Setting up a Beowulf Cluster: Fastforward to the last WEEK

Installation of Open SSH
First we have to install SSH in all the nodes. For this the following code needs to be performed:

$ sudo apt-get install openssh-server

$ sudo apt-get install openssh-client

Configuring Passwordless SSH
Now the SSH is installed, but it asks for password every time any code is run. This becomes a major problem when we have to run a large block of codes in MPI. So we have to make it passwordless. 
We have to generate an SSH key for all MPI users on all nodes, if we generate the SSH key for the Master node, all the nodes will automatically have an SSH key when we mount the home directory.

$ sudo ssh-keygen –t rsa

When asked for a passphrase leave it empty, hence the SSH will become passwordless. 

. Now for the master to login to the slave nodes, the public key of the master node needs to be added to the list of known hosts, i.e. ~/.ssh/authorized_keys of all the slav nodes. Since we are mounting all the slave nodes to the master’s home directory we have to do this just once and in the masters own home directory. Now we can SSH the nodes fom the master.

mpiuser@prithvi-desktop:~$ ssh jal02@192.168.5.102

mpiuser@jal02-desktop:~$ echo $HOSTNAME

jal02-desktop

We can see that we have to type the ipaddress as well, so to avoid this, we have to add the ipaddress and the names in the /etc/hosts file.

127.0.0.1 localhost

192.168.5.100 prithvi

192.168.5.101 vayu01

192.168.5.102 jal02

192.168.5.103 agni03

192.168.5.104 dharti04

192.168.5.105 jeevan05

For passwordless SSH configuration the following website was allot helpful:

http://www.tecmint.com/ssh-passwordless-login-using-ssh-keygen-in-5-easy-steps/

Compiling and Running MPI programs:

OpenMPI can be used with a variety of languages, two of the most popular are FORTRAN and ‘C’. If your programs are written in ‘C’, then you can either use mpicc instead of your normal ‘C’ compiler or you can pass the additional arguments directly to your ‘C’ compiler. With mpicc the arguments you pass are passed to your normal ‘C’ compiler.

If you want to use mpicc to compile a ‘C’ source file called helloworld.c:

$ mpicc helloworld.c

You can start a program on just one machine, and have it execute on multiple processors/cores. This is handy if you have a powerful machine or are debugging/testing a program. Alternatively, you can run the program over your cluster and request as many processes as you want to be run over it.

To run helloworld.c over five processes on the cluster using the .mpi_hostfile created above (Note that myprogram must be in the same location on each machine):

$ mpirun -np 6 --hostfile .mpi_hostfile ./a.out

Remember the .mpi_hostfile we created, yes the one with all the names of the nodes. Always remember to include this while u are running your mpi programs other wise the program wont run in the nodes. 

heloworld.c

#include "mpi.h"

   #include <stdio.h>

   int main(argc,argv)

   int argc;

   char *argv[]; {

   int  numtasks, rank, len, rc;

   char hostname[MPI_MAX_PROCESSOR_NAME];

   rc = MPI_Init(&argc,&argv);

   if (rc != MPI_SUCCESS) {

     printf ("Error starting MPI program. Terminating.\n");

     MPI_Abort(MPI_COMM_WORLD, rc);

     }

   MPI_Comm_size(MPI_COMM_WORLD,&numtasks);

   MPI_Comm_rank(MPI_COMM_WORLD,&rank);

   MPI_Get_processor_name(hostname, &len);

   printf ("Number of tasks= %d My rank= %d HELLO from %s\n", numtasks,rank,hostname);

   /*******  do some work *******/

   MPI_Finalize();

   }