optimization of Scanning Helmholtz Solver

[Sergejs]

To me that seems just paraview artefact of volume rendering. I seldom use that myself. I guess cuts are ok?

You are probably right - hopefully this is only visual problem in ParaView. It is quite misleading if you are not explicitly aware of where this pattern comes from. Perhaps there is a way to pre-merge the partial results so that ParaView does not need to run into this artefact?

From here: https://computational-acoustics.gitlab. ... lelisation
"note that the boundaries between partitions will probably be visible depending on the postprocessing you perform".

[Sergejs]

Hi

To split the analysis try this:

Code: Select all

$npart=5
$dn=10
$n0=(npart-1)*dn
Simulation 
...
  Simulation type = "scanning"
  timestep intervals = $dn
  Post File = "case.vtu"
  vtu: fileindex offset = $n0
...
Frequency = Variable "time"
  Real MATC "f0*1.1^(n0+tx-1)"

Solver n
  Procedure = "SaveData" "SaveScalars"
  Filename = f$npart$.dat
  ...

Alternating value of $npart you can run different part of the frequency sweep and catenate results
cat f*.dat > f.dat

-Peter

This code looks extremely useful - just what I need for writing a scheduler (based on "NumCalcManager.py") - will make initial tests tomorrow night.

[kevinarden]

Perhaps there is a way to pre-merge the partial results so that ParaView does not need to run into this artefact?

When I run parallel I get a vtu file for each partition and a combined one for the whole model

[raback]

Hi,

The issue is that each MPI process is working with its own piece. The nice feature of paraview is that it lets us write each of them separately and just provide the wrapper file (.pvtu). Then there is no communication needed in postprocessing. Imagine writinig everything to a single file there should be one process taking care of the writing.

So to me this is something that paraview should do - and does so for most visualizations. Maybe this is not made easy for paraview since each separate .vtu file uses local numbering. So the information could only be traced by searching nodes that have exactlty same coordinates.

-Peter

[Sergejs]

Hi

To split the analysis try this:

Code: Select all

$npart=5
$dn=10
$n0=(npart-1)*dn
Simulation 
...
  Simulation type = "scanning"
  timestep intervals = $dn
  Post File = "case.vtu"
  vtu: fileindex offset = $n0
...
Frequency = Variable "time"
  Real MATC "f0*1.1^(n0+tx-1)"

Solver n
  Procedure = "SaveData" "SaveScalars"
  Filename = f$npart$.dat
  ...

Alternating value of $npart you can run different part of the frequency sweep and catenate results
cat f*.dat > f.dat

-Peter

After some testing, Frequency data export/import is working great using ParaView to read in the group of individual files written by

Code: Select all

Filename = f$npart$.dat

But the command

Code: Select all

cat f*.dat > f.dat

can not be used unless filename will contain leading zeros (like .vtu files) - the numbering order during cat is not correct. And I could not figure out how to do this inside .sif file

Now working on a Python scheduler...

[Sergejs]

My Python scheduler is ready.

As it works for Scanning simulations with input frequency value I assume it is only useful for audio simulations and I called it

Code: Select all

ElmerScanManager.py

If there are other use cases I could still change the name of it

At the moment there is no point in sharing it as without the rest of the example files and some tutorial it makes no sense.
But thank you for the example code - it was what I needed to get this project to this level.

Will share the links to tutorial with scheduler when it is available. (may be useful to beginners like me and hopefully people will find bugs and improvement suggestions in it)