LAMMPS WWW Site - LAMMPS Documentation - LAMMPS Mailing List Archives
Re: [lammps-users] Problem with compute cluster/atom
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [lammps-users] Problem with compute cluster/atom


From: Axel Kohlmeyer <akohlmey@...24...>
Date: Thu, 15 Jun 2017 12:29:15 -0400

please always reply to the mailing list and not only individual people.


On Thu, Jun 15, 2017 at 11:36 AM, <1132033@...3237...> wrote:

Hi Axel,

Thanks for your valuable input. I upgraded my enviroment and did more tests today, here is the report.

1. Configuation 1 (old): Ubuntu 14.04.3;  gcc 4.8.4;  g++ 4.8.4;  mpich 3.0.4; lammps-31Mar17.  Error message as mentioned in last email.

2. Configuation 2 (new): Ubuntu 16.04.2;  gcc 5.4.0;  g++ 5.4.0;  mpich 3.2;  lammps-19May17.  Run smoothly.

3. Configuation 3 (new): same as 2, but with lammps-31Mar17. Error message. Interstingly, there are two kinds of error message when I mpirun -np 40 or -np 80. I posted the whole echo message on screen below for your information.

Anyway, I found a way to move forward. Thanks for you help. If you could reproduce the error message with lammps-31Mar17 as I do, please let me know. I am still intersted in this problem. Thanks again.

​please note, that the 19May2017 version of LAMMPS contains two bugfixes for "occasional neighbor lists", which is exactly what compute cluster/atom is using and where some of the call traces you copied are hinting the corruption is happening. please see http://lammps.sandia.gov/bug.html

the only other possibility consistent with the info your provide would be some issue of the communication hardware. for that you'd have to check whether the corruption is dependent on node placement. but that is quite unlikely.

axel.

 

Best regards,

Ethan



--
Dr. Axel Kohlmeyer  akohlmey@...24...  http://goo.gl/1wk0
College of Science & Technology, Temple University, Philadelphia PA, USA
International Centre for Theoretical Physics, Trieste. Italy.