FRAMES PER TASK, when suspended, will re-start the entire FRAMES PER TASK

[christian_kim]

Hi everyone!

i am coming from backburner (autodesk render manager). backburner has its own set of issues, but one thing that it does is that its smart enough to not restart the host application every frame.

so when i am rendering 1-100 frames, all the render nodes will pick up the next frame and render each frame consecutively, without having to restart the host each frame.

in DEADLINE, each task will restart the HOST it seems, and to alleviate this problem, I am forced into using FRAMES PER TASK.

However, what i find is that say if the “task” of 10 frames are rendered half way (5 frames), and if i have to suspend the job or pull the machine out from render nodes, ENTIRE task will be reset, and when the next render nodes pick up the job, it will try to re-render the entire 10 frames from start.

so it often comes down to this somewhat stressful balancing of putting as much frames per task, so i reduce the restart times, but not too big where each task cancellation or problem has enormous re-rendering.

1. is this the best way of rendering animation sequence? (i am using houdini + redshift)

2. is there any way DEADLINE can suspend task min way, and pick up where it failed or paused? rather than having to re-start the entire task every time?

thank you in advance for any advise!

[mois.moshev]

Consider that a frame that was stopped midway could be partially rendered and broken. iirc Deadline does not know which frames in the task were rendered correctly, so all are restarted.
But, the correct way is to simply use 1 frame per task, and you get the guarantees you want.

[anthonygelatka]

As Mois says, run the job as 1 frame per task.

I usually aim for a task time of 10-30mins, if a frame takes 1min, I’ll chunk 10 per task. If I have to suspend a job then at worst I’ve lost 9/10 frames or 9mins rendering.

Some render engines allow you to ‘skip existing frames’ which have completed, this doesn’t always work though if you’re overwriting frames already.

I don’t think Redshift has ‘checkpoints’ (V-Ray, Mantra, Renderman do), which are handy for resuming renders from specific points.

I think Deadline keeps the job loaded between tasks unless a different job is rendered, you may want to test that theory if you have massive load times.