[COMSOL_Users] Out of memory during LU factorization

[P.C.]

Has anyone here faced this type of problem. Can anyone shine some
light on this for me?
Appreciate it, Paulo.

Below other users with the same problem...

Have you found the solution for this problem? I am having the same
error message for a simple problem.
Thanks!

Hello everyone,
I am solving 3D fluid problem using Navier Stokes with

multigrid. I

got an error message "Out of memory during LU factorization" for

DOF

greater than 20000. I also tried using k-e turbulence model and the
maximum DOF allowed is around 30000. Previously I used

PC/windows XP

home with 512MB RAM. Now I am Using 3.4Ghz Pentium D

processor/windows

XP Pro with 4GB RAM and still getting the same error message.
I wonder if there are limitations on the model I used (k-epsilon
turbulence and Navier Stokes) or on the software itself.
Anybody, please help!

[freelsjd32289]

Windows and/or Linux 32-bit is limited to about 2GB problem size. So,
even if you have a machine with more than 2GB (like your 4GB), you
cannot solve a bigger problem than the code will hold with 2GB.
The largest addressible memory in 32 bit is 2^32==4 GB.

So, first thing to do to solve large problems is get a 64-bit machine.
Sorry, that is life in 3D. With a 64-bit machine, the largest
problem you can solve becomes related to the largest amount of memory
you can install in the computer, which becomes how much money you want
to spend to solve large problems.

There are some things you can do to solve large problems on small
machines. You have already discovered this by the fact you are using
LU decomposition as a preconditioner for an iterative method. Direct
methods on 3D problems of any reasonable size become impractical very
quickly even on 64-bit machines. If you insist on using
LU-decomposition, you can control the memory used by adjusting the
drop tolerance. However, if he drop tolerance becomes too large, then
the solution can become unstable. Each problem varies. You have to
experiment to see.

There are much better iterative methods within COMSOL that are more
efficient with memory. If you are solving 3D NS, then the recommended
method discussed and shown how to do in the COMSOL documentation is to
use iterative methods, GMRES or FGMRES, with GMG as a preconditioner,
and Vanka as a pre- and post- smoother. In addition, you can use a
segregated solver to split the problem into two parts to save
additional memory. SSRS can be used as a smoother for the turbulence
part of the segregation. See the docs for examples and create a
problem to examine the default values of the parameters, and adjust
from there.

Post by P.C.
Has anyone here faced this type of problem. Can anyone shine some
light on this for me?
Appreciate it, Paulo.
Below other users with the same problem...

Have you found the solution for this problem? I am having the same
error message for a simple problem.
Thanks!

Hello everyone,
I am solving 3D fluid problem using Navier Stokes with

multigrid. I

got an error message "Out of memory during LU factorization" for

DOF

greater than 20000. I also tried using k-e turbulence model and the
maximum DOF allowed is around 30000. Previously I used

PC/windows XP

home with 512MB RAM. Now I am Using 3.4Ghz Pentium D

processor/windows

XP Pro with 4GB RAM and still getting the same error message.
I wonder if there are limitations on the model I used (k-epsilon
turbulence and Navier Stokes) or on the software itself.
Anybody, please help!

[P.C.]

Thank you, I'll revise this matter and see if I can fix it!

Post by freelsjd32289
Windows and/or Linux 32-bit is limited to about 2GB problem size. So,
even if you have a machine with more than 2GB (like your 4GB), you
cannot solve a bigger problem than the code will hold with 2GB.
The largest addressible memory in 32 bit is 2^32==4 GB.
So, first thing to do to solve large problems is get a 64-bit machine.
Sorry, that is life in 3D. With a 64-bit machine, the largest
problem you can solve becomes related to the largest amount of memory
you can install in the computer, which becomes how much money you want
to spend to solve large problems.
There are some things you can do to solve large problems on small
machines. You have already discovered this by the fact you are using
LU decomposition as a preconditioner for an iterative method. Direct
methods on 3D problems of any reasonable size become impractical very
quickly even on 64-bit machines. If you insist on using
LU-decomposition, you can control the memory used by adjusting the
drop tolerance. However, if he drop tolerance becomes too large, then
the solution can become unstable. Each problem varies. You have to
experiment to see.
There are much better iterative methods within COMSOL that are more
efficient with memory. If you are solving 3D NS, then the recommended
method discussed and shown how to do in the COMSOL documentation is to
use iterative methods, GMRES or FGMRES, with GMG as a preconditioner,
and Vanka as a pre- and post- smoother. In addition, you can use a
segregated solver to split the problem into two parts to save
additional memory. SSRS can be used as a smoother for the turbulence
part of the segregation. See the docs for examples and create a
problem to examine the default values of the parameters, and adjust
from there.

Post by P.C.
Has anyone here faced this type of problem. Can anyone shine some
light on this for me?
Appreciate it, Paulo.
Below other users with the same problem...

Have you found the solution for this problem? I am having the same
error message for a simple problem.
Thanks!

