Note: This discussion is about an older version of the COMSOL Multiphysics® software. The information provided may be out of date.
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.
Cluster Computing - Solvers and Memory
Posted 23 janv. 2012, 10:32 UTC−5 Studies & Solvers Version 4.2a 1 Reply
Please login with a confirmed email address before reporting spam
I encountered two problems when trying to run a job off our PBS run linux cluster (using command line approach).
1) The first is regarding the solver. I got the following error:
Warning: PARDISO is not distributed. Switching to MUMPS.
How can I switch the PARDISO solved to distributed? Note that the mesh is very coarse and thus an iterative solver will not be able to find a solution (based on my investigation in a smaller subset of the geometry)
2) My second problem is that I tend to always get one of two memory related messages:
a) Out of memory LU factorization
b) Out of memory in Stationary Solver
Is this related to the RAM? Any advice considering I am stuck using a direct solver at the moment?
Here is the relevant section of the script file submitted:
nn=$(wc -l < $PBS_NODEFILE)
echo "--- mpd BOOT"
comsol -nn ${nn} mpd boot -f $PBS_NODEFILE -mpirsh ssh
echo "--- mpd TRACE"
comsol mpd trace
echo "--- parallel COMSOL RUN"
comsol -nn ${nn} batch -inputfile disttubecoarsepbs2.mph -outputfile disttubecoarsepbs2out.mph -batchlog batch_jan23.log
echo "--- mpd ALLEXIT"
comsol mpd allexit
echo "--- Job finished"
1) The first is regarding the solver. I got the following error:
Warning: PARDISO is not distributed. Switching to MUMPS.
How can I switch the PARDISO solved to distributed? Note that the mesh is very coarse and thus an iterative solver will not be able to find a solution (based on my investigation in a smaller subset of the geometry)
2) My second problem is that I tend to always get one of two memory related messages:
a) Out of memory LU factorization
b) Out of memory in Stationary Solver
Is this related to the RAM? Any advice considering I am stuck using a direct solver at the moment?
Here is the relevant section of the script file submitted:
nn=$(wc -l < $PBS_NODEFILE)
echo "--- mpd BOOT"
comsol -nn ${nn} mpd boot -f $PBS_NODEFILE -mpirsh ssh
echo "--- mpd TRACE"
comsol mpd trace
echo "--- parallel COMSOL RUN"
comsol -nn ${nn} batch -inputfile disttubecoarsepbs2.mph -outputfile disttubecoarsepbs2out.mph -batchlog batch_jan23.log
echo "--- mpd ALLEXIT"
comsol mpd allexit
echo "--- Job finished"
1 Reply Last Post 23 janv. 2012, 21:23 UTC−5
Please login with a confirmed email address before reporting spam
Posted:
1 decade ago
MUMPS is the direct solver that works best in distributed parallel processing mode. PARDISO does not work at all, and SPOOLES is much slower, but does use less memory. PARDISO only runs shared memory parallel.
If you accidently forget to use MUMPS while in distributed parallel mode and still have PARDISO set, COMSOL will automatically switch to MUMPS for you.
If you still run out of memory, you can use more compute nodes to reduce the amount of memory required per node, or you can change your model to use less memory.
If you use iterative methods, it will use far less memory, but run slower. It will run distributed parallel.
It is a tradeoff.
If you accidently forget to use MUMPS while in distributed parallel mode and still have PARDISO set, COMSOL will automatically switch to MUMPS for you.
If you still run out of memory, you can use more compute nodes to reduce the amount of memory required per node, or you can change your model to use less memory.
If you use iterative methods, it will use far less memory, but run slower. It will run distributed parallel.
It is a tradeoff.