RC's Blog: Rocks cluster installation

Table of content:1. Hardware requirements2. Installation and configuration of frontend node3. Installation of compute node4. Start computing5. Getting cluster console

1. Minimum Hardware Requirements:

1.1 Frontend Node:

• Disk Capacity: 20 GB

• Memory Capacity: 512 MB (i386) and 1 GB (x86_64)

• Ethernet: 2 physical ports (e.g., "eth0" and "eth1")

1.2 Compute Node:

• Disk Capacity: 20 GB

• Memory Capacity: 512 MB

• Ethernet: 1 physical port (e.g., "eth0")

2. Install and Configuration of Frontend:

This section describes how to install your Rocks cluster frontend.

The minimum requirement to bring up a frontend is to have the following rolls:

• Kernel/Boot Roll CD

• Base Roll CD

• HPC Roll CD

• Web Server Roll CD

• OS Roll CD - Disk 1

• OS Roll CD - Disk 2

You can use “area51+base+bio+condor+ganglia+hpc+java+kernel+kvm+os+perl+python+service-pack+sge+web-server+zfs-linux-6.1.x86_64.disk1.iso” for the installation.

Download link:

http://www.rocksclusters.org/wordpress/?page_id=449

Steps:

2.1. Boot the frontend system from the ISO. You will get the following screen. For a frontend node or master node, simply write “build” and press enter as shown in the screen.

It will start loading packages and your installation will start.

2.2 After loading the packages from the ISO, you’ll see a screen that looks like:

Here, you can give the hostname as per your choice. I am leaving it default.

Click on “CD/DVD-based rolls” to select the rolls.

2.3 You will find the list of rolls available.

2.4 Now, select the rolls you want to use and click “submit”.

2.5. Next screen will show you the selected rolls in the left.

Click next and you will get the screen like this. Click Next.

2.6. Then you’ll see the Cluster Information screen:

And click next.

2.7. The private cluster network configuration screen allows you to set up the networking parameters for the Ethernet network that connects the frontend to the compute nodes.

Here, you can give the IP for the both of Ethernet cards. First give IP for ‘eth1’ and click next. This IP will be used for ‘public network’.

2.8. Now give IP for ‘eth0’. This IP will be used for the ‘private network’.

2.9. Configure the Gateway and DNS entries:

2.10. Give the root password:

2.11. Configure the time:

2.12. The disk partitioning screen allows you to select ‘automatic or manual’ partitioning.

I am selecting automatic partitioning.

Now the installation will start.

3. Installing your Compute Nodes:

3.1. Login to the frontend node as root.

3.2. Run a program which captures compute node DHCP requests and puts their information into the Rocks MySQLdatabase:

# insert-ethers

If you get this message, wait for a while and run the command again. This presents a screen that looks like:

Take the default selection, Compute, hit ’Ok’.

3.3. Then you’ll see:

3.4. Now boot a VM with ISO attached and with PXE settings.

Boot order should be: CD/DVD, Hard-disk and PXE.

Now boot the VM.

When the client VM will boot, it will show the following window.

Now it will start taking the installation files from PXE.

Its MAC address will be captured by the frontend and you will get the screen like this:

Note: ‘Insert-ethers’ has discovered a compute node. The "( )" next to compute-0-0 indicates the node has not yet requested a kickstart file.

You will see this type of output for each compute node that is successfully identified by insert-ethers. The star “(*)” means that the node has picked the kickstart file.

3.5 When node installation is complete you can press F8 to quit the screen on the frontend.

3.6 After node installation, you will get the following screen for the installed node.

4. Start Computing:

4.1. Run MPI program in cluster:

If you don’t have a user account on the cluster, create one for yourself, and propagate the information to thecompute nodes with:

# useradd username

# passwd username

# rocks sync users

• Create a file in your home directory named machines, and put two entries in it, such as:

compute-0-0

compute-0-1

• Now launch the job from the frontend:

#ssh-agent $SHELL

#ssh-add

# /opt/mpich/gnu/bin/mpirun -np 2 –machinefile machines ./hello

The hello.c MPI program:

----------------------------------------------------------------------------------------------------------------

#include <stdio.h>
#include <mpi.h>

int main(int argc, char *argv[]) {
int numprocs, rank, namelen;
char processor_name[MPI_MAX_PROCESSOR_NAME];

MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &numprocs);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Get_processor_name(processor_name, &namelen);

printf("Process %d on %s out of %d\n", rank, processor_name, numprocs);

MPI_Finalize();
}
----------------------------------------------------------------------------------------------------------------

4.2. You need to compile the hello.c program first.

Run the command:

# mpicc hello.c –o hello

Note: Now, to execute the MPI program successfully across the cluster, you need to keep following things in mind:

1. The MPI program should be run from a non-root user.

2. After creating a new user, you have to sync the user to all of your nodes. To do that, run the command:

# rocks sync users

3. The same executable program must be present on all the nodes at the non-root user’s home directory.

4. You can run the program:

# mpirun –np 4 –H compute-0-0 ./hello

# mpirun –np 4 –H compute-0-0,compute-0-1 ./hello

# mpirun –np 4 –machinefile machines ./hello