How to install OpenMPI on Ubuntu (18.04) Cluster

Step 0: Create user and set ip addr

Step 1: Install ssh (ssh server and ssh client for all machines)

Step 2: Install nfs (nfs server on master, nfs client on nodes)

Step 3: Install OpenMPI in all machines

Step 4: Test

Step 0

Create user account for MPI and work on it

sudo adduser mpiuser
sudo usermod -aG sudo mpiuser
su mpiuser

Add all the hosts to server. Examples:

sudo nano /etc/hosts

Then add hostnames and ip addresses on Master machine.

Master machine is azcos1; others are worker nodes(azcos2, azcos3,azcos4,5,6)

127.0.0.1       localhost
10.1.20.1       azcos1
10.1.20.2       azcos2
10.1.20.3       azcos3
10.1.20.4       azcos4
10.1.20.5       azcos5
10.1.20.6       azcos6

You should add the master hostname on all nodes.

sudo nano /etc/hosts
10.1.20.1       azcos1

Step 1 — Install ssh

1. Run update for all

sudo apt-get update
sudo apt-get upgrade

2. Check ip addr and run

sudo apt-get install net-tools
ifconfig

3. Install ssh server and client for all machines

sudo apt-get install openssh-server openssh-client

4. Create directory .ssh for all machines, run

mkdir ~/.ssh
chmod 700 ~/.ssh

5. Generate ssh keys for passwordless ssh ( on Master machine)

cd ~/.ssh
ssh-keygen -t rsa
cp id_rsa.pub authorized_keys
chmod 600 authorized_keys
ssh-copy-id azcos2 (hostnames of all node machines, azcos3, azcos4, azcos5, azcos6)

6. Node machines — do it all nodes

cat id_rsa (do it on master machine)
vi id_rsa (paste it here)
chmod 600 id_rsa
vi authorized_keys
chmod 600 authorized_keys

7. Edit sshd_config file and restart ssh server

sudo nano /etc/ssh/sshd_config

Add this two lines.

PubkeyAuthentication yes
RSAAuthentication yes

Then restart ssh server.

sudo service ssh restart

8. Try to ssh the nodes without password

ssh azcos2 (and others)

Step 2 — Install nfs

1. Install NFS server on master machine

sudo apt-get install nfs-kernel-server

2. Create a directory and run:

sudo apt-get install nfs-kernel-server
sudo chown nobody:nogroup /home/mpiuser/Desktop/sharedfolder
sudo chmod 777 /home/mpiuser/Desktop/sharedfolder

3. Export files

sudo nano /etc/exports

then add these lines, save:

/home/mpiuser/Desktop/sharedfolder azcos2(rw,sync,no_subtree_check)
/home/mpiuser/Desktop/sharedfolder azcos3(rw,sync,no_subtree_check)
/home/mpiuser/Desktop/sharedfolder azcos4(rw,sync,no_subtree_check)
/home/mpiuser/Desktop/sharedfolder azcos5(rw,sync,no_subtree_check)
/home/mpiuser/Desktop/sharedfolder azcos6(rw,sync,no_subtree_check)

then

sudo exportfs -a

4. Restart NFS server, run

sudo systemctl restart nfs-kernel-server

5. Check firewall status(inactive), type:

sudo ufw status

6. We should install nfs client to all the nodes.

sudo apt-get install nfs-common
sudo mkdir -p /home/mpiuser/Desktop/sharedfolder
sudo chown nobody:nogroup /home/mpiuser/Desktop/sharedfolder
sudo chmod 777 /home/mpiuser/Desktop/sharedfolder
sudo mount azcos1:/home/mpiuser/Desktop/sharedfolder /home/mpiuser/Desktop/sharedfolder

7. Then test nfs

You can create a file in /sharedfolder. You can see this file on all machines where /home/mpiuser/Desktop/sharedfolder.

Step 3 — Install OpenMPI

1. Install gcc for all machines, run:

sudo apt-get install gcc
sudo apt-get install openmpi-bin openmpi-common libopenmpi-dev libgtk2.0-dev

2. Download and decompress file openmpi for all machines:

You can download file from official website (https://www.open-mpi.org/)

sudo wget https://www.open-mpi.org/software/ompi/v4.1/downloads/openmpi-4.1.1.tar.gz
sudo tar  -xvzf /home/mpiuser/openmpi-4.1.1.tar.gz

3. install OpenMPI for all machines

cd /home/mpiuser/Desktop/openmpi-4.1.1
sudo ./configure --prefix="/home/mpiuser/.openmpi"
sudo make
sudo make install

4. Export PATH, run:

export PATH=”$PATH:/home/mpiuser/.openmpi/bin”
export LD_LIBRARY_PATH=”$LD_LIBRARY_PATH:/home/mpiuser/.openmpi/lib”

5. Test by running mpirun or mpircc

If you see the below, it means successfully installed.

mpiuser@azcos6:~$ mpirun
--------------------------------------------------------------------
mpirun could not find anything to do.
It is possible that you forgot to specify how many processes to run
via the "-np" argument.
--------------------------------------------------------------------

Step 4 — Test

1. Compile mpi program

Go to the link , copy the code and paste it here:

vi mpi-prime.c

Compile mpi file.

mpicc mpi-prime.c -o ./outputfile

copy outputfile to the sharedfolder & check the file in the nodes sharedfolder

2. Run cluster

Go to shared folder on master machine, run:

cd /home/mpiuser/Desktop/sharedfolder
mpirun --hostfile /etc/hosts -np 5 ./outputfile

if you see like below, it is ok.

mpiuser@azcos1:~/Desktop/sharedfolder$ mpirun --hostfile /etc/hosts -np 5 ./outputfile
16 August 2021 04:31:31 PM
PRIME_MPI
  C/MPI version
An MPI example program to count the number of primes.
  The number of processes is 5
N        Pi          Time
1         0        0.006863
         2         1        0.000528
         4         2        0.000504
         8         4        0.000508
        16         6        0.000513
        32        11        0.000512
        64        18        0.000511
       128        31        0.000514
       256        54        0.000515
       512        97        0.000501
      1024       172        0.000741
      2048       309        0.001029
      4096       564        0.002891
      8192      1028        0.005456
     16384      1900        0.014796
     32768      3512        0.031765
     65536      6542        0.115813
    131072     12251        0.436057
    262144     23000        1.638243
    524288     43390        6.200569
   1048576     82025       23.463462
PRIME_MPI - Master process:
  Normal end of execution.
16 August 2021 04:32:03 PM
mpiuser@azcos1:~/Desktop/sharedfolder$

Thank you! :)