[Pw_forum] time for calculation of el-ph interactions of graphene supercell

[Axel Kohlmeyer]

On Sat, 6 Sep 2008, L.F.Huang wrote:

LFH> PostScript:
LFH>   Amount of memory: 2GB per node
LFH>   kind of interconnect: myrinet 2000
LFH>   No. of reduced k points: 52
LFH>   No. of nodes: 13 (with npool by default 1)

parallelization over k-points is always the 
first item to try. not all components in Q-E
support it fully, but when it is supported,
it is superior over the g-space parallelization,
which in itself has scaling limitationss. please
see how in my previous demonstration to vega,
the execution time goes up significantly, when
using too many nodes and the wrong parallelization
scheme. i would also check, whether your machine
is communicating correctly. i've seen myrinet machines
just stall occasionally, due to firmware crashes.

LFH>   Parallel environment: MPI 1.2

i assume it is MPICH-GM. mpi 1.2 does not make that
much sense (there is an MPI 1.2 standard but that 
numbering is usually independent of the MPI standard
version. e.g., i'm using OpenMPI 1.2.7 and it implements
all of MPI 2.0 ... ).

LFH>   Math library: NAG ACML 1.5、ATLAS 3.6

ACML is from AMD not NAG, and that would conflict 
with ATLAS as both implement BLAS/LAPACK.

LFH>   No. of cpus per node: 4

which type? looks like opteron dual-core?

cheers,
   axel.

-- 
=======================================================================
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.