[Lilly]

Dear all,

I am a bit confused about which kind of UDF I have to parallelize, when I use several nodes for calculation.
I was told that I don't have to parallelize a viscosity UDF (DEFINE-PROPERTY), since the calculation only uses information from the same cell.
Now, I also wrote a UDF for a velocity an mass fraction profile (DEFINE_PROFILE) at the inlet of my geometry and a UDF to define my mass diffusion coefficent depending on the strain rate (DEFINE_DIFFUSIVITY).
I didn't parallelize them, but they seem to work.
I now read that DEFINE_PROFILE udfs should be parallelized as well.
Is this true? And what about the diffusivity? And would I have gotten an error message, if the udf didn't have worked in parallel mode, or shouldn't I trust the results now?

Thank you for any help!
Lilly

[ghost82]

I'm also curious about this question, as I also worked with udfs and parallel cores without changing the code, and also to me it seems all was ok.

All I can say is that the need of parallelizing an udf depends also on the way the mesh is partitioned among the different nodes; for example, if the inlet belongs only to a single node and you apply the udf, I think it doesn't need to be parallelized; on the contrary it should be parallelized if the inlet is divided among different nodes (..but I'm not 100% sure).

Awaiting for other replies.

Daniele

[diamondx]

Hope this can help https://dl.dropboxusercontent.com/u/...n-parallel.pdf

[Lilly]

Thank you Daniele and thank you diamondx for your help!
Your replies helped me a lot and the link is really helpful!

So maybe you can also help me with another question (since I am really bad with C).
This is my udf:
#include "udf.h"
#include "math.h"

#define MU_INF 0.035
#define N_INF 1
#define DELTA_MU 0.25
#define DELTA_N 0.45
#define A 50
#define B 3
#define C 50
#define D 4
#define Dconst 1.5e-10

real mu;
real lambda;
real n;
real strain;
real d;


DEFINE_PROPERTY(cell_viscosity, c, t)
{
strain = C_STRAIN_RATE_MAG(c, t);
lambda = MU_INF+DELTA_MU*exp(-(1.+fabs(strain)/A)*exp(-B/fabs(strain)));
n = N_INF-DELTA_N*exp(-(1.+fabs(strain)/C)*exp(-D/fabs(strain)));
mu = 0.1*lambda*pow(fabs(strain),(n-1.));

return mu;
}

DEFINE_DIFFUSIVITY(diffusivity, c, t, i)
{
strain = C_STRAIN_RATE_MAG(c, t);
d = 0.53*Dconst+5.292e-9*strain;

return d;
}

• I just realized I didn't but a point behind the defined values of N_INF A B C D (so if I got it right it's an integer instead of a floating point number). I already used this udf in several simulations and I didn't realize something strange. Do you know whether Fluent does have problems with this or whether the solution of the udf would be different, if I would have defined A B C D and N_INF as integer?
• Concerning my original question: I hope I also got it right, that I don't have to parallelize the viscosity and diffusivity udf, since they are only working at each single cell and don't need to correspond? I guess it's the same for the Macro C_STRAIN_RATE_MAG(c, t) I used there, isn't it?

Thank you for any help!
Lilly

[ghost82]

Originally Posted by Lilly

• I just realized I didn't but a point behind the defined values of N_INF A B C D (so if I got it right it's an integer instead of a floating point number). I already used this udf in several simulations and I didn't realize something strange. Do you know whether Fluent does have problems with this or whether the solution of the udf would be different, if I would have defined A B C D and N_INF as integer?
• Concerning my original question: I hope I also got it right, that I don't have to parallelize the viscosity and diffusivity udf, since they are only working at each single cell and don't need to correspond? I guess it's the same for the Macro C_STRAIN_RATE_MAG(c, t) I used there, isn't it?

Hi, your undefined "integer" values will be automatically converted because you defined the other variables as real, and in C language you cannot perform operations between integer and real values.
Probably you would have error if you define that values as integer.

For the second point, reading diamondx's document I understand that you have not to parallelize the udf, since, as you're writing, they are only working at each single cell.

Daniele

[Lilly]

Thanks a million for this information and all your help, Daniele!
I am relieved to hear that! I hope I'll get better with C language and udfs soon...

I wish you all the best!
Lilly