clawpack / pyclaw Goto Github PK

File writing on Shaheen is very slow at present. This will help speed it up.

Benchmark test case still need to be defined. We are looking for it. It should be defined in ONE or MAXIMUM TWO DAYS. Based on the selected test case the next steps should be defined.

Right now, this only happens if the Controller is used. So the code will fail now unless one uses the Controller.

I don't think it will work as coded right now. It may not be worth supporting this combination anyway, but if not we should check for it and throw a helpful exception.

Setting the aux array is somewhat different from setting q because boundary values must also be set on initialization. Right now the user is required to create a numpy array of the appropriate size and fill it in. Aron suggested providing a simpler interface that would only require the user to provide a function setaux(x, y, dx, dy) that takes arrays of cell centers and dimensions and returns an array of aux values for the cells. Then the grid would have a method that applies this function automatically using the coordinates of the grid. It was agreed that this would be helpful and should be supported, but without removing support for the current approach. The grid could also have a method to impose common "boundary conditions" to generate the ghost cell aux values (zero extrapolation and periodic).

The user will be able to provide a function that accepts q and returns a vector of derived quantities. These two vectors are not necessarily the same size. As a simple example, the user may wish to only output one component of q.
In order to implement this efficiently for parallel runs, it is necessary to create another DA for the derived quantities. The user will be able to specify the frequency of writing both output and checkpoint files separately (the checkpoint files will contain just q).

The tutorial in the docs has not been updated to reflect how PyClaw works now.

This way we don't need commmunicate_CFL in PyClaw.

an unverified version of the test currently exists at: https://bitbucket.org/ahmadia/scipy_2011_pyclaw/src/d89be5055bb6/papers/petclaw_walla/experiment/SerialComparison/

The Makefiles are currently polluted with PETSc environment variables. We really need to clean this up before release.

FAILED (errors=1)
[0]PETSC ERROR: ------------------------------------------------------------------------
[0]PETSC ERROR: Caught signal number 11 SEGV: Segmentation Violation, probably memory access out of range
[0]PETSC ERROR: Try option -start_in_debugger or -on_error_attach_debugger
[0]PETSC ERROR: or see http://www.mcs.anl.gov/petsc/petsc-as/documentation/faq.html#valgrind[0]PETSC ERROR: or try http://valgrind.org on GNU/linux and Apple Mac OS X to find memory corruption errors
[0]PETSC ERROR: likely location of problem given in stack below
[0]PETSC ERROR: --------------------- Stack Frames ------------------------------------
[0]PETSC ERROR: Note: The EXACT line numbers in the stack are not available,
[0]PETSC ERROR: INSTEAD the line number of the start of the function
[0]PETSC ERROR: is given.
application called MPI_Abort(MPI_COMM_WORLD, 59) - process 0
[unset]: aborting job:
application called MPI_Abort(MPI_COMM_WORLD, 59) - process 0

This is not a pyclaw/petclaw issue but it affects us....

Manuel is implementing this and will send a patch to David.

SharpClaw is working on Shaheen, but classic is not.

Perhaps some kinds of objects are persisting from one test to the next?

Show how to create your own pyclaw/riemann repositories on Shaheen that use the library code installed in /project/k47

We have been bitten by this 3 times now. @Ahamadia, what is the best way to do this? Can we check the architecture we're on from inside Python? Or we could move the logging configuration files around in our setup scripts on Shaheen.

We should never need to communicate aux, since every processor knows how to fill its own ghost cells. But it seems very difficult to find a good way to code this without calling globalToLocal.

The current setup is very fragile and easy to break. The right solution is probably to move q and aux to a "state" object, out of the grid.

This may be challenging, since the 3D code has not been updated for some time. I think the most recent version is here:

http://www.amath.washington.edu/~claw/claw/clawpack/3d/lib/

This can give us:

-Automated building and testing
-Automated creation and uploading of docs
-More?

Possible tools:

-Buildbot
-Jenkins
-Others?

Possibilities:

clawpack.github.com/pyclaw

pyclaw.readthedocs.org

kaust.edu.sa/davidketcheson/pyclaw

Others?

Example usage:

state.q_names = ['density','momentum','energy']

state.q.density[i,j] # gives state.q[0,i,j]
state.q.momentum[i,j] # gives state.q[1,i,j]

These work in PetClaw but have not been tested in PyClaw.

Right now we only support the dimensionally-split 2D algorithms from Clawpack.

We need to expand our testing to cover some other common setups such as Ubuntu without gfortran and decide whether we can support them or not.

this makes parallel tests harder to debug, we need to figure out why this is happening and fix the test framework.

Most settings of the following work in theory but have not been tested:

-char_decomp
-lim_type

These are implemented in PetClaw but not yet in PyClaw.

These should be unified across all examples

Average q over blocks and then output, to speed up output and reduce file sizes in large runs.

We should be able to test for instance just the following:

-Pure Python tests only (no Fortran)
-PyClaw (no PetClaw)
-PetClaw with implicit timestepping
-Classic only
-SharpClaw only

I have added aux_global to the list of things that are pickled in petsc.py. It's not as clear where to put this in ascii.py.

It is supported by PETSc. This is also used in AMR.

Files still get automatically overwritten if this is set to False. Also, we should reconsider the default behavior.

In GeoClaw, one can set up gauges at particular locations, which cause the solution at that location to be written to a file at every timestep.

Remove mpi4py doc
Add installation petsc 64 bit