Distributed simulation and adptative distribution, frustum clipping simulation

Phoenix Requests
1

Look at this! Amazing!

Do you think you could make Phoenixfd simulations distributable? They say there is a limit of around 14 computers if I remember correctly that can be used before the sim gets slower because of the time taken to distribute the data needed between computers on the network

2

Currently we don’t have plans for distributed simulation; it might be something to research in the future, but it won’t be coming anytime soon.

Best regards,
Vlado

3

Maybe it could be used at least for resimulation? (without cutting the sim but distubuting the entire frames). I tried to resimulate to slow down my explosion and I used 2 computers but it didn’t work there was very noticable discrepancy on the frames where the sim was stopped and restarted. Even on the same computer if I press stop and then start again the sim at the next frame there was a difference in the result.

4

I don’t think it works this way. The resimulation is not done frame-by-frame independently; there is velocity information that must be carried from one frame to the next.

Best regards,
Vlado

5

Yes I know but if each calculation is always started on an existing frame of the main sim it should work? I just wonder how the sim can reach the state of the existing following frames?

6

some kind of resimulations can be distributed, but it does not worth the efforts.
the main goal is to distribute the full simulation, that is really a hard task.

7

I really wonder how they can do it. I will to play with houdini to believe it! Cause I know the computers send info about the borders velocity to each other but to know the values of each of those border velocity the calculation needs to be done first… and the calculation depnds on all that’s happening everywhee in the whole sim…

8

the same is true for multiple core CPU simulations, but all the simulator software support parallel calculations :slight_smile:
the real difference between multiple machines and multiple cores is the communication speed. even on single machine, the bottle neck is not the cpu speed, it’s the ram/bus speed, the information transfer is the slowest part in the simulation process. for distributed simulation this parameter changes to worse, that’s why we don’t offer distributed simulation.