[CPMD-list] cpmd352 on dual-PCs - shmem communication feasible?

[Christian Tuma]

Hello,

I am using CPMD 3.5.2 on a dual-processor PC together with MPICH 1.2.4.
As far as I can see parallel processes communicate per TCP/IP sockets
by default. This way a 2-processor job shows a speedup of something
between only 1.3 and 1.4 instead of nearly 2 (on a 2-processor PC).
The same 2-processor job on two different one-processor PCs shows a 
speedup of about 1.8 on the other hand! So I thought it would be nice 
to have the 2-processor job on the 2-processor PC communicate via 
shmem-routines instead of TCP/IP sockets. Then I rebuild MPICH with
the configure option "--with-comm-shared" and rebuild CPMD (linked
to the new MPICH, nothing else changed). When I try to start the 
2-processor job now as a shmem-job (with the -p4pg option to mpirun 
as decribed in the MPICH documentation) the jobs starts but crashes 
here:

...
    |  Number of Mesh Points :   511                           |  
    |  Pseudoatom Total Energy   -0.449882                     |  
    ============================================================  

 ****************************************************************
 *   ATOM       MASS   RAGGIO NLCC              PSEUDOPOTENTIAL *
 *      O    15.9994   1.2000   NO    KLEINMAN      S  NONLOCAL *
 *                                                  P     LOCAL *
 *      H     1.0080   1.2000   NO                  S     LOCAL *
 ****************************************************************

p1_8265:  p4_error: interrupt SIGSEGV: 11
P4 procgroup file is pgfile.

To me this error message looks as if CPMD causes a problem now.

Questions: 
Do I have to compile CPMD with some other switches now?
Can CPMD do the communication via shmem under Linux at all?

Did anybody manage to run parallel CPMD jobs on 2-processor PCs
communicating via shmem-routines or to get a speedup of around
the expected value of 2?

Any advice is welcome.  Thank you!

Christian.

-- 
Christian Tuma             Humboldt-Universitaet Berlin
ct at chemie.hu-berlin.de     Arbeitsgruppe Quantenchemie (Prof. Sauer)
phone: +49-30-20937140     Brook-Taylor-Str. 2, 12489 Berlin, GERMANY
fax:   +49-30-20937136     http://www.chemie.hu-berlin.de/ag_sauer