Hi,

On Tue, Oct 26, 2010 at 8:36 AM, R Tavakoli <[email protected]> wrote:
> It is a good new. Thanks for all efforts. Where is the release note?

We haven't released it in a new deal.II version, yet. We pushed the
work into the subversion repository, where it can be used by anyone
who is interested. I think we need to give it some time to mature
before it is officially released.

> though I have not thousands of nodes like yours (but two quad-core), some
> questions:
> Doe this mean that deal ii is now fully parallel, i.e. all internal
> functionalities work in parallel (e.g. MG, native linear algebra kernel,
> preconditioners)?

In short: no. For parallel preconditioners and linear algebra you need
to use PETSc or Trilinos. We have some thoughts on parallelizing the
geometric multigrid, but nothing is implemented yet.

> It is possible to do multi-threading (using pthread) in distributed manner
> now, i.e., domain decomposition based multi-threading (not
> simple loop splitting)?

If you want to parallelize in a distributed memory model (e.g. domain
decomposition), it doesn't make a lot of sense to use threads. You
would use MPI. MPI is very efficient on multi-core machines too, by
the way. I suspect the threading model will stay as it is right now,
i.e. multi-threaded loops and/or doing several independent tasks at
the same time.

Note that using multi-threading in deal.II (using TBB) and distributed
calculations (using MPI) won't make a lot of sense used together
(called hybrid parallelization). As we currently have no way to
interface to a linear algebra package that can do hybrid
parallelization, you need to use either multi-threading or MPI.

--
Timo Heister
http://num.math.uni-goettingen.de/~heister
_______________________________________________
dealii mailing list http://poisson.dealii.org/mailman/listinfo/dealii

Reply via email to