Control scheduling scheme design for real-time systems.

摘要

Recently, embedded systems have become commonly used in several manufacturing sectors, as well as mass in produced goods. The automotive industry is a good illustration for the use of embedded control systems. It is expected that embedded systems will become even more prevalent as technology advances. Multi-tasking environments are quite common for developing embedded systems. Resources in embedded systems, such as CPU and memory, are much more limited then that of the desktop environment due to cost and space. In this thesis, we propose a novel scheduling scheme for multi-tasking environments in a wide range of embedded control systems. The controlled target, in this thesis, is a real-time scheduler; thus, the proposed scheduling scheme is a feedback control-based real-time scheduler.; Due to the limited resources available to embedded systems, it is essential that the resource utilization should be improved. Although soft real-time constraints are applied to most embedded control systems rather than hard real-time constraints, it is necessary to minimize deadline constraint missing caused by the intrinsic uncertainty. In order to satisfy these two requirements, which are (1) improvement of the use of a given system's resources, and (2) reduction of the missing of deadline constraint, acquisition of accurate scheduling parameters is needed. It is for analyzing and scheduling multiply running tasks. The stochastic nature of a computing system in general hinders us from measuring the necessary scheduling parameters. As a remedy, real-time systems researchers traditionally have approximated scheduling parameters with the worst case execution time (WCET)-based parameters. Unfortunately, these parameters lead to a system that is very conservative and underutilized, as the worst cases of execution time occasionally occur.; To tackle the two real-time scheduler problems related to increasing resource utilization and increasing robustness with regards to uncertainties in embedded systems, feedback control-based real-time scheduling scheme is proposed. The advantages of adopting feedback control are two-fold: (1) it is not necessary to measure accurate scheduling parameters, as feedback control dynamically regulates the scheduling system to function within an operating range; (2) The intrinsic robustness of feedback ensures that the scheduler disregards uncertainties resulting from the stochastic nature of computing systems. In short, a feedback-controlled scheduling problem can be also defined as a resource assignment and management problem under the existence of uncertainties.; As a base scheduler for our feedback control-based scheduler, we employ the reservation-based scheduler, since it has the attractive property of isolation and comes with a mathematical model of a switched discrete system. We are also able to insert within the model a polytopic type uncertainty to model time-varying bounded uncertainty caused by task execution time variances. Hence, our target system becomes a switched discrete system with bounded time-varying uncertainty. In order to solve the control synthesis problem, we propose a linear matrix inequalities (LMI)-based Lyapunov method to tackle two characteristics of reservation-based scheduling systems: (1) the switched system characteristic, and (2) time-varying uncertainty characteristic. Feedback controllers in LMI-based form can be solved by LMI solvers. After modeling the system and synthesizing the controller, we carry out feedback-controlled scheduling experiments to illustrate the capability of the proposed scheme and to draw conclusions on the strength of the suggested approach.