[SciPy-User] ANN: SOLVCON 0.1

Yung-Yu Chen yyc@solvcon....
Fri Aug 12 06:20:44 CDT 2011


Hello,

On Fri, Aug 12, 2011 at 01:49, Klonuo Umom <klonuo@gmail.com> wrote:
> Interesting package. Congratulations on making milestone release
>
> I installed it and on a first look I can't see it workflow. On web
> portal I found tip to follow examples that come with this package, but
> those aren't trivial at all; I mean lot of classes and functions comes
> from nowhare and it's like no walkthrough provided
>

We have not put efforts to make the package user friendly.  SOLVCON
began with the idea to provide a framework to collect important
supportive functionalities needed by CFD codes, to enhance the
robustness and coding efficiency.  The proof of concept turned out to
be successful, and we realized that SOLVCON has great potentials to
facilitate a new category of practices of building high-performance
conservation-law solvers for high-fidelity solutions.

In the time being, SOLVCON is made for experts in computational
science.  In the foreseeable future, our collaborators and we will
make it more accurate, more scalable, and more versatile.  You can
find the plan for the forthcoming development at
http://solvcon.net/yyc/writing/2011/solvcon_0.1.html .  We hope
SOLVCON can be used to renew the technology used in high-end
calculations of PDEs, e.g., CFD, computational electromagnetism, etc.

> If I may explain myself, I got summer seminar assigment, starting from
> shallow water eqs to derive eq of absolute vorticity for nondivergent
> flow in linearized form using perturbation method. I've done it by
> hand, but would like to understand the process with some of Python
> packages if feasible, and as staring eqs are hyperbolic PDEs maybe I
> could use this package although it uses different method for solving.
> Is it good idea to try to use this package, and if answer is yes, can
> you maybe provide some starting point for this simple task?
>
>

The package was geared up for large-scale, complex calculations.
Using SOLVCON for very simple calculations would be an overkill.
Unfortunately, the short answer to your question could be no.

When dealing with multi-physics, we take the approach to model the
underlying numerical algorithm, mathematics and physics as much as
possible.  The approach looks complicated at the first glance, but is
actually concise from theories to implementations.  We believe the
conciseness or compactness is critically important for scaling SOLVCON
from thousands of CPUs to hundreds of thousand of CPUs.

The price for this approach is to prolong the path to generic
representation of PDEs.  We do hope to provide the capability to
compile the PDEs written by users in a symbolic form for SOLVCON to
execute automatically.  But this won't happen in foreseeable future.

If you want to know more about SOLVCON and its theoretical background,
you can check up with my dissertation at
http://solvcon.net/yyc/publications.html .

with regards,
Yung-Yu Chen

> Thanks
>
> On Fri, Aug 12, 2011 at 5:44 AM, Yung-Yu Chen <yyc@solvcon.net> wrote:
>> Hello,
>>
>> I am pleased to announce version 0.1 of SOLVCON.  SOLVCON is a Python-based,
>> multi-physics software framework for solving first-order hyperbolic PDEs.
>>
>> The source tarball can be downloaded at
>> http://bitbucket.org/yungyuc/solvcon/downloads .  More information can be
>> found at http://solvcon.net/ .
>>
>> This release marks a milestone of SOLVCON.  Future development of SOLVCON will
>> focus on production use.  The planned directions include (i) the high-order
>> CESE method, (ii) improving the scalability by consolidating the
>> distributed-memory parallel code, (iii) expanding the capabilities of the
>> existing solver kernels, and (iv) incorporating more physical processes.
>>
>> New features:
>>
>> - Glue BCs are added.  A pair of collocated BCs can now be glued together to
>>  work as an internal interface.  The glued BCs helps to dynamically turn on or
>>  off the BC pair.
>> - ``solvcon.kerpak.cuse`` series solver kernels are changed to use OpenMP for
>>  multi-threaded computing.  They were using a thread pool built-in SOLVCON for
>>  multi-threading.  OpenMP makes multi-threaded functions more flexible in
>>  argument specification.
>> - Add the ``soil/`` directory for providing building helpers for GCC 4.6.1.
>>  Note, the name ``gcc/`` is deliberately avoided for the directory, because of
>>  a bug in gcc itself (bug id 48306
>>  http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48306 ).
>> - Add ``-j`` command line option for building dependencies in the ``ground/``
>>  directory and the ``soil/`` directory.  Note that ATLAS doesn't work with
>>  ``make -j N``.
>>
>> Bug-fix:
>>
>> - METIS changes its download URL.  Modify SConstruct accordingly.
>>
>> --
>> Yung-Yu Chen
>> http://solvcon.net/yyc/
>> +1 (614) 859 2436
>> _______________________________________________
>> SciPy-User mailing list
>> SciPy-User@scipy.org
>> http://mail.scipy.org/mailman/listinfo/scipy-user
>>
> _______________________________________________
> SciPy-User mailing list
> SciPy-User@scipy.org
> http://mail.scipy.org/mailman/listinfo/scipy-user
>



-- 
Yung-Yu Chen
http://solvcon.net/yyc/
+1 (614) 859 2436


More information about the SciPy-User mailing list