Hi Kedi,

Yes, the solution process caused your issues.

Note that you are using the approximation order 2 in your field, so the linear system has not just 15242 active DOFs, which is OK for the direct solver you use, but it has 113933 active DOFs. I tried to use the MUMPS solver, and it required about 3.5 GB:

sfepy: nls: iter: 0, residual: 5.867625e+04 (rel: 1.000000e+00) sfepy: residual: 0.05 [s] sfepy: matrix: 2.28 [s] sfepy: solve: 14.80 [s] sfepy: warning: linear system solution precision is lower sfepy: then the value set in solver options! (err = 1.537797e-09 < 1.000000e-10) sfepy: nls: iter: 1, residual: 1.396787e-09 (rel: 2.380498e-14)

With the iterative PETSc Krylov solver, the solution can be obtained with peak memory requirement of about 800 MB:

from sfepy.base.base import Struct from sfepy.solvers import Solver

ls = Solver.any_from_conf(Struct(**{ 'kind' : 'ls.petsc', 'method': 'cg', 'precond': 'icc', 'i_max' : 200 }))

sfepy: nls: iter: 0, residual: 5.867625e+04 (rel: 1.000000e+00) sfepy: residual: 0.05 [s] sfepy: matrix: 2.28 [s] sfepy: solve: 17.64 [s] sfepy: warning: linear system solution precision is lower sfepy: then the value set in solver options! (err = 3.613957e+00 < 1.000000e-10) sfepy: nls: iter: 1, residual: 3.613957e+00 (rel: 6.159148e-05)

-> try using an iterative solver (and a suitable preconditioner - the 'icc' used above is not very good, but I am not expert in this to give a better advice).

BTW. for the approximation order 3, 375578 DOFs requires about 4.5 GB with PETSc CG/ICC. and 17.5 GB with MUMPS (both took about 1.5 minutes, but the direct solver was much more precise - note that the iterative solver had limited max. number of iterations to 200.)

Last note: the integral order should be 2 * approximation order (for constant material data).

Best regards, r.

On 15. 10. 20 3:18, Yang, Kedi wrote:

Hi guys,

Currently, I am using Sfepy to simulate a 3D geometry model with a uniaxial displacement to find out the von Mises strength and the total execution time for the simulation is about *40 mins with 8G RAM*. Besides, 94% of the RAM is occupied when the program is running. I am not sure what is causing this execution time to take that long and what I could do to shorten the execution time. The geometry model was inputted from an Abaqus input-file with 5514 vertices and 4417 cells. Also, I have checked the execution time line-by-line and found that the most time consumable step is defined as: state = pb.solve(status=status) In this case, I am supposing the solving process used most of the RAM and causing the timing issue.

The mesh file and our code is attached.

Thank you and best regards!

Kedi Yang

---

SfePy mailing list -- sfepy@python.org To unsubscribe send an email to sfepy-leave@python.org https://mail.python.org/mailman3/lists/sfepy.python.org/ Member address: cimrman3@ntc.zcu.cz