[CPMD-list] question on compile

[Axel Kohlmeyer]

On Thu, 22 Mar 2007, qfzhang wrote:

QZ> Hi,dear all!
   I just compile CPMD code on the new machine.But it seems that the effect of p
arallelism is very bad, 64 CPUs cost no less time than 32 CPU for the same job. 

no. it is not the parallelization, it is you inadequate network
bandwidth and too high communication latency. with the kind of setup
you have, you're probably throttled out by the network after using 
2-4 nodes. you'd need infinitband, dolphin/SCI, myrinet or quadrics
or something similar to scale beyond that. see, e.g.,

http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/cpmd-bench.html#parallel

for some very old, but still very true discussion on that topic.

cheers,
  axel.


there are really some problem during compiling.
The Makefile is attached here. and the hardware of the machine is as below.
   'Ada consists of 252 nodes with: 
   4 Xeon 3 GHz cores (dual dual core) 
   4GB RAM 
   ~100GB free local storage 
   MPI over gigbit etehernt'
 
   And the guide webpage is www.unicc.chalmers.se
   
   Best wishes
                                             Qianfan Zhang
                                              2007-03-22


-- 
=======================================================================
Axel Kohlmeyer   akohlmey at cmm.chem.upenn.edu   http://www.cmm.upenn.edu
   Center for Molecular Modeling   --   University of Pennsylvania
Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
tel: 1-215-898-1582,  fax: 1-215-573-6233,  office-tel: 1-215-898-5425
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.