lammps-19Jul11 : large virtual memory consumption

Dear LAMMPS user,

I recently compiled Lammps (lammps-19Jul11 version; Intel 11.0 64 bits, intel MPI 3.2 64 bits; FFTW 2.1.5) on a linux (running 2.6.18-53) cluster.

Dynamics run fine on multiple nodes, but appear to consume a huge amount of virtual memory (12.2 GB/core) as you can see below:

Cpu(s): 4.3%us, 0.6%sy, 0.0%ni, 95.1%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st
Mem: 32961556k total, 9381104k used, 23580452k free, 445748k buffers
Swap: 8385888k total, 0k used, 8385888k free, 4301780k cached

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
8522 debruint 25 0 12.2g 206m 7364 R 100 0.6 33:29.55 lmp_linux
8523 debruint 25 0 12.2g 210m 7160 R 100 0.7 33:32.88 lmp_linux
8524 debruint 25 0 12.2g 206m 7132 R 100 0.6 33:30.15 lmp_linux
8525 debruint 25 0 12.2g 202m 7372 R 100 0.6 33:32.14 lmp_linux
8527 debruint 25 0 12.2g 210m 7352 R 100 0.7 33:29.07 lmp_linux
8528 debruint 25 0 12.2g 210m 7372 R 100 0.7 33:32.84 lmp_linux
8529 debruint 25 0 12.2g 210m 7096 R 100 0.7 33:32.87 lmp_linux
8530 debruint 25 0 12.2g 212m 7364 R 100 0.7 33:32.95 lmp_linux
8531 debruint 25 0 12.2g 206m 7164 R 100 0.6 33:29.92 lmp_linux
8532 debruint 25 0 12.2g 211m 7388 R 100 0.7 33:30.74 lmp_linux
8533 debruint 25 0 12.2g 210m 7396 R 100 0.7 33:31.66 lmp_linux
8535 debruint 25 0 12.2g 211m 7372 R 100 0.7 33:31.55 lmp_linux
8536 debruint 25 0 12.2g 209m 7384 R 100 0.7 33:33.10 lmp_linux
8521 debruint 25 0 12.2g 209m 7084 R 98 0.7 33:29.29 lmp_linux
8526 debruint 25 0 12.2g 209m 7400 R 98 0.6 33:32.16 lmp_linux
8534 debruint 25 0 12.2g 211m 9716 R 98 0.7 33:32.63 lmp_linux
20239 debruint 15 0 10964 1100 664 R 2 0.0 0:00.01 top

FYI, the simulation box contains about 70000 atoms.

Is this a normal memory behavior?

Thanks in advance for your reply.
Kind regards
Theo de Bruin

Dr. Theo de Bruin
Research Engineer

IFP Energies nouvelles - Applied Chemistry and Physical Chemistry Division

1 et 4 avenue de Bois-Préau
92852 Rueil-Malmaison Cedex - France
Phone: +33 1 47 52 54 38 - Fax: +33 1 47 52 70 58
Email: theodorus.de-bruin@…2876…
Web: www.ifpenergiesnouvelles.fr

Dear LAMMPS user,

I recently compiled Lammps (lammps-19Jul11 version; Intel 11.0 64 bits,
intel MPI 3.2 64 bits; FFTW 2.1.5) on a linux (running 2.6.18-53) cluster.

Dynamics run fine on multiple nodes, but appear to consume a huge amount of
virtual memory (12.2 GB/core) as you can see below:

virtual memory is only reserved address space, not physical memory.
check out the RES column that lists the amount of it that is actually
_resident_ in physical memory (add to that shared, which is shared
across multiple processes, and thus difficult to account for).

it is _very_ likely that this is caused by your MPI library and that this
address space is reserved for backing communication. does you machine
have infiniband interconnect?

cheers,
    axel.

LAMMPS also prints out the amount of memory (per processor)
that it is using, at the end of the setup phase. Anything
way above that is system stuff (like Axel indicates), not
LAMMPS memory.

Steve