Hello.
we are 40+ Artists at EVE images (70+ Employees). We are using Corona as our Renderer of choice. We want a reliable Network rendering solution - based on Deadline.
What we are currently doing:
I wrote a custom submitter for Tile Jobs. It is submitting UHD cache Jobs seperately and takes care of important Corona Setting such as adaptive sampling. It does a lot more (like checking external files and stuff) but in a nutshell: it makes sure that all the tiles from rendered on different machines will fit without seams. so no bloom/glare, no camera distortion, no adaptivity, no time limit, no noise level limit and so on. we are basically disabeling a lot of cool corona features just to make it work.
Thats bad.
I recently wrote tools and defined a new standard for lighting at our office. As this is simulating lenses/cameras (and other stuff) the farm/distributing system we are looking for should be capable of all these things above. the normal tile/jigsaw rendering method of deadline can not produce that with corona. even if the camera distortion will work at some point, the sampling adaptivity might still not work and so on. while rendering nice bokeh, adaptive sampling is extremely important though... so my idea is the following now:
we have to use corona DR, as it seems to be capable handling all the requested features. But instead of just starting some corona DR slaves via Deadline, the job itself - so the master - should also render on the farm, not local.
This seems a bit tricky though. I would split up our 60 nodes into DR slaves and regular max job slaves. maybe even dynamically so it would adopt to the amount of jobs coming in (I guess this might somohow be possible with some code). but then? every submitted max job (with coronaDR internally enabled) would need to have some placeholder/wild-card rendernodes inside the CoronaDR slave list in order to render the job on just a fixed amount of free DR blades. "search lan during render" would result in one job using all free DR blades to render on, blocking other jobs. But using fixed blade names would result in fixed DR Blade groups that would not be dynamic enough.
I would like to find a solution that:
a: is capable of using all features of corona
b: utilizes and scales to all of the blades dynamically
c: is reliable
d: needs to be able to handle as much as 50+ Jobs a day
it does not matter if this involves some coding and stuff but it needs to be done. Also If someone has a better idea please feel free to share.
Thanks in advance :-)
PS. I know that Thinkbox/Amazon is working together with RenderLegion/ChaosGroup to solve some of these issues. This is all fine and good but we need a solution very soon, so sorry if you feel pushed ;-)
