On data in the DATABASES when running seisflows

[ytian159]

When I was using seisflows to run multiple forward simulations (same model, different sources), there are large files generated in the DATABASES. I think those files are mainly duplicated, since all the simulation are using the same mesh? I have turn the "save_mesh_file" and "save_forward" option off, those files are still there and occupying lots of spaces. Can we have an option to avoid using so much spaces?

[bch0w]

Hi @ytian159, thanks for bringing this up. Memory usage can be high/overwhelming when using high-resolution meshes and a large number of events. This is something I hope to deal with in the future.

I think those files are mainly duplicated, since all the simulation are using the same mesh?

This is true. Unfortunately it is not as simple as pointing all simulations to a single set of database files, because those simultaneous simulations end up running into some type of simultaneous file read error as they all try to read the same files at the same time.

I think a more likely solution is to take advantage of the simultaneous runs parameter in SPECFEM, which will allow us to do this more efficiently. This is related to #118 and an active development branch.

I have turn the "save_mesh_file" and "save_forward" option off, those files are still there and occupying lots of spaces

Currently save_forward is switched on by default. I will submit a new PR that removes this requirement for forward workflows, where the assumption is that you will not want to be performing adjoint simulations. Note however that if you do want to perform adjoint simulations in the future, you will have to re-run your forward simulations.

[ytian159]

Thank you for clarifying this up.
I saw the updated version, thank you for adding this so quickly.
I'm using dense mesh only for forward simulation, I will use a much coarser mesh for adjoint simulation in the future.

[bch0w]

Closing this as complete, please re-open if that's a mistake!