[CIG-MC] Aspect plans for the future (and things we've worked on lately)

Scott King sdk at vt.edu
Mon Feb 10 10:53:21 PST 2014

Thanks Wolgang and Timo, this is great.

My suggestion is to add 

We are having twice monthly conference calls with Eric and Raj to discuss CitcomS development and spherical convection benchmarking and we see this continuing during the following year.  We have near term goals of 1) removing python code from CitcomS without loss of functionality for users, a first step in the integration with PetSc; 2) integration of PetSc in CitcomS, and 3) spherical benchmarking, both addressing performance and accuracy, with a first report due at the Banff workshop in May.   Following these tasks we will revisit integrating the temperature stabilization used in Aspect and perhaps using what we learn in the benchmarking.  My thinking on this last point is whether it is worth using a different method than Aspect in order to give us greater diversity in available methods. Doesn't need to be decided now, but I think we should keep an open mind. 

The above paragraph will be fleshed out more for the strategic plan, but I like trying to keep it brief.

We might think about the following.   I believe that we have de-facto two working groups.   One more closely aligned with Aspect and one more closely aligned with CitcomS.   There is some cross-over.  Clearly, with things like benchmarking but even with regard to code development, cross-fertilization could be beneficial to all.  I think there is some, but I don't know how formal it is and if we should make it more formal.  I believe the EC and SSC are going to discuss working groups at some time in the future (should they be "refreshed", how to make sure new people are included, what is an optimal size, ...).   If anyone have thoughts, feelings, and/or ideas about this, it would be a great time to voice those so they can become part of the initial discussion.   Should we have one group (mantle convection) with two subgroups (Aspect and CitcomS) or should working groups be based on code and not discipline? (if so, what do we do about workshop planning?)

Anyone else have anything to add?


On Feb 10, 2014, at 10:29 AM, Wolfgang Bangerth wrote:

> All,
> Lorraine asked us to provide some feedback on what we have been doing with the Aspect code (http://aspect.dealii.org/) in the past few months and, in particular what our plans for the next year. Since Aspect is a code for the community, it makes sense to discuss these in the community. Below please find a draft of this document. We are looking for feedback whether that's sensible and/or what your opinions are!
> Best & thanks
> Wolfgang & Timo
> Activities in the past 6 months:
> ./ Worked on the Rayleigh-Taylor instability, explained the results
>   and reproduced the results reported in the literature
>   (combined effort over the mailing list)
> ./ Implementation in Aspect to (optionally) use PETSc instead of
>   Trilinos and necessary work in deal.II for this (Timo)
> ./ Training a student to work on improving temperature stabilization
>   scheme, ongoing progress (Timo)
> ./ As a result the student contributed smaller features and bugfixes in
>   Aspect (typos, hitting end time exactly) (Timo)
> ./ New Aspect virtual machine (Timo)
> ./ Implement periodic domains (Timo)
> ./ Cookbook: 3d convection (Wolfgang)
> ./ Cookbook: Rayleigh-Taylor instability (Wolfgang)
> ./ Implementation of models that treat the melt fraction as a function of
>   pressure and temperature (Wolfgang with Juliane Dannberg)
> ./ Implementation of models with latent heat (Wolfgang with Juliane Dannberg)
> ./ Fixes in the nonlinear solvers (Wolfgang)
> ./ Fixes for shear heating models (Wolfgang)
> ./ Fixes for compressible models (Wolfgang)
> ./ Several more cookbooks that show how to use Aspect and set up input files (Wolfgang)
> ./ Benchmarking to very large runs (up to 300+M unknowns on 6000+ processors,
>   with Rene Gassmoeller)
> ./ Big speedups in assembly (30-50%)
> ./ Better testing for 3d experiments
> ./ Youtube videos showing 2d and 3d results to help with publicity and outreach
> ./ A new section in the manual providing guidelines for faster computations
> ./ Some work on refactoring the tracer implementation (Eric, with some help by Wolfgang)
> ./ Better support for snapshot/resume
> Plans for the remaining months of the current project year and project year 5:
> ./ Development of Aspect tutorials that can be used for training sessions in
>   the future. Providing such training sessions at the CIDER (Timo), SEDI (Wolfgang)
>   and other meetings (e.g., at a European meeting later this year by Juliane Dannberg).
> ./ 3d shell computations (Wolfgang)
> ./ Participation in the 3d shell benchmarking effort
> ./ Reproducing more of the existing benchmarks
> ./ Improving temperature/composition stabilization scheme (Timo)
> ./ Active tracers (with Eric)
> ./ Better support for levelset-like compositional fields (accuracy, stability)
> ./ Parallel benchmarking of deal.II and Aspect to find bottlenecks
> ./ Working with members of the community to get their patches into Aspect,
>   in particular on melt models
> ./ More cookbooks
> ./ Nonlinear models: more testing and (likely) bug fixing
> ./ Porting to new platforms (e.g., the Bluegene/Q machine at Texas A&M and
>   the machines at the German High Performance Computing Centers)
> ./ Continued work on improving the manual
> ./ New Aspect releases
> ./ Work on a second Aspect publication detailing, among other topics, how
>   Aspect treats compressible models
> -- 
> ------------------------------------------------------------------------
> Wolfgang Bangerth               email:            bangerth at math.tamu.edu
>                                www: http://www.math.tamu.edu/~bangerth/
> _______________________________________________
> CIG-MC mailing list
> CIG-MC at geodynamics.org
> http://geodynamics.org/cgi-bin/mailman/listinfo/cig-mc

More information about the CIG-MC mailing list