The only change is an interface name change for some MPI variables. This
should not end up in this situation.
However, it seems like the tbtrans is not updated accordingly, a later fix
(rev 435) patches that, but this is not the issue you are seeing.

Please supply your output files.

Kind regards Nick


2013/10/22 <[email protected]>

>  Hello transiesta users and developers.
>
>  I have two versions of transiesta installed in two different machine
> architectures. I am running now the same calculation in both machines: same
> structure, same input. The number of transiesta steps, however, is
> ridiculously different in each. Let me summarize:
>
>  Case A:
> 16 nodes
> trunk 431
> 22 transiesta steps (ts-charge)
> finished
>
>  Case B:
> 32 nodes
> trunk 433
> 314 and running!
>
>  I had expected a similar if not identical performance. In fact, I
> expected the run in the 32 node cluster to be much faster wall-time wise
> than that in 16, but it seems that the transiesta loop is taking forever in
> case B.
>
>  @Nick: is transiesta/tbtrans_rep identical in both releases? If so, do
> you have any idea why the difference in performance?
>
>  Thanks for your attention and help.
>
>  All the best,
>
>  Diana
>
>   Diana M. Otálvaro
> PhD Candidate
>
>  Computational Material Science
> MESA+ Institute of Nanotechnology
> University of Twente.
> Enschede, Nederland
>

Responder a