[DFTB-Plus-User] MPI Version crashes

[Silvio a Beccara]

Dear Bálint,

thanks for the information. We recompiled the OpenMPI libraries with gcc 
4.9 (latest build) and parallel DFTB+ now works, giving exactly the same 
energies as the serial version.

However, unfortunately we're having some performance issues. We tried a 
system with about 5000 silver atoms, and this is the result of a test 
run: after 15 hours, on 32 cores with Infiniband connection, at 2.67 
GHz, only 8 SCC cycles were completed.

Moreover, if we double the number of the cores, the computing time 
increases slightly instead of decreasing, contrary to what would be 
expected.

What do you think? Do you have any idea as to why this is happening, and 
maybe some suggestion to optimise the calculation?
I am including our input and output files, together with the SKF file.

Thanks and regards

Silvio a Beccara


-- 
Silvio a Beccara
LISC - Fondazione Bruno Kessler
18 via Sommarive
38123 Povo TN
ITALY


> Dear Silvio,
>
>> I think that this has to do with the version of OpenMPI I am using: did
>> you re-compile also OpenMPI from scratch with the 4.9 compiler?
>>
> Yes, I did, as the .mod-files are not compatible between various
> compiler versions.
>
>    Best regards,
>
>    Bálint