Hello everyone,
I am solving 3D fluid problem using Navier Stokes with

multigrid. I

got an error message "Out of memory during LU factorization" for

DOF

greater than 20000. I also tried using k-e turbulence model

and the

Post by P.C.

maximum DOF allowed is around 30000. Previously I used

PC/windows XP

home with 512MB RAM. Now I am Using 3.4Ghz Pentium D

processor/windows

XP Pro with 4GB RAM and still getting the same error message.
I wonder if there are limitations on the model I used

(k-epsilon

Post by freelsjd32289

Post by P.C.

turbulence and Navier Stokes) or on the software itself.
Anybody, please help!

[satish_mech01]

Try again by decreasing the number of elements and solver

Post by P.C.
Has anyone here faced this type of problem. Can anyone shine some
light on this for me?
Appreciate it, Paulo.
Below other users with the same problem...

Have you found the solution for this problem? I am having the same
error message for a simple problem.
Thanks!

Hello everyone,
I am solving 3D fluid problem using Navier Stokes with

multigrid. I

got an error message "Out of memory during LU factorization" for

DOF

greater than 20000. I also tried using k-e turbulence model and the
maximum DOF allowed is around 30000. Previously I used

PC/windows XP

home with 512MB RAM. Now I am Using 3.4Ghz Pentium D

processor/windows

XP Pro with 4GB RAM and still getting the same error message.
I wonder if there are limitations on the model I used (k-epsilon
turbulence and Navier Stokes) or on the software itself.
Anybody, please help!

[jaczpan_360]

If you are working on COMSOL 3.5, you can try to use the PARDISO out of core solver and increase the amount of memory that the solver can use from the HDD, this will make the time of the solution take longer but you shouldn't have that error again.

An aditional detail, when you are using turbulence module in a computer with a low amount of RAM try to don't use SPOOLES solver or make shure it is not being used by the automatic solver selection, this solver may be a little bit more accurate than others, but requires more resources from your computer, personally I use PARDISO solver in both segregated solver fields.

Post by satish_mech01
Try again by decreasing the number of elements and solver

Post by P.C.
Has anyone here faced this type of problem. Can anyone shine some
light on this for me?
Appreciate it, Paulo.
Below other users with the same problem...

Have you found the solution for this problem? I am having the same
error message for a simple problem.
Thanks!

Hello everyone,
I am solving 3D fluid problem using Navier Stokes with

multigrid. I

got an error message "Out of memory during LU factorization" for

DOF

greater than 20000. I also tried using k-e turbulence model and the
maximum DOF allowed is around 30000. Previously I used

PC/windows XP

home with 512MB RAM. Now I am Using 3.4Ghz Pentium D

processor/windows

XP Pro with 4GB RAM and still getting the same error message.
I wonder if there are limitations on the model I used (k-epsilon
turbulence and Navier Stokes) or on the software itself.
Anybody, please help!

[Pu ZHANG]

COMSOL is itself memory consuming. Besides optimization of the memory cost
of your model, you should use 64 bit operation system so that all memory
installed can be utilized.

Post by P.C.
Has anyone here faced this type of problem. Can anyone shine some
light on this for me?
Appreciate it, Paulo.
Below other users with the same problem...

Have you found the solution for this problem? I am having the same
error message for a simple problem.
Thanks!

Hello everyone,
I am solving 3D fluid problem using Navier Stokes with

multigrid. I

got an error message "Out of memory during LU factorization" for

DOF

greater than 20000. I also tried using k-e turbulence model and the
maximum DOF allowed is around 30000. Previously I used

PC/windows XP

home with 512MB RAM. Now I am Using 3.4Ghz Pentium D

processor/windows

XP Pro with 4GB RAM and still getting the same error message.
I wonder if there are limitations on the model I used (k-epsilon
turbulence and Navier Stokes) or on the software itself.
Anybody, please help!

--
Pu, ZHANG 匠朎
Centre for Optical and Electromagnetic Research,
East Building No.5, Zijingang Campus,
Zhejiang University, Hangzhou 310058, China
E-mail: ***@coer.zju.edu.cn
***@gmail.com
Tel: [phone]
Fax: [phone]
Website of our center: http://coer.zju.edu.cn

[avin_mech]

Hi,

The best thing is to reduce the number of mesh elemnts. If its not acceptable, read further.....

The problen u r facing is because ur solver is in need of more memory from ur computer.

U can do this my increasing the size of ur virtual memory. Google abt this and U will gt to know how to do that. It pretty easy....

The other thing U can do is change ur solver itself. If U r using a direct solver, it takes a lot of memory. For N-S equations use FGMRES with Geometric Multi grid. Also in the Solver parameters > Advanced tab, check the option "Store solution on file"....

Hope this helps....

Post by P.C.
Has anyone here faced this type of problem. Can anyone shine some
light on this for me?
Appreciate it, Paulo.
Below other users with the same problem...

Have you found the solution for this problem? I am having the same
error message for a simple problem.
Thanks!

Hello everyone,
I am solving 3D fluid problem using Navier Stokes with

multigrid. I

got an error message "Out of memory during LU factorization" for

DOF

greater than 20000. I also tried using k-e turbulence model and the
maximum DOF allowed is around 30000. Previously I used

PC/windows XP

home with 512MB RAM. Now I am Using 3.4Ghz Pentium D

processor/windows

XP Pro with 4GB RAM and still getting the same error message.
I wonder if there are limitations on the model I used (k-epsilon
turbulence and Navier Stokes) or on the software itself.
Anybody, please help!