Zum Hauptinhalt gehen

⚠️ Please note that this topic or post has been archived. The information contained here may no longer be accurate or up-to-date. ⚠️

Distributed (network) rendering?

Eric311
Is this in the works?

What i mean here is utilizing multiple computers, as in a \"render farm\", to make high-volume batch processing more efficient over a network.

We deal with multiple shoots per day, with some clients requesting entire shootings as high-res tiffs, so you can understand the necessity...

Thanks
0

Kommentare

7 Kommentare

Datum Stimmen
  • faeyer
    I'm not sure if raw conversions can be easily adapted to distributed processing. The network will probably become a bottleneck as the amount of data to move from computer to computer can be quite large. Packaging of individual tasks is, - I guess - , not trivial for raw file demosaicing.
    On the Mac, xGrid provides a good starting point.
    Another point to consider is the licensing scheme. How much Phase One would charge for every seat in the farm.

    I'd be interested to read what Phase One has to say about it.
    0
  • Eric311
    Thanks for interesting reply, faeyer. I'm no expert, but it the 3D rendering world this is quite a standard concept - can it really be that much more complicated for RAW? Usually it's a small, standalone application that can only be used for the render farm. We've got C1 on every computer here, the network app would/should just come standard with every license.

    Anyone else? Phase One guys-- surely you've got some comments here!

    We're talking about high-volume, high demand workflow - how do you deliver a shoot with 1000+ images to your client in high-res tiff?!!
    0
  • faeyer
    I think that one of the problems would be to \"partition\" the rendering job is smaller and independant jobs. This \"partionning\" could be done by either the programmers or the compiler/linker/build system. The independance of small jobs is a tough issue, IMO. Only Phase One programmers can shed some light on this subject, they know how they handle the process.
    As said previously, the other tecnhical issue is network traffic. With 3D apps, I guess that the ratio between computing power vs amount of data is far more favorable than with RAW processing. You can get a glimpse of this problem if you use a remote server to store your RAW files and try to work with Capture One directly from that server. In other words, how long does it take to send a (partial) file to another computer and get the converted data back on the \"master\" computer?

    I'm not saying that it can't be done but I would strees that it's not a trivial task like turning on a compiler option in xCode! Having many G5 boxes around in the office, I would be the first one to appreciate an efficient distributed RAW file processing. 😂
    0
  • faeyer
    As for licensing scheme, this is just a \"political\" decision...
    It's not uncommon in parallel computing to base licensing costs upon the number of processor used or computer nodes in the farm.

    If you want to learn what Apple does on HPC you can visit this page:
    http://www.apple.com/education/science/resources.html
    0
  • Eric311
    Thanks again for the ideas - of course it's more technical than I'm trained for!

    That said, we do develop regularly from an XServe - all of our shoots go in, and we develop over the network (not always, but often). No real problems there.

    But yes, 3d files are much, much smaller, and then there's the problem of moving icc profiles, etc...

    As far as licensing, again in 3d - it's just a small app than can only do one task, which is to recieve information from one licensed full version, over the network (meaning you can't change anything, view files, etc...) So i don't see why that would be an issue.

    PhaseOne guys - is this really too far out there?
    0
  • Ulf Liljegren
    Problem is that we have no real chance of testing this out this is why I am hesitant to answer.

    But it should work fine when it works like a hierarchy like a old Windows NT4 domain structure where you have a Controller (PDC in NT4) and all other as Xgrid agents (BDC in NT4) so any job that the controller is working on will be distributed.
    Question is how this will work application / File wise.
    This I can not answer at current point.
    0
  • faeyer
    [quote="Gleebitz" wrote:
    ...That said, we do develop regularly from an XServe - all of our shoots go in, and we develop over the network (not always, but often). No real problems there...?


    Well, if you use xServes along with xServe RAID or other very high performance network connections, you should have eliminated the network bottleneck. 😄

    The only other issue is in Phase One's hands and as Ulf kindly responded above, they have not - yet - had the opportunity to test.

    Let's hope that someone at Phase falls in love with xGrid 🤭 🤭 🤭

    Faeyer
    0

Post ist für Kommentare geschlossen.