[Wien] Anyone ever compiled with pessl before?

[Jorissen Kevin]

The machines-file  you describe is suited only for fine grain.  For k-point, you'll have to write 32 lines, each corresponding to one processor ;-)
If you still have more k-points than processors, k-point parallellization is normally most efficient.  Only for very big systems, where there are just a few (or one) k-point, and where very large matrix sizes are necessary, fine grain par. is necessary.
 
Kevin.
 

	-----Oorspronkelijk bericht----- 
	Van: Chiung-Yuan Lin [mailto:cylin at buphy.bu.edu] 
	Verzonden: ma 3/22/2004 11:22 
	Aan: wien at zeus.theochem.tuwien.ac.at 
	CC: 
	Onderwerp: [Wien] Anyone ever compiled with pessl before?
	
	

	Dear all,
	 First I would like to thank Vladimir's advice on PESSL and SCALAPACK.
	
	 Here I have other questions:
	i)  Can I run k-point parallelization (a 4000-k-point case) on
	    a single queue (with 32 processors) of our machine? If yes, do I make
	    my .machines file
	
	    1:hostname:32
	
	    and then x run_lapw -c -p? Are any other commands like granularity or
	    extrafine recommended in my .machines file?
	ii) For fine grained parallelization, do I make the same .machines file
	    as i)? The command x run_lapw -c -p seems to initialize k-point
	    parallelization. Is there a way to run fine grained using run_lapw,
	    or I must use some other ways to initialize fine grained?
	
	 I ask because I want to know which paralleliztion scheme (purely fine
	grained or purely k-point parallelization) makes my case run faster?
	
	Thank you
	Chiung-Yuan
	
	
	Vladimir Timoshevskii wrote:
	>
	> Dear Chiung-Yuan,
	>
	> We have here IBM-SP4 machine with PESSL, and we tried to compile a
	> parallel version.
	> Unfortunately, PESSL is not completely compatible with SCALAPACK (on the
	> MPI level). You need either to change the code yourself, or to compile
	> the standard SCALAPACK.
	> However, standard scalapack appeared to be not very efficient for IBM.
	>
	> Good luck,
	> Vladimir.
	
	_______________________________________________
	Wien mailing list
	Wien at zeus.theochem.tuwien.ac.at
	http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
	
	

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/ms-tnef
Size: 5658 bytes
Desc: not available
Url : http://zeus.theochem.tuwien.ac.at/pipermail/wien/attachments/20040323/e645f649/attachment.bin