Re: [ESPResSo] Re: Parallel Simulations and System Initialization

[Axel Arnold]

Am Sonntag 03 Februar 2008 23:22 schrieb Lorenzo Isella:
> Hello,
> Thanks for the reply. Before delving into the intricacies of setting up
> the environmental variables in OpenMPI, I wonder if there is any
> parallelization library which is automatically recognized and supported
> by Espresso.
> I would like to keep it as simple as possible at this stage (also
> because I do not have a view of which library I should pick up right now).
> In other words: do you (or does anyone on the list) know a library which
> allows Espresso to take advantage of having 2 CPU's at its own disposal
> once it has been properly installed on the system? Does such a library
> exist as a Debian package?
> If the answer is not to both question, then I think I am left with
> learning how to set up the OpenMPI environment on my system.

There are two packages that Espresso is known to work with: LAM/MPI and MPICH. 
However, this will not solve your problems: LAM also requires you to prepare 
your computer for parallel runs; this has nothing to do with Espresso itself, 
but is a requirement of the parallel library. In case of LAM, for example, 
you need to execute lamboot before running your application, and lamhalt 
after running it. I think MPICH has a mode that does not require an external 
daemon, but is typical less performant than LAM.

The error that you obtain from LAM regarding "OpenIB" indicates that somehow 
your OpenMPI installation tries to use an non-existing infiniband-adaptor. 
Maybe you first want to check why OpenMPI wants to use Infiniband...

Regarding the random seed: you have to specify the seeds for the random 
generators on all nodes. If you want to specify just one number, then you can 
easily use the Tcl builtin RNG to generate seeds:
expr srand(<seed>)
set l ""; for {set i 0} {$i < [setmd n_nodes]} {incr i} {
 lappend l [expr int(32768*rand())]
}
eval t_random seed $l

Axel

-- 
Dr. Axel Arnold 
Fraunhofer Institute for Algorithms and Scientific Computing (SCAI)
Schloss Birlinghoven
53754 Sankt Augustin
Germany