[CPMD-list] CPMD parallel scalability

[Alessandro Curioni]

Maurice,

 due to your system size you should be away from the intrinsic scalability 
  limit of your test case -   so it is probably related to memory 
bandwidth saturation in your machine -
  even smallest 32 water molecules systems can scale up   to thousands 
processors  on more balanced machines -

in any  cases  -  since you have  a lot of memory available - try to add 
the following keywords :

REAL SPACE WFN KEEP  ( will  avoid a full loop of  ffts  keeping  real 
space rep of wfn in memory)
and /or
ALLTOALL SINGLE  ( will perform   3Dfft transposition in reduced 
precision)

if  the  limit causing  performance degradation  was intrinsic load 
imbalance in the 3D-FFT 
also using TASKGROUPS will help -  but I do not think is your case  ( 
indeed  it could become worst - being bandwidth the limitation) -



Alessandro



Alessandro CURIONI, PhD
Manager, Computational Sciences
IBM Research Division - Zurich Research Laboratory
Saumerstrasse 4
8803 Rueschlikon - Switzerland
e-mail: cur at zurich.ibm.com
www:    www.zurich.ibm.com
Tel: +41-44-7248633
Fax: +41-44-7248958




Maurice de Koning <dekoning at ifi.unicamp.br> 
Sent by: cpmd-list-bounces at cpmd.org
04/14/2008 03:58 PM

To
cpmd-list at cpmd.org
cc

Subject
[CPMD-list] CPMD parallel scalability






Hi all,

I´m running CPMD on an Altix 4700 system with 44 CPU´s and 88 Gb of RAM 
memory.
At the moment I´m running a CP MD run of a cell containing 96 water 
molecules using the
BLYP functional at 300 K. I noticed that the scalability is not very 
good. If I run on more than
about 16 CPU´s, the time per MD step starts even increases, such that 
the total time starts growing with the
number of CPU´s. Is there anything I can do about this?
Below is a part of the input script

Cheers,

Maurice

&CPMD
  MOLECULAR DYNAMICS CP
  RESTART WAVEFUNCTION COORDINATES VELOCITIES
  QUENCH BO ELECTRONS
  TIMESTEP
  4
  TEMPCONTROL IONS
  300.0 40
  MAXSTEP
  20000
  TRAJECTORY XYZ
  EMASS
  600
  STORE
  100
&END

&SYSTEM
  SYMMETRY
  8
  SCALE
  ANGSTROM
  CELL ABSOLUTE
  13.1493536801862057 15.1976493759670106 14.3247149284289197 0 0 0
  CUTOFF
  70
&END

&DFT
FUNCTIONAL BLYP
&END

_______________________________________________
CPMD-list mailing list
CPMD-list at cpmd.org
http://cpmd.org/mailman/listinfo/cpmd-list

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cpmd.org/pipermail/cpmd-list/attachments/20080414/ad195b1a/attachment.html