Some tests on Opteron cluster (strange???)

[ilya]

Hi, All  !!!

Finally. I've compiled it !!!
Very nice program for QMMM MD.

I've performed some tests on our Sun Fire X2100 cluster of Dual Core
Opteron 175 Processors.

I compiled with Intel Fortran 10.1.011
used MKL 10.0.2.018, FFTW3 3.1.2, OpenMPI 1.2.6

Benchmarked on H2O-32 from tests/QS/benchmark

Here the results (sorry for formatting)

 n thread tot | n proc  /  n thread per proc | cpu time per proc, sec
| wall time, sec |  speedup
----------------------------------------------------------------------------------------------------------------------------------
     1        |          1/1
|                          1703  |     1705      |    1.00
     2        |          1/2
|                          3282  |     1734      |    0.98
     2        |          2/1
|                            961  |      962      |     1.77
     4        |          4/1
|                            555  |      556      |    3.07
     4        |          2/2
|                          1787  |      971      |    1.76
     8        |          4/2
|                          1014  |      564      |    3.02
     8        |          8/1
|                            349  |      349      |    4.89

First of all setting OMP_NUM_THREADS=2 and running 1 MPI process per
node makes performance worse (lines 2, 5 and 6) compared to 2 MPI
processes per node (we have 1 dual core CPU per node).
Is it normal? CPU time still doubles.

And is it a normal scaling behavior? ( 4.89 on 8 procs is not very
good. I know the system is small but anyway).

Thanks.