This is the mail archive of the
mailing list for the Cygwin project.
- From: Tony Kelman <tony at kelman dot net>
- To: cygwin at cygwin dot com
- Date: Sun, 17 Jul 2016 11:56:55 +0000 (UTC)
- Subject: Re: Parallelization
- Authentication-results: sourceware.org; auth=none
- References: <email@example.com>
Thomas Koenig <tkoenig <at> netcologne.de> writes:
> I am working on a non-linear, boring system of ODEs, boring being
> defined as non-stiff and without other numerical surprises.
> So, CV_ADAMS works well.
> The only interesting part is that there are very many of the ODEs,
> around 30000 at the moment, and that each ODE depends on around half of
> the others, so a Jacobian (were I to calculate it) would be quite
> densely packed.
> Sundials works well for this problem, but it takes rather a lnog time
> and only utilizes a single processor on a 20-core machine.
> So, the question: Is there a way to parallelize the calculation for
> this? The references in the Sundials docs that I have seen only refer
> to parallelization of solving linear equations, which I do not need
> to do.
This is more of a general numerical ODE or Sundials usage question
rather than a cygwin specific one, but I would try openmp or mpi on
your function evaluations first, if that's taking most of the time.
Do you know whether sundials is thread-safe?
Problem reports: http://cygwin.com/problems.html
Unsubscribe info: http://cygwin.com/ml/#unsubscribe-simple