Dear users, thank you for clearing the situation!
Our admin told us to limit usage of memory to prevent something like memory overusage.
As far as I understand there are 4 processors per node sharing 4 Gb of memory.
850x4=3400 mb for the process and the rest is for the system.
2010/12/20 t t <[email protected]...>:
Dear users, thank you for clearing the situation!
Our admin told us to limit usage of memory to prevent something like memory
overusage.
As far as I understand there are 4 processors per node sharing 4 Gb of
memory.
850x4=3400 mb for the process and the rest is for the system.
this is actually more of a torque/pbs/linux question than something
lammps specific:
please note that you are restricting address space and not physical
memory and actually used memory. with modern virtual memory management
and shared libraries and use of copy-on-write allocation schemes, it
is very difficult to account for real memory use of a per process basis.
batch systems have not caught up to it. in any case, you probably
want to run your job across multiple nodes to have a decent throughput
and then the whole memory/address space concern should become
irrelevant due to the distributed memory scheme in lammps.
in any case, i would experiment with raising the limits and monitoring
how the jobs execute and when they start to swap. what your sysadmin
has been telling you is overly careful but may be well needed for
applications that do a lot of disk i/o (e.g. quantum chemistry jobs).
cheers,
axel.