segmentation faults

[esp]

I keep getting segmentation faults ... i thought it was because of an ldos calc so i tried a transmission calc .. same thing but after many hours .... do you think i should have someone check out the machine?  if it is simply that atk is using too much memory, why doesn't atk use only as much as needed, or as much is available?  why allow it to create this segmentation fault?

[Anders Blom]

You cannot limit a calculation in memory. If it needs to allocate 5 GB to store a matrix for instance, there is no alternative but to allocate 5 GB - and fail it that's not available.

[esp]

I suppose that makes sense, i gues i was thinking it could serialize when needed, when memory is not enough

[Anders Blom]

To write a code with that logic would be extremely difficult I think, because the MPI parallelization is done very deep down and the memory allocation occurs inside that part of the code. So you can't "bubble up" at that point to the top-level and redo everything again in serial.