Parallel NS-kepsilon-HT: defect in velocity field

Numerical methods and mathematical models of Elmer
Post Reply
Takala
Posts: 186
Joined: 23 Aug 2009, 23:59

Parallel NS-kepsilon-HT: defect in velocity field

Post by Takala »

Hello,

I am running a Navier-Stokes simulation with k-epsilon turbulence model coupled with heat transfer equation. When I do a serial run everything is fine, however with parallel run I get strange results:
One symptom is the defects in the velocity field. These defects take place in the interfaces between two partitions (take a look at the attached picture)

Please advice.

Br,

Eelis
Attachments
paralleldefect.jpg
(18.93 KiB) Not downloaded yet
raback
Site Admin
Posts: 4812
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: Parallel NS-kepsilon-HT: defect in velocity field

Post by raback »

Hi Eelis

The given equations are not the most robust in Elmer. Are you sure that you reach convergence i.e. the results are not from an unconverged state.

By the way, you might get a nicer partitioning using the dual graph in Metis, e.g.

Code: Select all

-partdual -metis 64 4 
-Peter
Takala
Posts: 186
Joined: 23 Aug 2009, 23:59

Re: Parallel NS-kepsilon-HT: defect in velocity field

Post by Takala »

Hi Peter

The parallel run is working well when I use this dual graph.

Thank you!

-Eelis
raback
Site Admin
Posts: 4812
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: Parallel NS-kepsilon-HT: defect in velocity field

Post by raback »

Hi Eelis

The old partiotioning should have worked as well. There may be two reasons

1) The partitioning is such that the preconditioners don't work as well and you didn't actually get convergence and the results just shows this.

2) You did the convergence but there is a bug in the parallel implementation.

If the case is 2) it is a little bit alarming though you overcome it with better partitioning.

-Peter
Takala
Posts: 186
Joined: 23 Aug 2009, 23:59

Re: Parallel NS-kepsilon-HT: defect in velocity field

Post by Takala »

Hi Peter,

I checked it again, and the Navier-Stokes linear system has no problems to converge. A Bug then it is?

Please tell me how to proceed in finding the problem.

Cheers,

Eelis
Takala
Posts: 186
Joined: 23 Aug 2009, 23:59

Re: Parallel NS-kepsilon-HT: defect in velocity field

Post by Takala »

Takala wrote:Hello,

I am running a Navier-Stokes simulation with k-epsilon turbulence model coupled with heat transfer equation. When I do a serial run everything is fine, however with parallel run I get strange results:
One symptom is the defects in the velocity field. These defects take place in the interfaces between two partitions (take a look at the attached picture)

Please advice.

Br,

Eelis

Hi,

I'm still experiencing this issue. Even if I don't have the K-epsilon equation.

Do you have any ideas how to track this?

Cheers,

Eelis
Takala
Posts: 186
Joined: 23 Aug 2009, 23:59

Re: Parallel NS-kepsilon-HT: defect in velocity field

Post by Takala »

Hello,


I think this problem has to do with the Result Output Solver:
I tested the Step_ke case from /tests and one can see the issue also there if Result output solver is used. However, in ep format there are no problems at all.

Edit. No it is not. Now I tested again with my own model and ep file has the problem as well....


Cheers,

Eelis
Takala
Posts: 186
Joined: 23 Aug 2009, 23:59

Re: Parallel NS-kepsilon-HT: defect in velocity field

Post by Takala »

some errors (see attached log) when trying to run parallel (NS-heat coupling).

Cheers,

Eelis
Attachments
log.txt
errorlog
(5.12 KiB) Downloaded 284 times
Takala
Posts: 186
Joined: 23 Aug 2009, 23:59

Re: Parallel NS-kepsilon-HT: defect in velocity field

Post by Takala »

Hello,

I found the problem:

I used to use this command with ElmerGrid:

Code: Select all

ElmerGrid 8 2 file.unv -autoclean -metis 32 4 -partdual
And it is faulty! Now I use these: first convert to serial run

Code: Select all

ElmerGrid 2 2 file.unv -autoclean
Then do the partitioning by using the serial elmer mesh format

Code: Select all

ElmerGrid 2 2 file.unv -metis 32 4 -partdual
Cheers,

Eelis
raback
Site Admin
Posts: 4812
Joined: 22 Aug 2009, 11:57
Antispam: Yes
Location: Espoo, Finland
Contact:

Re: Parallel NS-kepsilon-HT: defect in velocity field

Post by raback »

Hi Eelis

Thanx for figuring that out! I will look at the problem time permitting.

-Peter
Post Reply