Cbb/Baobab: Difference between revisions

From Computer Science Wiki
Jump to navigation Jump to search
← Older edit
VisualWikitext
Rhunter (talk | contribs)
Staff, Techstaff
321 edits
No edit summary
Rhunter (talk | contribs)
Staff, Techstaff
321 edits
No edit summary
 
(14 intermediate revisions by the same user not shown)
Line 5: Line 5:
The new cluster is made up of 6 nodes with 24 cores and 64 GB RAM on each node.
The new cluster is made up of 6 nodes with 24 cores and 64 GB RAM on each node.


Before submitting your jobs you need a PBS submission script


Test MPI program:
To use the baobab cluster setup your environment to use mpi.  I generally recommend you add this to your CBB .bashrc file before you get started.
<PRE>
<PRE>
/******************************************************************************
if [ ! -z `hostname | grep baobab` ]; then
* * FILE: mpi_hello.c
  module add mpi/openmpi-x86_64
* * DESCRIPTION:
fi
* *  MPI tutorial example code: Simple hello world program
</PRE>
* * AUTHOR: Blaise Barney
 
  * * LAST REVISED: 03/05/10
 
* ******************************************************************************/
Another recommendation is to create an ssh key pair for yourself, if you don't already have one. If your jobs are not running this is probably the reason why.
#include "mpi.h"
<PRE>cd .ssh
#include <stdio.h>
ssh-keygen -t dsa
#include <stdlib.h>
<enter><enter><enter> [just take the defaults and DON'T set a pass phrase]
#define  MASTER        0
cat id_dsa.pub >> authorized_keys</PRE>


int main (int argc, char *argv[])
{
int  numtasks, taskid, len;
char hostname[MPI_MAX_PROCESSOR_NAME];


MPI_Init(&argc, &argv);
After you have done this, you can ssh to the baobab headnode.  You can do this from any machine connected to the internal CBB network.
MPI_Comm_size(MPI_COMM_WORLD, &numtasks);
<PRE>
MPI_Comm_rank(MPI_COMM_WORLD,&taskid);
ssh baobab.cbb.lan
MPI_Get_processor_name(hostname, &len);
printf ("Hello from task %d on %s!\n", taskid, hostname);
if (taskid == MASTER)
  printf("MASTER: Number of MPI tasks is: %d\n",numtasks);
MPI_Finalize();
}
</PRE>
</PRE>


Example PBS script:
 
Add the node hostkeys to your known_hosts (Only needs done once)
<PRE>pdsh -w baobab-[1-6] "uptime"</PRE>
 
 
Please use the PBS submission system to run your jobs.  Here is a quick rundown of what you will need to know.
 
Make yourself a PBS submission script like the following:
<PRE>
<PRE>
#PBS -l nodes=6:ppn=24,walltime=00:01:00
#PBS -l nodes=6:ppn=24,walltime=00:01:00
module add openmpi-x86_64
## nodes = total number of nodes you need
########################################
## ppn = processors per node that you will need
## walltime = amount of time your job will be allowed before being forcefully removed
#####################################################################################
cd $PBS_O_WORKDIR
cd $PBS_O_WORKDIR
## How many cores total do we have?
## How many cores total do we have?
Line 48: Line 46:
## Main execution
## Main execution
echo "Job Started at: `date`"
echo "Job Started at: `date`"
mpiexec -np $NO_OF_CORES -machinefile $PBS_NODEFILE hello2
mpiexec -np $NO_OF_CORES -machinefile $PBS_NODEFILE hello
echo "Job Ended at: `date`"
echo "Job Ended at: `date`"
</PRE>
</PRE>


To submit the jobs to the cluster from the headnode:
To submit the jobs to the cluster from the headnode:
Line 56: Line 55:
qsub <yoursPBSscript>
qsub <yoursPBSscript>
</PRE>
</PRE>


To check the status of your jobs:
To check the status of your jobs:
Line 61: Line 61:
qstat -a
qstat -a
</PRE>
</PRE>


To delete your jobs from the queue:
To delete your jobs from the queue:
Line 66: Line 67:
qdel <Job_Number/Identifier>
qdel <Job_Number/Identifier>
</PRE>
</PRE>
For more examples look here: https://wiki.cs.vt.edu/wiki/Cbb/Baobab/examples

Latest revision as of 09:16, 11 August 2015

The Baobab Cluster

CBB's internal cluster.

The new cluster is made up of 6 nodes with 24 cores and 64 GB RAM on each node.


To use the baobab cluster setup your environment to use mpi. I generally recommend you add this to your CBB .bashrc file before you get started. 

if [ ! -z `hostname | grep baobab` ]; then
  module add mpi/openmpi-x86_64
fi


Another recommendation is to create an ssh key pair for yourself, if you don't already have one. If your jobs are not running this is probably the reason why. 

cd .ssh
ssh-keygen -t dsa
<enter><enter><enter> [just take the defaults and DON'T set a pass phrase]
cat id_dsa.pub >> authorized_keys


After you have done this, you can ssh to the baobab headnode. You can do this from any machine connected to the internal CBB network. 

ssh baobab.cbb.lan


Add the node hostkeys to your known_hosts (Only needs done once) 

pdsh -w baobab-[1-6] "uptime"


Please use the PBS submission system to run your jobs. Here is a quick rundown of what you will need to know.

Make yourself a PBS submission script like the following: 

#PBS -l nodes=6:ppn=24,walltime=00:01:00
## nodes = total number of nodes you need
## ppn = processors per node that you will need
## walltime = amount of time your job will be allowed before being forcefully removed
#####################################################################################
cd $PBS_O_WORKDIR
## How many cores total do we have?
NO_OF_CORES=`cat $PBS_NODEFILE | egrep -v '^#'\|'^$' | wc -l | awk '{print $1}'`
##
## Main execution
echo "Job Started at: `date`"
mpiexec -np $NO_OF_CORES -machinefile $PBS_NODEFILE hello
echo "Job Ended at: `date`"


To submit the jobs to the cluster from the headnode: 

qsub <yoursPBSscript>


To check the status of your jobs: 

qstat -a


To delete your jobs from the queue: 

qdel <Job_Number/Identifier>


For more examples look here: https://wiki.cs.vt.edu/wiki/Cbb/Baobab/examples

Retrieved from "http://wiki.cs.vt.edu/index.php?title=Cbb/Baobab&oldid=3708"