Hi Pratik,

May I ask what kind of connection do you have between the two nodes when you run in parallel? Is it Gigabit Ethernet or something else?

I would like to do the same thing but now have to pause as your experience.

Thanks,
Rolly

On 05/25/2016 06:50 PM, PRATIK DAS wrote:
Dear QE users,
I am not a new user of quantum espresso. Previously I worked on a single machine of 4 core. I simply worked with mpi by 'mpirun -np 4', and the job was faster than a serial run. Now I have a 2 node cluster. Each node consists of 2 processors of 12 core each. Hence I made a hostfile(for a single node) by specifying
"node1 slots=24".
Now my command is
"mpirun --hostfile my_hostfile -np 24 ph.x <input.in <http://input.in>> output.out". This run takes 9hours and 39 minutes.
But when I want to use both nodes by
"node1 slots=24
node2 slots=24"
& command
"mpirun --hostfile my_hostfile -np 48 ph.x <input.in <http://input.in>> output.out -ni 6 -nk 2", It is taking too much time to complete the job. Even it is now almost 26 hours over. 2nd q point calculation is going on.

NB: The job has only 4 q points.

I have also seen this in pw.x run. In a single node, it is taking only 2mins. Command: mpirun --hostfile my_hostfile -np 24 pw.x <input.in <http://input.in>> output.out.

But in case of both nodes, it is taking almost half an hour.
Command: mpirun --hostfile my_hostfile -np 48 pw.x <input.in <http://input.in>> output.out

Can anyone suggest me that why the jobs running on 2 nodes are slower than the jobs in single node?

Thanks in advance.


Regards
Pratik Kumar Das
/*Pratik Kr. Das*/
Research Fellow
High Pressure and Temperature Lab
Faculty of Science
Jadavpur University
Kolkata 700032



_______________________________________________
Pw_forum mailing list
[email protected]
http://pwscf.org/mailman/listinfo/pw_forum

--
PhD. Research Fellow,
Dept. of Physics & Materials Science,
City University of Hong Kong
Tel: +852 3442 4000
Fax: +852 3442 0538

_______________________________________________
Pw_forum mailing list
[email protected]
http://pwscf.org/mailman/listinfo/pw_forum

Reply via email to