Pebbles motion

Questions and discussion about applications, input, output and general user topics
Ville Valtavirta
Posts: 541
Joined: Fri Sep 07, 2012 1:43 pm
Security question 1: No
Security question 2: 92

Re: Pebbles motion

Post by Ville Valtavirta » Tue Sep 21, 2021 1:41 pm

Hi,

I think uploading scripts to e.g. github and linking the page here should be OK.

-Ville

yrob
Posts: 39
Joined: Tue Oct 01, 2019 9:50 pm
Security question 1: No
Security question 2: 19

Re: Pebbles motion

Post by yrob » Wed Sep 22, 2021 6:15 am

Thank you.
Here is the code. I might update it if needed.

https://github.com/yvrob/Serpent_Restart_files.git

yrob
Posts: 39
Joined: Tue Oct 01, 2019 9:50 pm
Security question 1: No
Security question 2: 19

Re: Pebbles motion

Post by yrob » Thu Sep 30, 2021 2:54 am

Hello,

I was wondering now: if I wanted to do the same with domain decomposition, would it work?
How do decomposed restart files work in the latest version, in particular handling materials/memory?

Because I would have materials that do not always correspond to their previous domain mainly due to recirculation.

Thank you.

Ana Jambrina
Posts: 683
Joined: Tue May 26, 2020 5:32 pm
Security question 1: No
Security question 2: 7

Re: Pebbles motion

Post by Ana Jambrina » Thu Sep 30, 2021 10:13 am

Just for clarification: the domain decomposition methodology remains the same between versions (implemented in Serpent 2.1.31), and in version 2.1.32, the restart feature was enabled to work with the domain decomposition. The restart works in the most simple/straightforward way with the domain decomposition - splitting the writing/reading between MPI tasks/domains.

The domains are set based on the geometry and, once the calculation starts, the partial-materials structures are defined within the MPI task that is going to handle that domain and from that point on, each MPI task contains only its own partials info.

The single binary restart file that is regularly generated given the 'set rfw' option is split into the multiple restart files, one per domain (written by its own domain) - containing the info relative to the parent's materials and the partials' materials defined within the domain.

When the reading time comes, 'set rfr' option, Serpent again divides the geometry/materials into domains (that information is not directly carried out by the restart files) and, each MPI task reads the corresponding restart file, which is assumed that since comes from a previous calculation is coherent with the current simulation - reason for which the number of MPI tasks/domains should match.

As previously commented, if any modification is made to the restart files, Serpent does not see it. There is no reindexing or reassignment of materials. That is why the mapping feature comes in handy with anything related to redefining geometry.
- Ana

Post Reply