The reason for research in scheduling is an economical one, viz., to optimise the utilisation of resources. Up to the present time, almost all interest is directed towards processor scheduling. Departing from the requirements holding for real-time computing, in this paper it is shown that maximum processor utilisation has become obsolete as an optimisation criterion for industrial real-time systems. It is also shown that the earliest-deadline-first discipline and certain modifications thereof provide a satisfactory and final answer to all real-life scheduling needs. To this end, all intrinsic properties of this discipline are compiled and discussed in order to show that it is the most advantageous scheme at hand, characterised by efficiency and allowing for predictable system behaviour. It is then pointed out how the method naturally extends to the scheduling of tasks having non-pre-emptable regions due to resource-access constraints. A sufficient condition is presented, which allows, at any arbitrary point in time and under observation of resource constraints, to check the feasible schedulability of the tasks competing for processor allocation. This condition applies to entirely non-pre-emptable tasks as well. Then, by taking industrial practice and actual cost relations into account, evaluation criteria and design principles for real-time computing systems are developed. The paper closes with pointing to those open optimisation questions, scheduling research ought to address if it wants to deal with practically relevant problems, viz., minimisation of software costs, software complexity, and complexity of schedules, synchronisation sequences, inter-task communication, etc. In other words, simplicity is to be maximised to enhance system dependability and predictability of system behaviour.