Tweet
Problem: We often render large print renders, sometimes we use distributed on the farm and sometimes we don't, depending on availability and time. Sometimes the farm frees up midway through a render and it would be nice to be able to enable the farm from there out without having to stop to enable distributed and re-render.
Note: Our farm has about 550 cores, so there is a lot to manage and keep track of.
Possible Solution 1: Make the distributed buckets always active. If we were to fire off every render with distributed on and every machine on the farm always checked it would be a simple solution. Unfortunately it seems like after the initial firing off of a render, the vray server if busy is ignored from then out, even if it frees up later. So you can't just always leave this checked. It can seem to overwhelm the server on the farm too if multiple renders try to access it, sometimes crashing it. A solution would be to have it check availability at a regular interval, or have an "server free" variable that is always checked for. Making this regularly check instead of just initially checking for free nodes would be a tremendous time saver and would cut out a lot of farm time-management issues. Then we could just leave all of the farm and distributed always checked, and we can all share it freely and it would always be fully booked.
Possible Solution 2: Make the Distributed Render window and features always active, even during render, so you can re-open it midway through a render and enable machines on the fly. This would be ideal for manually controlling the environment and including something like a priority in this dialogue would be a cool bonus. This would solve distribution problems and would really help in crunch times. Both of these options together would solve all of our large distributed render problems.
I hope that is clear enough. I'd be happy to help answer questions or develop a solution if I can. Just let me know.
-Joel E.
Note: Our farm has about 550 cores, so there is a lot to manage and keep track of.
Possible Solution 1: Make the distributed buckets always active. If we were to fire off every render with distributed on and every machine on the farm always checked it would be a simple solution. Unfortunately it seems like after the initial firing off of a render, the vray server if busy is ignored from then out, even if it frees up later. So you can't just always leave this checked. It can seem to overwhelm the server on the farm too if multiple renders try to access it, sometimes crashing it. A solution would be to have it check availability at a regular interval, or have an "server free" variable that is always checked for. Making this regularly check instead of just initially checking for free nodes would be a tremendous time saver and would cut out a lot of farm time-management issues. Then we could just leave all of the farm and distributed always checked, and we can all share it freely and it would always be fully booked.
Possible Solution 2: Make the Distributed Render window and features always active, even during render, so you can re-open it midway through a render and enable machines on the fly. This would be ideal for manually controlling the environment and including something like a priority in this dialogue would be a cool bonus. This would solve distribution problems and would really help in crunch times. Both of these options together would solve all of our large distributed render problems.
I hope that is clear enough. I'd be happy to help answer questions or develop a solution if I can. Just let me know.
-Joel E.
Inter-process communication is always tricky.
Comment