[Dune-devel] Parallel solvers in dune-istl
Oliver Sander
oliver.sander at tu-dresden.de
Mon Jan 8 19:03:28 CET 2018
Hi Martin,
stupid me, that's quite an obvious answer!
Thanks,
Oliver
On 08.01.2018 15:05, Martin Nolte wrote:
> Hi Oli,
>
> The dune-istl solvers print verbose information on every rank you ask it to. If
> you pass in verbose = 2*(rank == 0), you will end up with the desired result. I
> checked the solvers and the implementation does not require verbose to be equal
> on all ranks. I cannot, however, tell you whether this behavior is specified or
> intended in any way.
>
> Best,
>
> Martin
>
> On 01/08/2018 02:27 PM, Oliver Sander wrote:
>> Dear dune,
>>
>> I am trying to implement a distributed Poisson solver using only
>> dune-grid and dune-istl. The algorithm shall be a preconditioned
>> CG method. My current understanding of dune-istl is that I should
>> use the standard CGSolver implementation for this, and parallelization
>> happens in dedicated ScalarProduct, Preconditioner, and LinearOperator
>> implementations. This approach works, but the CGSolver class prints
>> all information about the current residual, estimated conv rate etc.
>> on _all_ processes. Is this the expected behavior? Why? It gets
>> annoying pretty quickly.
>>
>> Did anybody else ever implement a Poisson solver using only dune-istl
>> and dune-grid? My own works, but it does not look very natural to me.
>> I would appreciate a look at alternative implementations.
>>
>> Thanks,
>> Oliver
>>
>>
>>
>> _______________________________________________
>> Dune-devel mailing list
>> Dune-devel at lists.dune-project.org
>> http://lists.dune-project.org/mailman/listinfo/dune-devel
>>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 5158 bytes
Desc: S/MIME Cryptographic Signature
URL: <https://lists.dune-project.org/pipermail/dune-devel/attachments/20180108/68e65e4f/attachment.bin>
More information about the Dune-devel
mailing list