Problem with using PyAns linear solver

[HazemEldahshan]

Hi,

I am using the linear algebra functionality of the PyAns/PyMAPDL library to perform some computations. The workflow is as follows: I first retrieve the stiffness matrix, factorize it, and then perform a solve step, as shown below:

full = pymapdl_reader.read_binary(inputFile)
dofref, k0, m = full.load_km()
mm = mapdl.math
factorizedStiff = mm.factorize(k)

mm.set_vec(rhs, "rhsAnsMath")
rhsAnsMath = mm.vec(name="rhsAnsMath")
t = factorizedtiffnessMatrix.solve(rhsAnsMath)
This workflow runs correctly for relatively small models. However, when I use a large model with about 3 million DOFs, the computation stops at the solve step (the factorization appears to complete successfully), and I obtain the following error message in the log file (identical for all processes):

====================================================================================
BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES
RANK 0 PID 11700 RUNNING AT WL0036338
EXIT STATUS: -1 (ffffffff)

I suspect this may be related to a memory overflow or resource limitation, especially since similar calculations run without issues in Mechanical. Could you please advise how to resolve this problem or how to configure the solver/PyMAPDL to handle such large models?

Best regards,
Hazem

[germa89]

Hi @HazemEldahshan

You can try to start an MAPDL instance with more resources:

from ansys.mapdl.core import launch_mapdl()
mapdl = launch_mapdl(ram=10*1024, nproc=8) 

full = pymapdl_reader.read_binary(inputFile)
dofref, k0, m = full.load_km()
mm = mapdl.math
factorizedStiff = mm.factorize(k)

mm.set_vec(rhs, "rhsAnsMath")
rhsAnsMath = mm.vec(name="rhsAnsMath")
t = factorizedtiffnessMatrix.solve(rhsAnsMath)

MAPDL normally allocates as much memory as possible and need, but sudden increases in memory usage might for the OS to kill the processes.

I would run for 2 million and then check the memory footprint.