Getting Started
From UniCluster
This document provides the instructions for downloading and installing the UniCluster software into your cluster. You will perform this work on a single node in the cluster, called the Installer Node. There is one Installer Node in a cluster. It holds all provisioning and configuration data for all nodes in the cluster. Its purpose is to provision and deploy software to the other nodes (the Compute Nodes). You perform all provisioning and configuration on the Installer Node, which then "pushes" these changes out to the Compute Nodes. System administrators and users typically have direct network access to the Installer Node, which acts as a gateway between the users and the compute nodes. The Installer Node may also be referred to as the "Master" or "Main" node.
Installer Node Requirements
- Installed Linux Operating System - choose one of these supported operating systems:
- RedHat Enterprise Linux 5, 32 or 64 bit.
- CentOS 5, 32 or 64 bit.
- Oracle Enterprise Linux, 32 or 64 bit.
- 2 gigabytes RAM
- 30 gigabytes free disk space
- 1 network interface (2 recommended)
- The yum-utils package
- The Development Tools package group
- 32 bit Operating Systems work, but are not fully supported or tested
Recommendations
UniCluster can be installed onto a machine with a single network interface, however we recommend configuring the Installation Node with two network interfaces:
- an interface for connecting the installer node to the public network for admin/user/desktop access, and
- an interface to reach the compute nodes on a dedicated network. Since a DHCP server will be running on this interface, it must have a static IP address.
UniCluster Installation Instructions
- Install RHEL, CentOS or Oracle Enterprise Linux onto a machine. If you already have a spare machine with the Operating System installed you can use that machine.
- Create the provision network for your cluster. You will need to know the gateway IP address, the broadcast, and the netmask for this network.
- Set up the network interface on the installation node from which compute nodes will boot. Use a static IP address; this will become the address from which the compute nodes will boot. The easiest way to modify the network is using the setup program.
- Download http://peaches.grid.org/unicluster/unicluster-repo-4.1-0.noarch.rpm
- Install the RPM:
rpm --install unicluster-repo-4.1-0.noarch.rpm
- Run the uninetwork-register script to get access to the UniCluster software. A valid grid.org username and password will be needed:
/opt/tortuga/sbin/uninetwork-register
- Begin installing UniCluster:
yum install unicluster
- Set shell environment (or just log out and back in):
source /etc/profile.d/unicluster.sh
- Complete installing UniCluster:
/opt/tortuga/sbin/unicluster-setup
The unicluster-setup command begins by checking the network interfaces on the Installer Node to identify the one(s) to use for provisioning the Compute Nodes. Part of the provisioning process invloves running a DHCP server on the Installer Node to respond to requests from the Compute Nodes. Unicluster-setup needs to know which interface(s) to run the server on. Unicluster-setup warns you if it appears that you risk running DHCP on a public network, and gives you a chance to stop; you probably do not want the Installer Node to run a DHCP server on any network(s) other than those dedicated to the Compute Nodes.
If unicluster-setup finds more than one network interface configured with static IP addresses, it will ask you which one(s) you want to use for provisioning.
Once the unicluster-setup command completes, you are ready to begin adding and provisioning Compute Nodes.
Note that unicluster creates some files and directories in certain system locations, e.g. /var/log, /var/run, /etc/init.d, /var/www/html, etc.