[Wien] question about parallel run

[Stefaan Cottenier]

> Altough
> the results now are the same than in the single mode run, I get 10 of all
the
> files which get the suffix of the processors (case.dmat,case.dmat,
> case.energy,case.scf1,case.scf2, etc) but I only have 8 processors.
> My .machines is:
>
> Can I ignor this or does anyone can tell me what is going on there?

This is correct behaviour. Your number of k-points is not commensurate with
your number of processors. Imagine you have 4 identical machines and 50
k-points. With this .machines:

1:machine1
1:machine2
1:machine3
1:machine4

you will e.g. (depending on the granularity) give 12 k-points to each
machine, after which two machines will execute each one k-point (while the
other two are idle). So you will have 6 jobs. If you don't like this, you
can control yourself the distribution of k-points by putting the actual
k-point numbers (and granularity:1) in the .machines:

13:machine1
13:machine2
12:machine3
12:machine4
granularity:1

Now you will have only 4 files of each. If you accidentally have another
free machine available (an old slow one would be ideal in this case), you
could do it like this:

12:machine1
12:machine2
12:machine3
12:machine4
2:slowmachine
granularity:1

Such a file is most efficient: all fast machines are always working, never
waiting for each other, and this thanks to your old slow machine.

> Also the testpara1 and testpara2 are showning me that sometimes on of
these
> computers are used more than twice, but they only have two processors.

They are used twice *one after the other*, not simultaneously, see above.

> Another question: Why is the .machines file like
>
> granularity:1
> 1:linux4gb:2
> 1:linux3gb:2
> 1:linux2gb:2
> 1:linux1gb:2
>
> is not running? I thought I am not using MPI but the error message is:

Guess: omit the :2 and list dual-cpu machine twice:

1:linux4gb
1:linux4gb
1:linux3gb
1:linux3gb
1:linux2gb
1:linux2gb
1:linux1gb
1:linux1gb
granularity:1

Stefaan