Automatic Configuration

Most obviously, the effort required to configure several hundred machines manually from a graphical interface is not normally acceptable. However, most sites will also want to have a reasonable confidence in the correctness of their configurations; misconfigured systems represent serious security problems, as well as leading to unpredictable failures. Manually configured systems, with no explicit representation of the configuration, are notoriously difficult to guarantee correct. Early attempts to overcome these problems were often based on a cloning procedure where a single machine is configured by hand and the resulting disk image is copied directly onto a set of other machines. This is usually followed by execution of some scripts to apply any machine-specific differences. This process is useful for large numbers of very similar machines which do not change regularly, such as those in a student laboratory. It is also widely used in Windows environments.

In many installations, such as our own, both the variety of different configurations, and the rate at which they change, makes cloning impractical. We support a range of machines from file servers, to student laboratory clients, to researcher's laptops, and the hardware and software requirements are all very different. New machines arrive continuously, old machines are re-allocated, and systems are rebuilt after hardware failures or OS upgrades; all of these imply a reconfiguration, and we estimate that, on average, about 10% of our machines are completely reconfigured each week. Small configuration changes also occur very frequently in a complex environment; for example, changing a server or gateway can imply configuration changes for many other hosts. Software updates also occur at an average rate of several tens of packages per day.