> [email protected]
>>>>>> http://lists.opencastproject.org/mailman/listinfo/matterhorn-users
>>>>>>
>>>>> _______________________________________________
>>>>> Matterhorn-users mailing list
>>>>> [email protected]
>>>>> http://lists.opencastproject.org/mailman/listinfo/matterhorn-users
>>>>>
>>>>>  _______________________________________________
>>>> Matterhorn-users mailing list
>>>> [email protected]
>>>> http://lists.opencastproject.org/mailman/listinfo/matterhorn-users
>>>>
>>> _______________________________________________
>>> Matterhorn-users mailing list
>>> [email protected]
>>> http://lists.opencastproject.org/mailman/listinfo/matterhorn-users
>>>
>>>
>>
>>   --
>>
>> ________________________________________________
>> Rüdiger Rolf, M.A.
>> Universität Osnabrück - Zentrum virtUOS
>> Heger-Tor-Wall 12, 49069 Osnabrück
>> Telefon: (0541) 969-6511 - Fax: (0541) 969-16511
>> E-Mail: [email protected]
>> Internet: www.virtuos.uni-osnabrueck.de
>>
>>
>> _______________________________________________
>> Matterhorn-users mailing list
>> [email protected]
>> http://lists.opencastproject.org/mailman/listinfo/matterhorn-users
>>
>
>
>
> _______________________________________________
> Matterhorn-users mailing 
> [email protected]http://lists.opencastproject.org/mailman/listinfo/matterhorn-users
>
>
>
> --
>
> ________________________________________________
> Rüdiger Rolf, M.A.
> Universität Osnabrück - Zentrum virtUOS
> Heger-Tor-Wall 12, 49069 Osnabrück
> Telefon: (0541) 969-6511 - Fax: (0541) 969-16511
> E-Mail: [email protected]
> Internet: www.virtuos.uni-osnabrueck.de
>
>
> _______________________________________________
> Matterhorn-users mailing list
> [email protected]
> http://lists.opencastproject.org/mailman/listinfo/matterhorn-users
>
>
_______________________________________________
Matterhorn-users mailing list
[email protected]
http://lists.opencastproject.org/mailman/listinfo/matterhorn-users

Reply via email to