[CP2K:912] Re: parallel distribution of data

[Teodoro Laino]

Ciao Nic,

you're definitely right.. that array is not scaling linearly, but  
quadratically as you noted. And if atoms are
molecules, like in your case, the memory is just exploding.

While somebody, somewhere will fix this issue, you can follow the  
suggestion of Matt (about
the topology generation).. but keep in mind that there's no safe way  
to generate a topology..
It may easily end up with a failure.

So there area mainly two alternatives:
1) Trying to use the internal topology generator with a section like  
that:
&GENERATE
      CREATE_MOLECULES
      REORDER
&END

2) Provide a PSF for the 8Ksomething water molecules. PSF is a kind  
of file that describes the
connectivity of a system, In your case once you build the one for a  
single water molecule  you can, with
some scripting, extend it to NK water molecules easily.

Let us know if that by-passes the problem.
Teo

On 1 Apr 2008, at 08:44, Matt W wrote:

>
>> Could the 2D_MOLECULAR_DISTRIBUTION TRUE be a workaround? Do I  
>> define the
>> molecule through
>> some connectivity file?
>
> I've no idea, in principle it should help, I think.
>
> __ROOT__%FORCE_EVAL%SUBSYS%TOPOLOGY%GENERATE
>
> looks like it might do what you need.  I'd test on a smaller than
> 25,000 atom system though;)
>
> Matt
> >