Nextsimdg

[aydogduali]

This PR should close the PR #7 in terms of the fixes for gricad environments. It is tested and performing an ensemble simulation using nextsimDG. No DA test yet. @yumengch

[yumengch]

I've added a few comments in this PR. They're mainly for the purpose of explaination. In hindsight, perhaps I should've put some comments directly in the code.

[myying]

@aydogduali @yumengch, Thanks for the efforts in testing and modifying the code. Could you send me some initial condition and forcing data, so that I can also perform a test on my side?

[aydogduali]

I see that you have access on dahu. The IC and BC are here: /bettik/yumengch-ext/DATA If you test it on gricad the code should use them by default. Let us know.

[myying]

I see that you have access on dahu. The IC and BC are here: /bettik/yumengch-ext/DATA If you test it on gricad the code should use them by default. Let us know.

My account has expired, I've asked Stephanie to revive it. I don't think I have access to compute nodes, so I'll test on my Norwegian machines.

Before I get my account back, maybe you can send me the metadata information for the IC and BC files? Just send me the output from "ncdump -h *.nc" for one of the members, and I can start preparing some files already.

[aydogduali]

I see. Actually you can find everything here. Other SASIP related data are also open access here. The configuration files that we use should be this one. Let me know.

[myying]

Now I've managed to run the nextsimdg model with the IC/BC files from /bettik/yumengch-ext/DATA.

I'm testing the perturbation codes, it seems work in single processor mode (where I bypass the scheduler code and just call the model.preprocess). However, when I run multiprocessor mode, the perturbations are generated but not added successfully to the nc files (no difference for all members), I'm suspecting that threading.Lock() is not quite working...

[myying]

I've tested the code on betzy.sigma2.no computers using slurm.py submitter, there are some bugs in model.preprocess(), slicing_nc and perturbation code. After debugging with yumeng, we managed to make the ensemble forecast part work okay on my computer. I trust that it will also work fine on gricad with oar.py submitter. I will follow up in a new PR with the bug fix commits and cherry pick the updates from my other_features branch to go forward for the DA part.