order of assign MPI_rank to nodes

[Federico Paladin]

> True again.  This sounds like a more risky approach as now you have to
> hope all the nodes stay up between your jobs, and convince the scheduler
> to allocate the same set of nodes again, in exactly the same order.  I'm
> not sure if anything in PBS guarantees the same node _order_ for a given
> node _set_.  Mpiexec just uses what it is given by PBS.
> 
> I suppose you could manage a different order by remembering the original
> set of nodes then building a -conf file in your second batch job to
> explicitly set the order:
> 
>     % cat conf
>     node03 : myjob
>     node01 : myjob
>     node02 : myjob
> 

On this point, I have just done some tests with some stupid mpi program
(myrank), but it seems not work: indipentently of the order of the conf
file, the rank seems to be assigned respecting the order of
$PBS_NODEFILE.

For example in a PBS interactive session:
> cat $PBS_NODEFILE
node003
node002
node001
> mpiexec myrank
rank 0 on node003
rank 1 on node002
rank 2 on node001
> cat conf
node001 : myrank
node003 : myrank
> mpiexec -conf conf
rank 0 on node003
rank 1 on node001

>From the last command I would expect:
rank 0 on node001
rank 1 on node003

So it seems that same node _set_, same _order_!

Thanks, 
  Federico