AWS Thinkbox Discussion Forums

Deadline with AI ;o)

I have some idea …


  1. When we launch multiple big render … like 333 frames by group of 5 frames … deadline should be able to detect free slave to dispatch frames that are waiting to render … like the second frame is rendering then dispatch the 3 others on free slave


  2. Deadline should evaluate time needed to render 1 frame for a job to know how much time is needed to do the complete job then if u have a job that needs 30min a frames and takes all slaves and u have one job that need 1min a frame then it should dispatch that job on some slave to get that job done because the other one.



    I know all that is doable by someone managing the render but can be cool to have a kind of Adaptive Mode …



    my 2 cents

Hi Daniel, I know this is not the kind of answer you might like to hear, but i have a suggestion for your 1 min render problem.



What I did here is setup a pool called small_jobs, this pool is reserved for mattes renders, normal pass, etc.



By putting this pool in the top of the list in the machine pool list, it forces these small jobs to render first.



I also try to enforce that any task of any given task doesn’t take more than 45 mins if possible, again to make sure that if my farm is running 100% and a small job get sent, the user only need to wait max 45 min for the small job to run, and by using large chunk on the small jobs, they render quite fast.



Hope this helps a bit



Sylvain Berger | Technical Director | Alpha Vision


Privacy | Site terms | Cookie preferences