#1 Parallelization
I
went through the mailing list wherein it has been mentioned that SfePy can
support multicore via numpy/scipy multicore support, but compute cluster
support is not available which requires knowledge of MPI. So we can add
computing cluster support where jobs need to communicate with each other and
exploit the high performance computing in order to make it scalable. In Python
it can be done using mpi4py module.
#2 Pre-processing and Post-Processing combined with SfePy
One
of the reason why people use proprietary software is their ease to use. We can
build complete simulation platform with powerful frontend for pre-processing to
analysis to post-processing. This will include the script support wherein
people can write scripts to post process the data on the platform itself. Also
it will be made modular so as to make it extensible. We can look into integrating
it with CAD packages like PythonCAD or built our own pre-processor or
postprocessor using powerful GUI toolkit to provide the complete simulation solution.
#3
Incorporating
Phase changing material simulation which has never been done in any simulation software
package before. Since it is my research area I will develop its model based on
various research paper and my own research and encorporate it into SfePy.
Hi,
we are trying to sign up under the PSF umbrella for this year's Google Summer
of Code because of an e-mail from Ankit Mahato, who expressed interest to help
developing SfePy as his GSoC project this summer.
So let us discuss possible project ideas here. I will post results of the
discussion to [1].
Ankit's ideas are (my summary):
#1 parallelization - cluster support using mpi4py
#2 pre- and post-processing GUI frontend
#3 incorporating phase changing materials (his research area)
Ankit, could you post full text your ideas into this thread? The pdf you sent
me does not allow selecting text.
My comments:
For me, #1 is something that I was planning to do "soon" anyway as I am going
to need it for my research work - a help would come really handy, but we will
have to think carefully about the implementation. I think I prefer having a
parallel layer above the current serial FEM, so that the current code can stay
as it is, unaware that it runs in parallel. I am not sure yet how difficult it
is going to be, but it won't be trivial.
#2 would be nice, but IMHO it is not so important as having a solid and
reasonably fast FEM core.
#3 would IMHO be the most useful for Ankit, and a nice addition to modelling
capabilities of SfePy.
Other possible topics can be found in our issues list ("enhancement" label).
IMHO it would be good to prospective student(s) to try tackling some of the
issues listed below to get acquainted with SfePy code before the GSoC starts:
#196 Document properly term evaluation modes and postprocessing/probing.
#195 describe how to add Neumann BC in a diffusion example and tutorial
(tutorial part done by Alec)
#167 improve gallery page
#164 Python 3 compatibility
#154 automatic testing of terms
#140 test schroedinger.py
#133 Provide examples for SfePy Terms
Implementing the other enhancements would be, of course, also very useful, but
those IMHO too difficult for someone trying to learn the code. They are
certainly quite difficult for me, as they are not done yet =:) (shell elements!)
Cheers,
r.
[1] http://sfepy.org/doc-devel/development.html