Learning Forum

[Jingyu Liu]

Hi,

I used the LES model to simulate the single phase fluid flow in the pipe. And I use the FW-H model to calculate the noise generation in the receiver.

The plot shows the simulation time vs the total number of steps I have run. The computational time is very wired.

In the plot, for the first 10000 steps, it takes 10 hrs, but for the second 10000 steps, it takes 50 hrs. And for the later steps, the computational time increase exponentionally. 

There are totally 5, 000 cells of my model. I tried, 5 10 15 20 30 cores to run the model, but all the tests obey the this relationship between computation time and number of time step. Which means, the computation time does not decrease even though I increase the cores.

What is the reason for this phenomenon? And what can I do to improve my simulation efficiency?

 

Best regards,

Leo

[Federico]

Hello Leo, 

if your model consists of only 5000 cells, then increasing the number of cores might not help since you will effectively be spending most of your computation time communicating the solution between cores/partitions.

Each GB of RAM should be able to handle in the order of 10^5 cells. So it is likely that you would not see much improvement by parallelizing this simulation.

[Federico]

Regarding the slowdown of computatoin with number of time steps, this can depend on several factors: depending on what data you are storing/displaying, files you are writing, models, etc.