Thermal core taking up more time than Stress Balance or Mass Transport

wadewei

Running a MOLHO model with 13 vertical layers and iterative solver using 40 cores. Is this normal? Thank you!

Wade

Core solution profiling
elapsed time : 20417.5 Seconds
number of flops : 2.01042e+11 Flops
memory used : 9.88856e+07 Bytes

Individual core profiling
Thermal core elapsed time: 9806.72 sec
Stress balance core elapsed time: 7719 sec
Mass transport core elapsed time: 2145.98 sec
SMB core elapsed time: 24.4233 sec

md.toolkits

ans =

List of toolkits options per analysis:

DefaultAnalysis:
toolkit: 'petsc'
mat_type: 'mpiaij'
ksp_type: 'fgmres'
pc_type: 'bjacobi'

RecoveryAnalysis:
toolkit: 'petsc'
mat_type: 'mpiaij'
ksp_type: 'preonly'
pc_type: 'lu'
pc_factor_mat_solver_type: 'mumps'
mat_mumps_icntl_14: 120
mat_mumps_icntl_28: 2
mat_mumps_icntl_29: 2

mathieumorlighem

you are funny 🙂 There is no information about the number of elements, or how long your model is running for (1 day or 100 years?). The output you are showing looks good to me though.

wadewei

mathieumorlighem
My bad. There are 3.6 million elements in total. A total of 1000 steps took 20000 seconds, i.e. each step needs 20 seconds with 40 cores.
My main concern is that the thermal core spent more time than stress balance calculation -- I thought the non-linear MOLHO calculation would be most costly. Perhaps there's a better-suited iterative solver for thermal solution?

mathieumorlighem

oh ok that makes sense. MOLHO is a depth-integrated model so it is actually pretty cheap. It is not surprising that the enthalpy model is the bottleneck in your case.