[Dune] publications about Dune parallel scalability and performance

[Marian Piatkowski]

Hi Ziad,

good to hear that incompressible Navier-Stokes fellows are interested in 
DUNE.
The research group of Peter Bastian has done quite some work on 
high-order DG methods, DG incompressible Navier-Stokes, and 
convection-diffusion-reaction type equations.
As you said, a DG code may be slow for such equations. To obtain a 
high-performance finite element code, necessary steps need to be taken 
that include
- avoid superfluous gather/scatter operations during assembling, in 
particular for DG methds
- exploit tensor product structure in finite elements on 
quadrilateral/hexahedral meshes, sum-factorization
- matrix-free methods

Those optimizations have been realized within the PDELab discretization 
framework for example. The publications to be listed below show 
performance and scaling results using this code basis:

[1] https://arxiv.org/abs/1711.10885
[2] https://www.sciencedirect.com/science/article/pii/S0021999119303973

cover convection-diffusion-reaction type equations where the former 
shows performance and scalability results and the latter matrix-free 
solver effectiveness.
During my PhD project I have developed a large-scale incompressible 
Navier-Stokes code on top of PDELab, using splitting schemes like Chorin 
as you said, and using a DG discretization. It is described in the 
following references:

[3] https://www.sciencedirect.com/science/article/pii/S0021999117308732
[4] https://arxiv.org/abs/1912.10242
[5] http://archiv.ub.uni-heidelberg.de/volltextserver/26674/

where the last item includes performance and scalability results for the 
Navier-Stokes solver.
We are also working on a code generator for the PDELab discretization 
framework based on UFL. Here I want to share the following two publications:

[6] https://arxiv.org/abs/1812.08075
[7] http://archiv.ub.uni-heidelberg.de/volltextserver/27360/

Hope this helps as a start-up.

Best regards,
Marian

On 12/30/19 4:27 PM, Ziad Boutanios wrote:
> Hi Christoph,
>
> I have an extensive background in multiphase flows using both FEM and 
> FVM. My interest in DG for incompressible multiphase simulations stems 
> from the suitability of the method for hyperbolic problems. It is also 
> quite good for volume fraction transport equations where the volume 
> fraction needs to be bounded between 0 and 1.
>
> I recently wrote a CG/DG code for a Eulerian-Eulerian formulation 
> using UFL with good results but it is very slow compared to FVM. Such 
> a code can be used for aero-icing, sediment transport, drifting snow, 
> particulate flow through porous media, etc. Right now I am looking to 
> modify the equations and make the code faster, maybe use DG for all 
> equations if it helps parallel runs.
>
> My interest in Dune is due to improved mesh handling and availability 
> of automatic mesh adaptation. This is a research project I'm doing 
> part-time for lack of funding but I am open to collaborations if 
> someone has a common interest and could get involved full-time.
>
> Kind regards,
>
> Ziad
>
> On 2019-12-28 2:20 p.m., Christoph Grüninger wrote:
>> Hi Ziad,
>> my impression is, that there is not must interest in large-scale
>> Navier-Stokes calculation within the Dune community. Most applications
>> are elliptic or parabolic.
>> The only exception is some work by Robert Klöfkorn (et al.?), but that
>> was about shock fronts that needed grid refinement along the front. So
>> not incompressible. And quite some time ago.
>> Some weather simulations were or are done using Dune. I don't know what
>> type of equations they solve.
>>
>> Why are you asking? Do you want to use Dune for this kind of appliation?
>> Do you want to compare its performance to other tools? What's your
>> background? How much would you be willing to invest?
>>
>> Bye
>> Christoph
>>
>> _______________________________________________
>> Dune mailing list
>> Dune at lists.dune-project.org
>> https://lists.dune-project.org/mailman/listinfo/dune
>
> _______________________________________________
> Dune mailing list
> Dune at lists.dune-project.org
> https://lists.dune-project.org/mailman/listinfo/dune

-- 
Dr. Marian Piatkowski
Interdisciplinary Center for Scientific Computing (IWR)
Heidelberg University

Im Neuenheimer Feld 205
69120 Heidelberg
Germany

Tel: +49 6221 54 14539
Web: https://conan.iwr.uni-heidelberg.de/people/marian/