Closed robertsawko closed 10 months ago
This is on the stable/archlinux branch that I maintain, so I'd like to merge this. The test failures are due to a mismatch with the hashdist version. I'll look into whether I can fix that.
Hi Chris,
This is really just a quick question rather than an issue. Is there a way to invoke "calculateSolution" in parallel from within python? You already pointed me to a relevant, I think, test file which is
tests/ci/test_meshPartitionFromTetgenFiles.py
I am working with it now and trying to adjust my serial cases for instance this one:
https://github.com/robertsawko/pde-and-uq/blob/master/riemann/mesh_and_discretisation_test.py
Could you give me a description of what needs to be done to make it work in parallel?
Some features I spotted are: * Comm module is imported in mesh partition example * There's a parallel switch in numerical settings and it seems to be forcing flux and solver settings e.g. nLayersOfOverlapForParallel parallelPartitioningType, linearSolverConvergenceTest
At the end of the day I would like to run
mpirexec -np 10 python mesh_and_discretisation_test.py
in order to accelerate calculations.
Please let me know,
Not everything that's true can be proven so! http://en.wikipedia.org/wiki/Godel%27s_incompleteness_theorems
Hi Robert,
Yes, no problem. I'm going to switch this conversation over to the proteus issue tracker and will CC you on that.
Thanks, Chris
On Wed, Mar 30, 2016 at 8:33 AM, Robert Sawko notifications@github.com wrote:
Hi Chris,
This is really just a quick question rather than an issue. Is there a way to invoke "calculateSolution" in parallel from within python? You already pointed me to a relevant, I think, test file which is
tests/ci/test_meshPartitionFromTetgenFiles.py
I am working with it now and trying to adjust my serial cases for instance this one:
https://github.com/robertsawko/pde-and-uq/blob/master/riemann/mesh_and_discretisation_test.py
Could you give me a description of what needs to be done to make it work in parallel?
Some features I spotted are:
- Comm module is imported in mesh partition example
- There's a parallel switch in numerical settings and it seems to be forcing flux and solver settings e.g. nLayersOfOverlapForParallel parallelPartitioningType, linearSolverConvergenceTest
At the end of the day I would like to run
mpirexec -np 10 python mesh_and_discretisation_test.py
in order to accelerate calculations.
Please let me know,
Robert
Not everything that's true can be proven so! http://en.wikipedia.org/wiki/Godel%27s_incompleteness_theorems
— You are receiving this because you commented. Reply to this email directly or view it on GitHub https://github.com/hashdist/hashstack/pull/926#issuecomment-203434763
The symbolic link created by the fix_include step was hard-coded and pointing to a different version. Also, version 2.1 was breaking grib compilation.