Strategies to reduce memory usage

[MarcusAW]

Hello,

I am running a large LES simulation that requires small time steps. I am having trouble with memory usage when using a setup similar to the tutorials.

Besides increasing the allocated memory on HPC, are there any other strategies I could try?

Thanks,
Marcus

[mkraposhin]

Hello @MarcusAW , all possible strategies will require rewriting of the code:

1. If you do MPI simulation, we can use several nodes (and even distribute computations amongth them).
2. It is possible to dump results to hard disk and run processing latest.
3. Or we can think on compressing the data.

If you have other ideas, please add them in reply.

[MarcusAW]

Hello,

I am quite interested getting the library to work for MPI.
Is there anyone in the research group working on this?

Thanks,
Marcus

[mkraposhin]

Hi, @MarcusAW ! Currently, I'm working in another organization (and in another area), so I'm not aware of what happens inside unicfd group. I might try to implement MPI support for libAcoustics someday as a hobby.

[michaelmrq]

How can I dump results to be processed later?

[mkraposhin]

Well, unfortunately, this feature is still in the wish list ((

[mkraposhin]

How can I dump results to be processed later?

Hello, I made very simple and preliminary version of libacoustics that dumps calculated data to disk. If you are interested, you can look here: https://github.com/mkraposhin/libAcoustics/tree/digitef-dev-2312-dump

There's still a lot of work todo. I thought maybe you could check source code .