[DFTB-Plus-User] DFTB+ 20.1 memory leakage
NaOH
naohjohn at qq.com
Mon Nov 9 12:22:08 CET 2020
Dear DFTB+ developers/users,
I'm experiencing the same memory leakage problem during MD runs in DFTB+ 20.1 as previous thread "DFTB+ 19.1 memory leakage"(quoted below).
The code is compiled using Intel MPI/MKL without ELSI or ELPA. I do NVT simulation with OMP_NUM_THREADS=1. The system consists of Au C H S O atoms. The memory grows gradually as MD steps increases. After several ten thousand steps the program crashes with the same error below. On contrary, non-MPI version is without memory problems.
I do not see the exact system in the previous thread(114 C atoms and 10 H atoms), but it is quite different from this system (Au C H S O). So I think the problem may not be related to the system but could be MPI-related.
Could someone resolve this issue?
Thank you very much!
"
> I am experiencing memory leakage when doing molecular dynamics runs with
> DFTB+ on my machine. > My system configuration is: > > * Ubuntu 18.04 LTS > * Intel Xeon CPU > * 256 GB RAM > > I compiled DFTB+ with the ELSI library and have this issue with the ELPA > solver. However, I also experience the memory leakage when using the QR > solver. > The system I'm simulating consists of 114 C atoms and 10 H atoms (nothing > too wild here, no spin polarization). > > After setting up my starting structure I do a NVE simulation (see attached > input file), starting with 'mpiexec -np 24 dftb+'. I set OMP_NUM_THREADS=1 > for the ELPA solver. > I have logged my memory and it shows an exponential growth of memory > usage, until after ~ 77.000 MD steps my complete 256 GB RAM are full and > the simulation crashes with an MPI error > > > =================================================================================== > = BAD TERMINATION OF ONE OF YOUR APPLICATION PROCESSES > = RANK 0 PID 19101 RUNNING AT xxxxxxxx > = KILLED BY SIGNAL: 9 (Killed) > > =================================================================================== > > Has anyone experienced the same problems or could help me resolve this > issue? > Thanks in advance > > > Best regards, > Paul Schwarz"
Best regards
Jian Shao
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.zfn.uni-bremen.de/pipermail/dftb-plus-user/attachments/20201109/4078a44d/attachment.htm>
More information about the DFTB-Plus-User
mailing list