LAMMPS and sparse dump files

Hello,

Does anyone know if there are certain situations that could cause LAMMPS to produce
sparse dump files? I’m using the lammps-30Jul16 version with a custom potential model.

Occasionally, our LAMMPS simulations produce TB-sized dump files.
Closer investigation revealed that these were in fact apparent sizes: the actual file
sizes were in the 10 GB range, as expected.

What’s odd is that sometimes the simulations produce dump files that are non-sparse
(normal behaviour) and at other times these files are sparse. There seems to be a
non-deterministic element to this problem. This seems like a potential bug, however, I
couldn’t find a description of a known bug that matches the issue I’m seeing
(http://lammps.sandia.gov/bug.html).

Thanks in advance,
Michael

Hello,

Does anyone know if there are certain situations that could cause LAMMPS
to produce
sparse
<https://administratosphere.wordpress.com/2008/05/23/sparse-files-what-why-and-how/> dump
files? I’m using the lammps-30Jul16 version with a custom potential model.

Occasionally, our LAMMPS simulations produce TB-sized dump files.
Closer investigation revealed that these were in fact *apparent* sizes:
the actual file
sizes were in the 10 GB range, as expected.

What’s odd is that sometimes the simulations produce dump files that are
non-sparse
(normal behaviour) and at other times these files are sparse. There seems
to be a
non-deterministic element to this problem. This seems like a potential
bug, however, I
couldn’t find a description of a known bug that matches the issue I’m
seeing
(http://lammps.sandia.gov/bug.html).

​this is more likely a bug/problem with the file system hosting the files
than with LAMMPS.
LAMMPS is using standard i/o functions from the stdio library part of the c
library.

axel.

i should add to my statement, that i am assuming that you are using the native text mode LAMMPS dump style (atom or custom).
LAMMPS supports a large variety of output formats and dump styles, but since you didn’t mention anything specific it implies the standard format.

if you are using a format based on MPI-IO, you may want to update your LAMMPS version in order to make certain, you have all bugfixes on those, since those use more complex ways to perform i/o. that said, it is still most likely that you have exposed a problem in your local systems file system code, or are using the machine in an unsupported way.

axel.