Dear Admin,
I am using the latest version of LAMMPS compiled on Bluegene/Q. Currently, using the “dump atom” command, the output dump file ends up having lines with gibberish, for instance:
12560 14 -74.349 -55.6431 -43.6212 0 0 0
12561 4 -74.0956 -54.8929 -44.1983 0 0 0
12562 4 -75.09 -56.0251 -44.113 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 2.54639e-313 0 0 0
I have simultaneously dumped .xtc and .dcd files and these were perfectly fine. May I know where the issue might lie? Is it a problem with how I have compiled LAMMPS?
Best Regards,
JJ
**Disclaimer** The sender of this email does not represent Nanyang Technological University and this email does not express the views or opinions of the University.
Dear Admin,
I am using the latest version of LAMMPS compiled on Bluegene/Q. Currently, using the “dump atom” command, the output dump file ends up having lines with gibberish, for instance:
12560 14 -74.349 -55.6431 -43.6212 0 0 0
12561 4 -74.0956 -54.8929 -44.1983 0 0 0
12562 4 -75.09 -56.0251 -44.113 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 2.54639e-313 0 0 0
I have simultaneously dumped .xtc and .dcd files and these were perfectly fine. May I know where the issue might lie? Is it a problem with how I have compiled LAMMPS?
there is not enough information here to know anything. one can make
guesses, but even for that, you don't provide much useful data. ...and
how can anybody know how you have compiled LAMMPS, when you don't
say??
your first two steps should be to figure out whether the same input
run with the same settings produces the same inconsistency on any
other machine and the second whether this only happens with a
particular input or also with the example inputs from the LAMMPS
distribution.
for the future, please check out the "ten commandments" for posting
http://lammps.sandia.gov/guidelines.html
axel.
This looks a like LAMMPS is writing data to file from a buffer that is uninitialized. This suggests a run-time problem with the MPI or threading library, probably due to incompatible compiler and run-time environments. Note:
sort = id for dump styles dcd, xtc, and xyz
Trying running the problem on a trusted machine to see if the behavior is different.
Aidan
I apologise for my tardiness in diagnosing the problem. As you guys have suggested, I attempted to replicate this issue in the shear example on the BGQ but I did not observe the same problem. Would this be related to the fact that my system has upwards of 200k atoms?
The same command was used in both cases:
dump 1 all atom 1000 file.dump
I have also attached my makefile. LAMMPS was compiled with “make yes-standard no-gpu no-kim no-kokkos no-meam no-poems no-python no-reax no-voronoi no-compress”
I’ve ran the same job on BGQ, another SGI Altix UV 1000 cluster with SUSE, and on my workstation running Linux Mint. Only the BGQ runs had this issue.
**Disclaimer** The sender of this email does not represent Nanyang Technological University and this email does not express the views or opinions of the University.
Makefile.cumulus (3.72 KB)
ATT00001.txt (3.48 KB)
I apologise for my tardiness in diagnosing the problem. As you guys have suggested, I attempted to replicate this issue in the shear example on the BGQ but I did not observe the same problem. Would this be related to the fact that my system has upwards of 200k atoms?
The same command was used in both cases:
dump 1 all atom 1000 file.dump
I have also attached my makefile. LAMMPS was compiled with “make yes-standard no-gpu no-kim no-kokkos no-meam no-poems no-python no-reax no-voronoi no-compress”
I’ve ran the same job on BGQ, another SGI Altix UV 1000 cluster with SUSE, and on my workstation running Linux Mint. Only the BGQ runs had this issue.
this hints very strongly at a bug in either the compiler or the MPI
library on the blue gene machine. if you have the GCC toolchain
installed, you can try using that instead of XLC. the XLC compilers
have a notoriously bad reputation for miscompiling C++ code. if you
cannot switch the compiler, it might help to reduce optimization and
vectorization/multi-threading.
axel.
To clarify, I should recompile with either gcc or simplify CCFLAGS = -g -O2?
JJ
To clarify, I should recompile with either gcc or simplify CCFLAGS = -g
-O2?
either is an option worth trying. it is extremely difficult to debug
such a thing from remote.
your best shot to get proper help is to contact the HPC staff of the
bluegene machine and tell them that you are concerned about your
program being miscompiled or being affected by a bug in the MPI
library or the compilers and then work with them to resolve it.
axel.