[p.norouzi.gv]

Hi all.


I am doing a simulation for clot formation and I developed 28 UDS. I have to simulate it for 60 minutes. The biggest time step that I can have is 10^-5s. I am using parallel simulation with 2 cores. (In this situation, the simulation will take a couple of months) When I increase the cores there is no effect on simulation speed. I would appreciate it if any of you could give any advice on how I can increase the speed of my simulation.
I already tried to decrease the under-relaxation factor (0.3, 0.5,08) and decreased element number, but they didn't have any effect

[arjun]

"I have to simulate it for 60 minutes. The biggest time step that I can have is 10^-5s."

You did not mention anything about the models involved and what is restricting this time-step.


If you are open to other softwares then you can try wildkatze. I am sure it could go much faster there (or made to go faster ). (I develop wildkatze so i know what is possible there).

(60 minutes it still high but usually you can get away with much larger timesteps in wildkatze).

[JBeilke]

The recommended strategy is to increase the relaxation factors and reduce the number of inner iterations.

[bigfootedrockmidget]

Originally Posted by p.norouzi.gv

Hi all.
take a couple of months) When I increase the cores there is no effect on simulation speed.

Your parallellization is not working correct it seems. If you are using user defined functions, make sure they are properly parallellized. Also check with a simple (but large) test case that parallelization is working correctly.

[p.norouzi.gv]

Originally Posted by JBeilke

The recommended strategy is to increase the relaxation factors and reduce the number of inner iterations.

Thanks for your response. I tried it and no effects !

[p.norouzi.gv]

Originally Posted by bigfootedrockmidget

Your parallellization is not working correct it seems. If you are using user defined functions, make sure they are properly parallellized. Also check with a simple (but large) test case that parallelization is working correctly.

While using 2 core the process is faster in compare with 1 core. But after increasing to higher cores there is no effect. It seems that parallelization I correct. I have a small geometry with 50k messh element. And it is very wired that why convergence is not good . Do you have any idea ?

[naffrancois]

Originally Posted by p.norouzi.gv

While using 2 core the process is faster in compare with 1 core. But after increasing to higher cores there is no effect. It seems that parallelization I correct. I have a small geometry with 50k messh element. And it is very wired that why convergence is not good . Do you have any idea ?

This is weird. How many cores do you have in your machine ? How many cores do your licence allow ? With a 50k elements mesh, you should see an improvement in speed when increasing core count, given your machine has more than 2 cores...

[p.norouzi.gv]

Originally Posted by naffrancois

This is weird. How many cores do you have in your machine ? How many cores do your licence allow ? With a 50k elements mesh, you should see an improvement in speed when increasing core count, given your machine has more than 2 cores...

I am not sure about the number of cores which my license allows, I am using student version. but I can increase the cores up to 32 on my system. I am trying to simulate it on my Laptop which has a high speed RAM, the speed of simulation is very high in compare with PC.
I am solving 29 equation which are coupled with each there. It seems that the RAM is effect the speed. what do you think?

[Hamidzoka]

Originally Posted by p.norouzi.gv

Hi all.


I am doing a simulation for clot formation and I developed 28 UDS. I have to simulate it for 60 minutes. The biggest time step that I can have is 10^-5s. I am using parallel simulation with 2 cores. (In this situation, the simulation will take a couple of months) When I increase the cores there is no effect on simulation speed. I would appreciate it if any of you could give any advice on how I can increase the speed of my simulation.
I already tried to decrease the under-relaxation factor (0.3, 0.5,08) and decreased element number, but they didn't have any effect

Hi,
Solution algorithm has an impact on convergence speed.
E.g. coupled algorithm takes more CPU time compared with SIMPLE.

best

[p.norouzi.gv]

Originally Posted by JBeilke

The recommended strategy is to increase the relaxation factors and reduce the number of inner iterations.

Thanks for your suggestion, but I already changed the under relaxation between 0.2 and 1 and there was no effect.