Building a Linux-Based High-Performance Compute Cluster
As you enter data on these screens, the installation routine is building a small MySQL database that details all of the component configurations in your cluster. The various tables Linux needs to run (like /etc/hosts) will be generated as an SQL report from this database. If you want to make changes in the system's configuration, the tools that Rocks provides actually change the database first, then run the appropriate reports to regenerate the system configuration files. This significantly reduces the chance for errors to creep into these files. It still is possible to edit the automatically generated system files manually, but remember that the next time you use the Rocks tools to reconfigure the cluster, your manual changes will be overwritten by the automatically generated SQL report versions.
The next screen (Figure 6) allows you to enter information about your cluster. If the cluster will be connected to your enterprise network, you should enter a fully qualified hostname to be consistent with your domain. The cluster name you enter in the Cluster Name field will appear in the management screens during cluster operation. Once you are satisfied with your entries, click Next to go to the configuration of the head node network connection to the private network (eth0).
The next screen (Figure 7) lets you configure the cluster's network. The installation routine automatically selects 10.1.1.1 as the IP address for eth0 on the head node. Because this is a private network, you probably won't need to change this setting. If your public network also happens to be in the 10.1.X.X configuration, change this to something that doesn't conflict with your existing network. Clicking Next brings up the head node public network connection configuration screen.
Figure 8 shows configuring the “public” connection of the head node, its connection to the rest of your systems. The public connection for the head node must be configured with a fixed IP address. The public network for this example is configured as 192.168.0.X with a netmask of 255.255.255.0. Make sure the head node does not conflict with other servers and workstations on the public network. On the following screen (Figure 9), configure the local Gateway and DNS Server IP addresses for the head node to use.
On the next two screens (Figures 10 and 11), enter the root password and configure the time zone and NTP server for the head node.
A Note about Time
All of the systems in the cluster must be synchronized as closely as possible to each other. This is accomplished using the Network Time Protocol (NTP). The head node synchronizes itself with one of the members of the public NTP pool then acts as a local time server through the private network connection to each member of the cluster. If a member of the cluster is slightly slow or fast, the NTP dæmon on that machine will “slew” its clock over a period of time to bring it in line with the rest of the cluster.