Thanks Brian!!
I think that it works now.
So I suppose that  I should eave that lamboot commands inside every script I run
and not worry about lamboot outside the pbs environment at all!!

I do have a few other questions that I hope you can answer. I have two slave
nodes and four processors. PBS recognized all of them at setup. Now using the
mpirun -np 2 option,  i see that the process runs on two processors of the same
node. How do I make the process run on both the slave nodes and on all four
processors? I tried a variety of options but none of them seem to work. I am
pretty sure that it is fairly easy but I don't seem to get it. All I need to do
is control the number of processors the job runs on.

Thanks.

Senthil



"Brian W. Barrett" wrote:

>
> This is an easy one (although probably not obvious until you stop and
> think about it).  When using PBS, you are never supposed to do anything
> "outside of PBS" - you should use PBS for everything, including starting
> up the LAM runtime environment (running lamboot).  This is actually pretty
> easy - in your case, all you need to do is have a PBS script that looks
> something like this:
>
> ----
> #PBS -S /bin/sh
> #PBS -l nodes=1
> #PBS -q normal
> #PBS -N tielema
> #PBS -j oe
> echo "I ran on `hostname`"
>
> # bring up LAM environment on allocated nodes
> # (which are listed in the filename contained in $PBS_NODEFILE)
> lamboot $PBS_NODEFILE
>
> # use mpirun to run my MPI binary with all allocated nodes
> mpirun C ls
>
> # shut down LAM environment
> lamhalt
> ---
>
> You were getting the "no lamd running" error message because LAM tries to
> be smart under PBS.  If setup a certain way, PBS can allocate two seperate
> jobs to the same node.  If LAM wasn't smart, the two jobs would collide -
> so we have to play some tricks to avoid colliding - hence the error
> message that you saw.  Since your lamboot was not under the same PBS job,
> it was completely invisible to the mpirun running under PBS.
>
> > Hope to get this "computer" problem fixed ASAP so that I can worry about
> > "science"
>
> That's what we're here for.  Hopefully, this should be the end of it:).
>
> Brian
>


_______________________________________________
Oscar-users mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/oscar-users

Reply via email to