Running in parallel on multiple nodes

kalyangoparaju · January 18, 2012, 00:33

Friends,

I was trying to run a case by using resources of 2 computers by using the following command.

mpirun --hostfile <machines> -np <nprocs> snappyHexMesh -parallel

When I script without the hostfile on 1 node with 8 processors, I don't get any errors but when I run the same script on 2 nodes using the machines names in the machines hostfile, I get an error saying

cannot find points in directory polymesh from 0 down to constant.

I tried checking if the constant directory had the polymesh directory and the points file in it and apparently, it does.

Can someone please help me. Where is it that I am going wrong?

regards,
Kalyan Goparaju

bastil · January 18, 2012, 05:01

Hi,

you need to decompse your model in order to run in parallel. Run decomposePar in order to do this.

Regards Bastian

kalyangoparaju · January 18, 2012, 10:22

Bastil,

I did do that. The following are the steps I followed.

1. blockMesh
2. decomposePar
3. mpirun --hostfile machines -np <nprocs> snappyHexMesh -parallel

The problem what I mentioned is when I do the third step.

Kalyan

elvis · January 18, 2012, 11:41

Hi,

do you have the files for the "simulation calculation" on both machines?
And do you have the same filesystem structure on both machines?

kalyangoparaju · January 18, 2012, 12:13

Elvis,

As I understand, using MPI with machine files doesn't require us to have the working folder in both the system.

To answer your question, no I don't have the file in the slave nodes. But I will give it a shot now and see if it works.

Kalyan

Update - Elvis, I did put the folder in both the nodes and tried running. I get the same error

olivierG · January 18, 2012, 12:36

hello,

In fact you need to have the working folder in both system (usually via nfs shared file system).

And you also need that snappyHexMesh can work on both system (so same as the working folder, you need openfoam on nfs, or install openfoam on the same dir on each machine).

You aslo need to source your bashrc on each node. On way to do this is to use foamExec.

And take a look at ssh access with shared key (doesn't need a password for each node).

regards,
olivier

January 18, 2012, 00:33	Running in parallel on multiple nodes	#1
kalyangoparaju Member Kalyan Join Date: Oct 2011 Location: Columbus, Ohio Posts: 53 Blog Entries: 1 Rep Power: 15	Friends, I was trying to run a case by using resources of 2 computers by using the following command. mpirun --hostfile <machines> -np <nprocs> snappyHexMesh -parallel When I script without the hostfile on 1 node with 8 processors, I don't get any errors but when I run the same script on 2 nodes using the machines names in the machines hostfile, I get an error saying cannot find points in directory polymesh from 0 down to constant. I tried checking if the constant directory had the polymesh directory and the points file in it and apparently, it does. Can someone please help me. Where is it that I am going wrong? regards, Kalyan Goparaju

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
problem of running parallel Fluent on linux cluster	ivanbuz	FLUENT	15	September 23, 2017 20:12
running OpenFoam in parallel	vishwa	OpenFOAM Running, Solving & CFD	22	August 2, 2015 09:53
multiple parallel jobs on one machine	joeybernard	CFX	0	December 16, 2010 11:10
Running dieselFoam in parallel.	Palminchi	OpenFOAM	0	February 17, 2010 05:00
Minimum number of nodes to run CFX in parallel	Rui	CFX	3	April 11, 2005 21:46

January 18, 2012, 05:01		#2
bastil Senior Member BastiL Join Date: Mar 2009 Posts: 530 Rep Power: 20	Hi, you need to decompse your model in order to run in parallel. Run decomposePar in order to do this. Regards Bastian

January 18, 2012, 10:22		#3
kalyangoparaju Member Kalyan Join Date: Oct 2011 Location: Columbus, Ohio Posts: 53 Blog Entries: 1 Rep Power: 15	Bastil, I did do that. The following are the steps I followed. 1. blockMesh 2. decomposePar 3. mpirun --hostfile machines -np <nprocs> snappyHexMesh -parallel The problem what I mentioned is when I do the third step. Kalyan

January 18, 2012, 11:41		#4
elvis Senior Member Elvis Join Date: Mar 2009 Location: Sindelfingen, Germany Posts: 620 Blog Entries: 6 Rep Power: 24	Hi, do you have the files for the "simulation calculation" on both machines? And do you have the same filesystem structure on both machines?

January 18, 2012, 12:13		#5
kalyangoparaju Member Kalyan Join Date: Oct 2011 Location: Columbus, Ohio Posts: 53 Blog Entries: 1 Rep Power: 15	Elvis, As I understand, using MPI with machine files doesn't require us to have the working folder in both the system. To answer your question, no I don't have the file in the slave nodes. But I will give it a shot now and see if it works. Kalyan Update - Elvis, I did put the folder in both the nodes and tried running. I get the same error

January 18, 2012, 12:36		#6
olivierG Senior Member Olivier Join Date: Jun 2009 Location: France, grenoble Posts: 272 Rep Power: 18	hello, In fact you need to have the working folder in both system (usually via nfs shared file system). And you also need that snappyHexMesh can work on both system (so same as the working folder, you need openfoam on nfs, or install openfoam on the same dir on each machine). You aslo need to source your bashrc on each node. On way to do this is to use foamExec. And take a look at ssh access with shared key (doesn't need a password for each node). regards, olivier