Re: [ESPResSo-users] 24 Espresso threads consume 32GB memory in 1 hour.

[Olaf Lenz]

Hi!

I have certain problem understanding the actual question in your mailing.

> If I delate the vtf output lines, the whole program will consume
> 32GB memory in 10 hours. When i check using top command, it shows
> that the mpirun thread consumes 97% of the whole memory.

What do you mean by "delate"? And I guess you mean that the generated
file will have a size of 32GB, rather than the simulation process
consuming 32GB of memory? And you write that the mpirun thread takes 97%
of the memory. How much memory does the machine have?

> writevsf $vtf radius {0 2.0 1 auto 2 1.0}

> Did I do anything wrong?

Looking at your script, I note that you use the command "writevsf"
within the main loop of the simulation. "writevsf" will write out the
structure of the system (i.e. which atoms and bonds are in the system).
The structure should be contained *only once* in every VTF file.
I think what you want to do is to write out the structure of the system
once before you start the main integration loop, and during the
integration loop you only write out the coordinates of the system via
the command "writevcf".

> What's the most elegant way to save the current status automatically 
> so that I can easily resume it when the computer crashes?

To write out the state and restart a simulation, you should use the
"blockfile" commands. The VTF format is *not* suitable for that purpose,
it is intended for visualization of the system via VMD.


Olaf

-- 
Dr. rer. nat. Olaf Lenz
Institut für Computerphysik, Pfaffenwaldring 27, D-70569 Stuttgart
Phone: +49-711-685-63607

olenz.vcf
Description: Vcard