Abstract
Today's hard real time systems are designed to control unpredictable environments. They are required to be dynamic, and at the same time, highly predictable, adaptive, flexible, reconfigurable, and fault-tolerant. Among these many characteristics, dynamism and predictability, two apparently antagonistic characteristics, are the most dominant ones in the design of such systems. A dynamic system can truly be predictable only when all its policies, mechanisms, and services supporting task execution in a distributed environment are themselves predictable: resources needed for computation must be allocated to contending tasks in a predictable way, decisions to dynamically schedule new tasks must be taken in a predictable time, and allocation of resources needed for inter task communication must be allocated in assessable delays.
The RTDOS project was initiated in 1989 to investigate policies and mechanisms that support the design and implementation of predictable dynamic hard real time systems. In this project, we strove as much as possible to follow a pragmatic approach, trying to always implement and experiment with new concepts before reaching final conclusions. The objective of this paper is to present the experience gathered throughout this project, emphasizing the many original approaches to the predictability problems developed for the RTDOS system. Finally, we succinctly present performance evaluation results to consolidate our targeted objectives.