The comparison of results: serial and parallel (elasticity)

[tomx]

Hi,

I am testing Elmer code for linear elastic case. I am using simple beam model which contains 24 nodes and 30 tetrahedral elements (1st order). All 4 nodes at one side are fixed (by dirichlet conditions) and on oposite side there is force attached to two seperate nodes. For serial run of Elemer I have obtained results as in reference. So I have run this case parallel and I achieved diffrent results. For various number of domains the different displacement results are observed. For instance, at 24 node, the Y displacement for serial Elmer is equal -5,070E-06, for 2 domains run -7,523E-06, for 3 domains run -9,738E-06 and for 4 domains run -1,191E-05. This is simple case and I haven't expected such problem especially that in the near future I want to test more complex cases. Did someone has faced with that issue?

Also I have marked to write in results file the displacements load (Calculate Loads = True in Solver section) and I have noticed that the loads at shared nodes have same direction. Does the direction of force in adjacent domains should have the inverse sense in order to balance force between domains? The value of these forces is equal to submitted load.


Tom

[Juha]

Hi

Generally this should work out of box, (if you get convergece). if you would care to share your case, i'll have a look?

Regards, Juha

[tomx]

Hi again & sorry for late response.
I have attached the configuration (sif) file and the log file.
I hope that it would help you. If you need any other file or information, please let me know. I will send it to you via this post or private message.

Tom

[Juha]

Hi,
I'm sorry that i have not followed up on this earlier...
Could you also attach the mesh?
Regards,Juha

[tomx]

Here you have mesh files

Tom

[Juha]

Hi,

this is now fixed in the SVN. The trouble was that each partition added the 'Displacement n Load'
to a local node closest to given "Target Coordinates".

Thanks for the report!

Regards, Juha

[tomx]

Hi,
I have run priviausly case with new verision of Elmer and I have obtained following message:

Code: Select all

ELMER SOLVER (v 7.0) STARTED AT: 2012/11/14 13:02:27
ELMER SOLVER (v 7.0) STARTED AT: 2012/11/14 13:02:28
ParCommInit:  Initialize #PEs:            2
MAIN: 
MAIN: =============================================================
MAIN: ElmerSolver finite element software, Welcome!
MAIN: This program is free software licensed under (L)GPL
MAIN: Copyright 1st April 1995 - , CSC - IT Center for Science Ltd.
MAIN: Webpage http://www.csc.fi/elmer, Email elmeradm@csc.fi
MAIN: Library version: 7.0 (Rev: 5917)
MAIN:  Running in parallel using 2 tasks.
MAIN: =============================================================
MAIN: 
MAIN: 
MAIN: -------------------------------------
MAIN: Reading Model: beam_from_cp-case3-4.sif
Model Input:  Unlisted keyword: [displacement 1 load] in section: [boundary condition 5]
Model Input:  Unlisted keyword: [displacement 2 load] in section: [boundary condition 5]
Model Input:  Unlisted keyword: [displacement 3 load] in section: [boundary condition 5]
Model Input:  Unlisted keyword: [displacement 1 load] in section: [boundary condition 6]
Model Input:  Unlisted keyword: [displacement 2 load] in section: [boundary condition 6]
Model Input:  Unlisted keyword: [displacement 3 load] in section: [boundary condition 6]
Loading user function library: [ElasticSolve]...[ElasticSolver_Init0]
MAIN: -------------------------------------
Loading user function library: [ElasticSolve]...[ElasticSolver_Init]
Loading user function library: [ElasticSolve]...[ElasticSolver]
OptimizeBandwidth: ---------------------------------------------------------
OptimizeBandwidth: Computing matrix structure for: elasticsolver...done.
OptimizeBandwidth:  Half bandwidth without optimization:            8
OptimizeBandwidth: 
OptimizeBandwidth: Bandwidth Optimization ...done.
OptimizeBandwidth:  Half bandwidth after optimization:            5
OptimizeBandwidth: ---------------------------------------------------------
MAIN: 
MAIN: -------------------------------------
MAIN:  Steady state iteration:            1
MAIN: -------------------------------------
MAIN: 
ElasticSolve: 
ElasticSolve: 
ElasticSolve: -------------------------------------
ElasticSolve:  ELASTICITY ITERATION              1
ElasticSolve: -------------------------------------
ElasticSolve: 
ElasticSolve: Starting assembly...
ElasticSolve: Assembly done
DefUtils::DefaultDirichletBCs: Setting Dirichlet boundary conditions
SetNodalLoads: Checking loads for: displacement 1
WARNING:: ListFind: 
WARNING:: ListFind:  Requested property: [Target Nodes], not found
WARNING:: ListFind: 
WARNING:: ListFind: 
WARNING:: ListFind:  Requested property: [Target Nodes], not found
WARNING:: ListFind: 
SetNodalLoads: Checking loads for: displacement 2
WARNING:: ListFind: 
WARNING:: ListFind:  Requested property: [Target Nodes], not found
WARNING:: ListFind: 

And when I am changing Target Coordinates to Target Nodes, the solution is equal 0.


Tom

[Juha]

Hi,

the ElmerSolver should add the "Target Nodes" internally when "Target Coordinates"
are used. Seems to work for me... Don't really have a clue what this might be about.

Anyway, I made some more corrections after my first reply, could you still try updating your
sources and re-install ?

Regards, Juha

[tomx]

Hi,
ok, when I have run Elmer (5921) with first case, there was no last issue but I haven't obtained the convergence. When I have changed the constrains from Target Coodinates to Target Nodes the computations has been performed but the results are the same as with first version (different from serial run of Elmer).

The solver input files and log files are attached.

Tom

[Juha]

Hi

could you try changing the preconditioning from ILU2 -> ILU0 and
rerunning? I had similiar trouble with two partitions. I don't think
the "Target Nodes" thing is even supposed to work in parallel.

Juha