Problems with large meshes

Extension of Elmer in computational glaciology
Post Reply
Evan
Posts: 12
Joined: 12 Aug 2014, 14:42
Antispam: Yes

Problems with large meshes

Post by Evan »

Hello,

I am currently trying to run a large (read: ice sheet sized) simulation, which demands a fairly fine resolution mesh. I am having problems with Mumps when using the Navier Stokes solver:

Code: Select all

ListToCRSMatrix: Matrix format changed from CRS to List
 problem with NIV2_FLOPS message  -1.1920928955078125E-007           3  -575999060.00000012
From what I understand, the problem is related to the amount of memory allotted to Mumps, and there is a keyword available to increase it.

Code: Select all

  mumps percentage increase working space = integer 60
This did indeed fix the problem with some of my tests, but when I scaled up the mesh size, it still did not work. Any advice?
fgillet
Posts: 46
Joined: 30 Sep 2010, 16:58

Re: Problems with large meshes

Post by fgillet »

Hello;

which version of MUMPS are you using?
looking from the MUMPS versions changelog.:
from 4.8.4 to 4.9: 18. Message "problem with NIV2_FLOPS message" should no more occur
from 4.10.0 to 5.0.0: 24. Message "problem with NIV2_FLOPS message" should no more occur (there was still an occasional problem in 4.10.0)


I'm using verson 4.10.0 (I used 4.9.2 before)
When there is a problem with the workspace in MUMPS, the error returns something like : INFO(1)=-9; and according to the mumps user guide the work space should be increased vi an input parameter of MUMPS; this is what is done by this keyword "mumps percentage increase working space".
Have you tried higher values?

If you are running very large simulations you may want to give a try to the ParStokes solver?

Fabien
Evan
Posts: 12
Joined: 12 Aug 2014, 14:42
Antispam: Yes

Re: Problems with large meshes

Post by Evan »

Hi Fabien, thanks for the response.

I believe the supercomputer I am using only has Mumps 4.9. My laptop has 4.10 and has never run into this problem. I'll insist more that they upgrade Mumps, and hopefully that will solve the problem. Unfortunately using Parstokes is not an option, since we are using our out customized version of the Stokes solver.
Evan
Posts: 12
Joined: 12 Aug 2014, 14:42
Antispam: Yes

Re: Problems with large meshes

Post by Evan »

BTW, if anyone else runs into this problem, upgrading to Mumps 5.0 fixed it.
Post Reply