In order to elucidate the goals of computer configuration and maintenance, it will be necessary to identify the main characteristics of computer systems at a suitable level of abstraction. These include finding:
The basic model of a computer is that of a dynamical community of processes and resources, coupled to an external environment (an external source or force). The source includes the stochastic influences of all of the users of the system, and any other computer systems which communicate with hosts within the perimeter of our own system. As pointed out in ref. [5], the issue of networking does not increase the complexity of the administration problem, only its localization and perhaps its magnitude. A set of networked hosts, sending external messages, is simply a single virtual host with internal inter-process communication.
The variables, important in characterizing the usage of a computer system, are measures of average behaviour, such as rate of work, numbers of processes, network connections and so forth[2]. Other measures, such as average service latencies, affect the system only at the level of the network. Latencies are very complex phenomena and are unlikely to be predictable by any simple model.
Invariances refer to the independence of qualities and values to changes. In the long run, there are no features of a computer system which are fully constant, but for long periods of time, certain things can be considered invariant. For instance, the software tools one uses to edit a file usually make no difference to the outcome, thus the outcome of an editing operation may be considered invariant with respect to differences in software used; the CPU efficiency of the software used makes no difference to the result in most cases. Invariance could also mean that a particular piece of software never changes (is never upgraded), or that the content of a configuration file is fixed with respect to other changes. In the space of changes, such invariances may be considered to be ignorable coordinates.
Persistent structures are, like invariances, values or qualities which do not change over appreciable periods of time. This includes checksums of important software, kernel profiles of software; it might also include numbers of user accounts. Persistent structures are not expected to change. Changes in these structures might be considered anomalous behaviour.
An important characteristic of computer systems is that they are strongly coupled to human users' behaviour patterns. The majority of human users follow strict daily and weekly work patterns and this is reflected in many measurements of system resource behaviour. A consequence of this is that measurements which are periodically constrained are distributed according to a Planck spectrum. The Planck spectrum can therefore be considered a general characteristic of computer statistics in many cases.