[Pw_forum] error on LINUX-MPI

[Giovanni Cangiani]

Hi Sergey,
	you have first to start the lam demon on the nodes
where you want to run the program:

# lamboot mynode

 
LAM 6.5.9/MPI 2 C++/ROMIO - Indiana University
 
 
---
where mynodes is a file containg the list of machines where you want
to run the program as for example:

cnode0
cnode1
cnode2
cnode3

If this works you can check the status of lam with

# lamnodes
n0      cnode0:1
n1      cnode1:1
n2      cnode2:1
n3      cnode3:1
---

At this point you can run your program with the standard mpirun command:

# mpirun -np 4 /home/cangiani/bin/pw.x < in > out


Once you no longer need LAM you can stop it with 
# lamhalt

LAM 6.5.9/MPI 2 C++/ROMIO - Indiana University

---


 
Please note that 
 1) the executable must be accessible by all nodes (e.g. give an
absolute path and put it in a shared directory or copy it on each node).
 2) with the option -np NN you specify as usual the number of processes
you intend to run. I think that in your case (dual cpu nodes) you simply
have to ask for NN=twice the number of nodes you intend to use.
 3) If you have a queing system like PBS, then it is more complicated
because you have to know which nodes has been assigned to your job.
Normally, there should be a way to obtain the node list from the queing
system.


Hope this helps. If not, please read LAM documentation at 
http://www.lam-mpi.org/using/docs/   which is very good.


giovanni


On Tue, 2003-07-08 at 11:56, Sergei Lisenkov wrote:
> Dear Giovanni ,
> 
> Thank you for your help, I compiled it. I have one question: How to run PW on PC-clusters using lam?
> My cluster is 15 dual PC-pentium3 (2x15=30 CPU). I know that running PW is machine-dependent. Can you help me?
> 
> Thank you,
> Best wishes,
> Sergey
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum