Discussion Closed This discussion was created more than 6 months ago and has been closed. To start a new discussion with a link back to this one, click here.

Editing in core memory

Please login with a confirmed email address before reporting spam

I am having a memory issue running a 3d Navier Stokes Incompressible flow model on my 32 bit Linux machine. It has 2 processors and 4 GB RAM. The COMSOL users guide claims that you can edit the amount of in core memory for LU factors using the "In core memory edit field". The default value is 512 MB. I can't find anything resembling an "in core memory edit field" for the life of me. I'm wondering if this may fix my memory problem. Does anyone have any suggestions?

2 Replies Last Post 30 déc. 2009, 22:44 UTC−5
Ivar KJELBERG COMSOL Multiphysics(r) fan, retired, former "Senior Expert" at CSEM SA (CH)

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 30 déc. 2009, 17:21 UTC−5
Hi
have you read through the COMSOL Installation and Operations Guide ?

personally I prefer to use the client server modus when my models get close to the RAM limit, you gain a little as you separate the pre/postprocessing to the solver part, and one can be swapped out while the other is running, this will also change the heap sizes by default

God luck
Ivar
Hi have you read through the COMSOL Installation and Operations Guide ? personally I prefer to use the client server modus when my models get close to the RAM limit, you gain a little as you separate the pre/postprocessing to the solver part, and one can be swapped out while the other is running, this will also change the heap sizes by default God luck Ivar

Jim Freels mechanical side of nuclear engineering, multiphysics analysis, COMSOL specialist

Please login with a confirmed email address before reporting spam

Posted: 1 decade ago 30 déc. 2009, 22:44 UTC−5
Toby, I also started with 32-bit linux for COMSOL (actually, FEMLAB at the time. I have been a linux user since 1994, and a COMSOL user since FEMLAB 3.0). What I learned is that, similar to Windows but a bit better than Windows actually, 2GB is about as much as you can stuff "in-core" for a single application. You can tweak the java heap size and it will help, but eventually you will overwhelm your computer. It is not a limitation of COMSOL, but of the 32-bit operating system (2^31 ~ 2x10^9). For a 3D NS problem (7 variables per node if turbulent), it will not be difficult to create a mesh (think of a cube 100^3) that will be too large for a 32-bit machine. Even with iterative solvers and/or disk caching, a problem such as this is too large for a 32-bit computer IMHO. Therefore, I recommend 64-bit machines and operating system (Linux or Windows) for 3D NS problems. Further, I would put as much memory as you can afford in that machine (I started with 16GB and now run 64GB per node). Machines are relatively inexpensive these days, and memory runs about $50/GB or less. Last benchmark I ran was about 5x10^6 dof problem, 40 GB memory used, GMG iterative solver, and about 10 days solution time on 8-core machine. This was a highly-nonlinear, tightly-coupled problem.
Toby, I also started with 32-bit linux for COMSOL (actually, FEMLAB at the time. I have been a linux user since 1994, and a COMSOL user since FEMLAB 3.0). What I learned is that, similar to Windows but a bit better than Windows actually, 2GB is about as much as you can stuff "in-core" for a single application. You can tweak the java heap size and it will help, but eventually you will overwhelm your computer. It is not a limitation of COMSOL, but of the 32-bit operating system (2^31 ~ 2x10^9). For a 3D NS problem (7 variables per node if turbulent), it will not be difficult to create a mesh (think of a cube 100^3) that will be too large for a 32-bit machine. Even with iterative solvers and/or disk caching, a problem such as this is too large for a 32-bit computer IMHO. Therefore, I recommend 64-bit machines and operating system (Linux or Windows) for 3D NS problems. Further, I would put as much memory as you can afford in that machine (I started with 16GB and now run 64GB per node). Machines are relatively inexpensive these days, and memory runs about $50/GB or less. Last benchmark I ran was about 5x10^6 dof problem, 40 GB memory used, GMG iterative solver, and about 10 days solution time on 8-core machine. This was a highly-nonlinear, tightly-coupled problem.

Note that while COMSOL employees may participate in the discussion forum, COMSOL® software users who are on-subscription should submit their questions via the Support Center for a more comprehensive response from the Technical Support team.