Installation Bootstrapping

Providing the ability to install new machines with the absolute minimum of manual intervention is very important. This allows failed machines to be replaced, and new machines to be installed, quickly and correctly, by unskilled staff ¹.

Installing an operating system on to bare hardware requires some sort of bootstrap process. Typically:

• A minimal version of the operating system or other install program is loaded from the network, or removable media.
• Once booted, this program partitions the system disk and installs a copy of the operating system.
• There will usually be some additional software installation and configuration, the first time that the machine reboots from the newly installed system.

The first operation tends to be operating system-dependent and the original LCFG made use of Solaris Jumpstart [15,14] to boot an initial image from the network. The current Linux port requires a boot floppy (or CD) for the same purpose, since not all hardware supports network booting (Kickstart [12] is not used).

Desktop machines may not have CD drives, and they use BOOTP/DHCP to mount the root filesystem from NFS. Laptops cannot use NFS so easily in this way because of the need for PCMCIA drivers, and they are installed using a bootable CD which contains an equivalent image. Once the system is booted, access to the network is necessary to retrieve the configuration parameters from the NIS and the RPMs for building the system disk.

When the minimal system has booted, an update script runs automatically to partition the system disk according to the LCFG resources and load the software. In the original Solaris implementation, a small hand-crafted image was first copied directly from the network onto the system disk, but this is difficult to maintain. Under Linux, the update script builds the root filesystem completely from a set of RPMs. Although it is aware of the context, this uses exactly the same process which is used nightly to update the software on a running machine (see 4.2.1), ensuring a consistent interpretation of the LCFG resources at install and update time.

When the software has been installed, the system reboots from the new image. The LCFG subsystem scripts start normally and perform the remaining configuration.

Once the install operation has been started, it runs completely unattended, allowing whole laboratories of machines to be installed easily by one person. However, the server load imposed by large numbers of clients performing simultaneous installations can be a performance problem. A ``helper'' CD is sometimes used to provide a local copy of many of the packages. If the CD becomes out of date, and newer versions of some packages are available on the network, the newer versions will automatically be installed instead. It would be interesting to look at ways of improving simultaneous large-scale installations, perhaps by using multicast to distribute the RPMs.

Footnotes

... staff¹

This does not include the restoration of any user data from backup