[Za-ck]

Hi all,

from time to time I have the problem, that some of my simulations crash before solving the first equation.

In my actual simulation it runs fine on 1 processor, 16 processors and 144 processors, but it crashes on 72 processors.

I am using scotch for the decomposition. Using 72 processors, each processor has ~60000 cells. Using the same case the problem occurs also on different computers. Most of the time I am only using 16 processors for similar mesh sizes. With 16 processors the problem never occured.

For me this error seems to be very arbitrary and destroys an efficient workflow completely.

Any ideas?

Best regards,
Za-ck

[Za-ck]

Ok, I think I have found the problem. FYI: It happens, when turbulence->validate() is called. I am initializing the velocity with zero and accelerate it afterwards. Maybe a small velocity should be initialized from beginning.

[cryabroad]

Originally Posted by Za-ck

Ok, I think I have found the problem. FYI: It happens, when turbulence->validate() is called. I am initializing the velocity with zero and accelerate it afterwards. Maybe a small velocity should be initialized from beginning.

If the simulation crashes when you called the turbulence->validate() function, I think it's more related to the turbulence fields. For example, if you use RANS k-epsilon model, the turbulence->validate() function calls correctNut(), which then uses the k and epsilon fields to calculate nut. Keep in mind that according to the code, nut = Cmu*sqr(k)/epsilon. Therefore, if your epsilon is initialized to be 0, this causes the typical dividing zero error.

Whatever turbulence model you are using, you can dig into the source code and find out what it is doing in the turbulence->validate() function. This should give you enough hints.

Ruiyan

[damon707]

Hello all,

Za-ck could you please elaborate how exactly you found out that the error occurs when turbulence->validate() is called?

Kind regards,
George