Hi Chris,

Answers inline.

2011/9/8 Christopher Brooks <[email protected]>

> Hi Ruben,
>
> Sounds exciting, I had it in mind to try something like this too, but a
> little different.  Some questions for you:
>
> 1. Network congestion is a big issue for us.  Are you ingesting to a
> central admin service then farming it back out to the capture agents,
> or are you ingesting directly to the agent the media is on?  This
> latter method could lead to huge performance gains.


We first tried to use the capture agents "as are", meaning deattached from
the core, except for the REST services. That implied using the
workspace-stub profile, but It turns out that we found out a nasty error
doing so (HTTP copy only supports 2GB [1]). Therefore now we deploy our
agents with the workspace, serviceregistry and worker profiles. The files
are ingested via HTTP [2] and once in the shared storage, there are no big
copies of files (it's everything in the same share). I've just realised that
this could be improved by slightly changing the agents ingestion to, for
instance, copying the zipped mediapackage directly in the inbox (which is
also in the share). Thanks for asking this! You've helped a lot!

[1] Is that a bug, by the way? Because it is mentioned in MH-7123's comments
and it seems like a known issue.
[2] If arbitrarily large files are ingested via HTTP, why is [1] failing? Or
we cannot ingest > 2GB files via HTTP?

2. What hardware are you using for your agents?  Did you try the

reference one to see how that would work?  It's the one I've been
> considering.
>

We started using Hauppauge cards for recording video, and Epiphans for the
VGA grabbing. Now we've given up on the Hauppauge because we've found plenty
of HD webcam models which give a considerably better image quality (HD says
it all, I guess). This comes at a price, of course: the agent needs to
decode the camera's mjpeg output to an xvid/avi video (mpeg consumes an
intolerably huge amount of space). That's one of the reasons we need to use
a more powerful machine for the agent, also.

3. How (or are) you scheduling the capture agents so they don't process
> while they are capturing?
>

Not yet. We will try and see how the machines respond. If we see the need,
then we will think of some mechanism (the easiest being limiting the
processing hours to the night time).


>
> Regards,
>
> Chris
>
_______________________________________________
Matterhorn-users mailing list
[email protected]
http://lists.opencastproject.org/mailman/listinfo/matterhorn-users

Reply via email to