[Wien] machines file ... again

[Torsten Andersen]

Hello,

it is ALWAYS more convenient (and much faster) to use k-point 
parallisation if it is at all possible (when one has more than one 
k-point). The k-point parallelisation generates a case.klist for each 
set of k-points and runs the usual cycle on each of them, with 
appropriate summing of different quantities in between.

Whether you use one or two processors on a two-processor machine is not 
so important when the two programs doesn't have to "talk" to each other, 
like in k-point parallelisation. Also, here the network speed is not 
such a big issue - just use a separate local scratch partition on each 
machine. With MPI you should avoid the network (use a good crossbar in 
stead), and forget all about myrinet, Gbit ethernet, and all the other 
ways of getting money out of people for a meager performance increase.

Best regards,
Torsten Andersen.

Griselda Garcia wrote:
> Hello! Kevin
> 
> Yes, I remember that discussion however I considered that was important when 
> you try to use the two processors of the same machine but in my case I am 
> trying to use one processor of diferent dual machines.
> 
> But, if what you cited it is true, It is convenient to use th K-point 
> parallelization in any way. 
> So, I have other question, when the system to study have several atoms and 
> one K-point, how do you run the code in parallel?, the running time will be 
> similars in serial and K-point-parallel version?
> 
> Thanks!!
> 
> Griselda.
> 
> 
> 
> _______________________________________________
> Wien mailing list
> Wien at zeus.theochem.tuwien.ac.at
> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
> 

-- 
Dr. Torsten Andersen        TA-web: http://deep.at/myspace/
AG Hübner, Department of Physics, Kaiserslautern University
http://cmt.physik.uni-kl.de    http://www.physik.uni-kl.de/