Program does not finish when two nodes are used

Question

When I run my program in one computer, with 4 processes, it will terminate almost instantly. When I try to run it in a cluster of two computers (the cluster was checked and was OK), it just won't finish!

I had done a run a long time ago in the cluster and I remember that it was slower than in the run on the one pc, but it would terminate!

Here is my run.sh:

#!/bin/bash

start=100
end=100
for ((i = $start; i <= $end; ++i )) ; 
do
        mpiexec -f machinefile -n 4 ./test ../../l_matrices/Lmat_755.mtx 1 755 755 $i $i 2 2 0 0
done

and I did check that two processes are spawned in every node.

Here is my machinefile:

hostname1.gr:2
hostname2.gr:2

What is happening?

Oh yes @πάνταῥεῖ, I was planning to post some code too, but I didn't, so it should be removed, thanks! — gsamaras, Aug 03 '15 at 18:24
are you sure that the submission script is correct? the loop will enter terminate after one iteration? intentional? — Sidharth N. Kashyap, Aug 03 '15 at 20:15
@Sidharth.N.Kashyap, yes the script it's correct. It works fine with one computer, the problem appears only when I use two. — gsamaras, Aug 03 '15 at 23:34

score 1 · Accepted Answer · answered Aug 04 '15 at 11:44

1

(Presuming that the script is correct and the start and end variable values are intentional, this will not do anything meaningful as mpiexec will execute two copies of the same file with the same arguments once)

Check the paths - You have used relative paths, that can lead to problems as the execution happens in the default directory after login, typically, your home directory.

answered Aug 04 '15 at 11:44

Sidharth N. Kashyap

299
2
6

1

Just for the history, I have changed something in the code, which causes this behaviour. I just tried an older version of my project and it works just fine. – gsamaras Aug 04 '15 at 12:40

Program does not finish when two nodes are used

1 Answers1