Now that I have described what we did at Siggraph - I thought I would parcel out a small braindump of some of the things that I have been thinking about, to see what interest lies out there.
DRAFT
Since draft is built in, and you have liceneses already - I was considering using Draft to generate H264 or other Streaming formats of renders as they completed [or perhaps even partially complete]? or even JPEG thumbnails as jobs finish so that you can quickly review remote frames and a. know that they are working b. reduce transmission costs and bandwidth reduction to a fraction.
I don’t know that this is important to a garage facility - but if you have 30 artists, and everyone is submitting jobs to a hybrid farm including cloud - i’m sure you don’t want to have many concurrent streams of giant EXR files clogging up your bandwidth. Nor do you want incorrect jobs to waste expensive cycles on the cloud, so this may be a way to mitigate it to a greater or lesser extent…finally, why bring down EXRS at all, if you are going to run a comp script in the cloud? just leave the data there, review jpegs and then launch your nuke/fusion/ae job via deadline.
now, I realize that many people will not go to the cloud until they are ‘sure’ but as a former facility owner, anything that speeds up the review process so that I know the data is good is smart. I think Draft fits in here somewhere. Comments?
XMESH
We have clients who are using XMESH to save out ‘layers’ of their scenes from disparate physical locations and then use MAX to ‘composite’ those layers. So rather than render out layers and comp, they render out a full scene. Its an interesting approach, and I feel like XMESH could be a good way to cache out assets on the cloud that aren’t changing, and submit max jobs ands scripts that combine new assets, cameras, lighting, materials etc to reduce the subsequent differences or vectors.
With Maya XMESH coming, this may be something to consider.
jpeg proxies -> Nuke -> Final Render would be a no go for us for a variety of reasons:
No overbrights. We often grade down or up passes.
No 2.5D rendering. Even 16bit isn’t good enough for position passes for relighting.
No Alpha channel.
No AOVs if you bundle render elements. I hate multi layer EXRs but people still make them.
Compressed EXRs are often smaller than JPEGs thanks to data windowing.
Xmesh:
Seems like there is some good potential to reduce bandwidth but doesn’t that increase not decrease the complexity of relinking and version control?
I don’t know that you fully understood what I meant -
cloud machines render something. say it is big files for whatever reason.
draft on cloud makes h264 proxy, or thumbnails or whatever of images as they render.
you view proxies /thumbnails of cloud render before necessarily agreeing to download the exrs or whatever
OR you see the assets look correct and trigger the nuke job in the cloud [pointing to new assets] without even bothering to download new files.
i’m thinking all sorts of scenarios for this. you work locally at 1/8th res, render at 5k on cloud and leave files there and feed directly to nuke job on cloud, never downloading assets locally except for final deliverable etc. depending on what deliverable is, you may review it and send it directly from cloud.
Xmesh:
Seems like there is some good potential to reduce bandwidth but doesn’t that increase not decrease the complexity of relinking and version control?<<
sure, it complexity is increased over NO system, but imho far easier than a complete version control system. we had one we built in production for max, and dealing with all the possibilities and circular dependencies etc is a pain. if you could rely on xmesh to cache out assets and break down your scene into a specific number of elements…i think it would be helpful. then you would have a job reassemble it on deadline. Some day you should ask bobo about RPV or next time you see me with a table in hand, i’ll show you our video of it in action on gijoe.
thanks for the feedback, i’m interested I hearing the problems…