Setting maximum memory usage, or dimensional refinement?

Hi There!

First time posting here after getting most of the way there through the user manual.

In large 3D simulations I’ve been running into memory issues, the model keeps refining and eventually my cluster runs out of memory and I get a segmentation fault.

My work around has been to use a global initial enhancement factor of five and then prevent coarsening but this is not quite optimum and leads to lower than desired resolution in some important parts of the model domain.

I’m wondering if there is functionality to say “only use this amount of memory” or “set this as the maximum number of elements”?

I’m also wondering if at refinement it’s possible to ask only for refinement in the x and y dimensions? My domain is fairly tabular, so I have plenty of vertical resolution, and would like to use say 5 in the z direction, but 6 in x and y.

At the moment I’m stuck with all 5 (too coarse) or all 6 (not really computationally feasible for my set up).

Many thanks,
Rob

Dear Rob,

I’m wondering if there is functionality to say “only use this amount of memory” or “set this as the maximum number of elements”?

Not that I know of. I’m not sure how one would impose a particular memory restriction, and setting a maximum number of elements would lead to requested refinement steps not running to completion. Neither of these seems like great options.

An alternative: have you tried the “Mesh refinement / Maximum refinement” function?: Mesh refinement — ASPECT 2.6.0-pre

You might need to play around a bit to get the right numbers.

I’m also wondering if at refinement it’s possible to ask only for refinement in the x and y dimensions? My domain is fairly tabular, so I have plenty of vertical resolution, and would like to use say 5 in the z direction, but 6 in x and y.

It is often sensible to set up your desired x, y, z resolutions at the beginning of the simulation, via the Repetitions parameter: Geometry model — ASPECT 2.6.0-pre

Let us know if you need any more assistance. If so, it might be useful to attach your prm file.

Best wishes,
Bob