[Tfug] ZoneMinder resource requirements

[Bexley Hall]

Hi Zack,

On 11/5/2012 3:09 PM, Zack Williams wrote:
> On Mon, Nov 5, 2012 at 9:18 AM, Bexley Hall<bexley401 at yahoo.com>  wrote:
>> Any data points from folks running ZM on otherwise
>> "un/lightly-loaded" machines?
>
> I'd tell you but the one I'm running on it died yesterday and I'm

<grin>  "The more stuph you've got, the more stuph you've got to *fix*!"
<frown>

> rebuilding it.  I'll let you know when I get it back up and have some
> data.

OK.  No rush, here (just investigating what sort of hardware I
need to throw at the problem)

> Are you thinking IP cameras shipping a JPEG stream, or composite video
> into framegrabbers?  Motion detection?

First appraisal is to do the image analysis (motion detection)
*at* the camera along with evaluating the alarm criteria.
In parallel, do any compression/coding of the video stream
for delivery to a storage service "elsewhere" (based on whatever
criteria you have defined: every frame, every other frame, etc.).

So, if you want to just archive video at some frame rate
*and* capture "events" at some different frame rate, the
camera can deliver what you need -- instead of making the
"back end" ask for it.

That way, the number of cameras can scale more readily (because
one "server" isn't doing all the work).  And, you can ship the
video off to "wherever" is convenient (instead of forcing one
machine to handle it all, etc.).

Finally, it allows for the "server" (whatever *that* is in this
scenario) to be overloaded without losing events.  I.e., you
can jump up and down in front of cameras A, B and C (using
lots of THE server's resources to detect that activity and
archive it) yet camera D (which is watching the bank vault) can
*still* detect events of interest to itself.

You would then have to prioritize which of these video streams
gets *recorded* (if the storage service is overloaded) but any
number of cameras can still "detect" events of interest regardless
of the (total) processing load involved (as it would be if a single
server did all the work).

I'll see if I can get a sample video stream to "batch" process
and see what their motion detection algorithm costs.

Thx!
--don