Real-time scheduling - Mono-processor scheduling

Computer systems can be divided into three categories: transformational, reactive and interactive. Transformational systems transform input data into output data at their own pace, without interacting with their environment. Interactive systems typically represent the class of programs with a graphical interface. They respond to a user's requests, and it is welcome that they respond quickly. Reactive systems are in charge of controlling processes. A process is a physical system to be controlled, with its own environment. Consequently, the reactive system has to react quickly enough to the dynamics of the controlled process and its environment. In contrast to the other categories, reactive systems do not exhibit recurring behaviors: as the process evolves with its own dynamics in an often complex environment, it is rare for reactive system/process evolution in its environment execution scenarios to be reproducible.

IT systems controlling critical physical processes, particularly in the fields of transportation, autonomous exploration, etc., are typically embedded in the process itself. These systems are referred to as critical embedded systems. Such systems are often subject to time constraints inherent in the dynamics of the controlled process. They are then said to be real-time. A real-time system is a reactive system, in charge of ensuring that a process (drone, automotive or avionics system, etc.) is maintained in a desired state, while ensuring system reactivity consistent with the time constraints inherent in the process and its environment. Despite the increase in computer power, first in frequency, then in the number of computing cores since the early 2000s, on-board computers are dimensioned as closely as possible to computing power requirements, in order to minimize the energy consumed, and therefore the autonomy of the process.

Two assumptions are commonly made when partitioning models, methods and issues: the synchronous assumption and the asynchronous assumption. The synchronous hypothesis considers that processing is triggered directly by the arrival of events. Treatments are considered to have a duration of zero, or at any rate less than the minimum distance separating two successive triggering events. The asynchronous assumption is based on the multitasking programming paradigm: as in the synchronous case, events can trigger processes. However, processing is considered to be of non-zero duration. As a result, one process can be pre-empted by another, i.e. interrupted, and put on hold, to allocate computing resources to another higher-priority process, for example.

In this article, we consider the asynchronous assumption. We consider this assumption to be realistic, particularly in the case of embedded systems. In this case,...