[SciPy-user] Parallel linear solver.

Frank Lagor dfranci@seas.upenn....
Thu Jul 24 11:15:27 CDT 2008


One last thing--

If you just want parallel interaction between your python codes, and you
think that petsc4py is overkill with all of it's solvers and such, then just
use mpi4py to exchange messages, perform reductions, etc. in parallel.  It
is written by the same author for petsc4py (but it is not a requirement for
petsc4py) and you may find it a bit easier to use if you want to run
basically sequential codes in parallel and exchange results.

-Frank

On Thu, Jul 24, 2008 at 12:06 PM, Frank Lagor <dfranci@seas.upenn.edu>
wrote:

> Hi Nils,
>
> Typically users come to find PETSc when they need it-- that is if they are
> doing a lot of scientific computations and parallel processing is therefore
> definitely needed. For me, I had access to a small 64 processor cluster and
> sequential codes that would take days to run, so it was a natural fit.  I
> could definitely see it being used  in smaller settings (and I'm sure many
> people do), like on a desktop machine with a few processors, but that is not
> what I use it for.  I'm not sure about your needs or your access to a
> cluster, so you'll can probably be the best judge of if it is for you.
>
> For software requirements-- PETSc uses BLAS, LAPACK, and an MPI
> distribution as a mimimum.
>  It can also interface with countless other packages (e.g. SCALAPACK,
> ATLAS, SPRNG, etc.), but I don't bother with all this.  For me, there was an
> OpenMPI implementation of MPI already installed on my cluster (as should be
> the case for most clusters), so I just linked to it.  And the BLAS and
> LAPACK on the cluster were not working currently, so I told PETSc to
> download and install BLAS and LAPACK automatically.  It did and it works
> fine. Anyways, that's my story-- I hope it helps.
>
> -Frank
>
>
>
>
> On Thu, Jul 24, 2008 at 11:48 AM, Nils Wagner <
> nwagner@iam.uni-stuttgart.de> wrote:
>
>> On Thu, 24 Jul 2008 11:31:53 -0400
>>   "Frank Lagor" <dfranci@seas.upenn.edu> wrote:
>> > Yes, there will be a new release of PETSc soon (I just
>> >asked one fo the
>> > developers), but they don't know exactly how long it
>> >will be. The  2.3.3
>> > release of PETSc is still current.  They use patches for
>> >a lot of their
>> > small changes.  For example, just a two months ago or
>> >so, I downloaded a
>> > version that was 2.3.3-p8.  The current patched version
>> >is 2.3.3-p13.  It is
>> > still very active, and in my opinion, major releases are
>> >not a good thing,
>> > because it may end up affecting your code.  Actually, I
>> >think this is
>> > probably why you asked (so you could wait until a new
>> >release so you
>> > wouldn't have to worry about code changes for a while).
>> >
>> > Hope this help,
>> >Frank
>> >
>> Fank,
>>
>> Thank you very much for your prompt. So far I have used
>> serial programs solely.
>> How do I benefit from parallel code in python ?
>> I mean what are the minimal requirements to run codes in
>> parallel (hardware/software) ?
>> What software packages are needed to configure petsc in
>> that context ?
>>
>> Nils
>>
>>
>> _______________________________________________
>> SciPy-user mailing list
>> SciPy-user@scipy.org
>> http://projects.scipy.org/mailman/listinfo/scipy-user
>>
>
>
>
> --
> Frank Lagor
> Ph.D. Candidate
> Mechanical Engineering and Applied Mechanics
> University of Pennsylvania
>



-- 
Frank Lagor
Ph.D. Candidate
Mechanical Engineering and Applied Mechanics
University of Pennsylvania
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://projects.scipy.org/pipermail/scipy-user/attachments/20080724/0dd6f24d/attachment.html 


More information about the SciPy-user mailing list