Possible to simulate specific frames to Deadline?

[chris_widdowson]

Is there a workaround for this? Apparently you cannot distribute a job across multiple workers, which is quite disappointing; 1 job = 1 worker. Is it possible to split the simulated frame range so for the same simulation, you manually queue frames 0-50 on worker_01 and frames 51-100 on worker_02?

[Svetlin.Nikolov]

Hey,

I'm afraid every frame of the simulation depends on the result of the previous. Distributed simulation are much pickier than distributed rendering. In the past we did some experiments with distributing the calculations over the network, but the nature of simulations is that they require A LOT of transfer of data between the different areas of the simulator box, so our results showed some improvement in some cases, and even slower performance on many machines compared to just one machine in other cases. Of course there are networks that are optimized for speed, but very people have the luxury to have such at their disposal...

What would be a more practical choice right now is to run completely unrelated simulations on many machines in parallel, or run variations of the same simulation on many machines and then pick the best results.

Cheers!

[chris_widdowson]

Interesting, I suppose the best solution is to just configure our render nodes to maximize the per job compute speed. Thanks!

[chris_widdowson]

[resolved]