Institutional Repository
| 多处理器反馈控制实时调度研究 | |
| Alternative Title | Feedback Control Scheduling of Real-time |
| 傅勇 | |
| 2008-01-19 | |
| Degree Grantor | 中国科学院软件研究所 |
| Degree Level | 博士 |
| Place of Degree Grantor | 软件研究所 |
| Keyword | 实时系统 分布式系统 操作系统 实时调度 |
| English Abstract | 多处理器实时系统是一种为了满足大规模实时应用而设计的系统。由于其广泛的应用，近年多处理器实时系统来得到了广泛的研究。实时调度问题是实时系统的核心理论，但传统的实时调度理论着重于解决任务特性明确的应用，如航空电子系统，雷达系统。而多处理器实时调度系统应用范围宽广，各类应用的任务负载情况各不相同，并且有些应用的任务特性是无法预知的。在这种情况下，传统的实时调度理论作用大大降低。反馈控制实时调度是一种新出现的实时调度理论。它结合了传统调度理论和经典的反馈控制理论，产生了一种创新的实时调度体系。反馈控制实时调度理论侧重于解决在系统任务特性未知的情况下，如何高效的满足实时和性能的要求，并具有广泛的应用前景。反馈控制实时调度具有理论基础坚实、系统分析能力较强及利于实际应用等特点。本文深入研究了反馈控制实时调度的理论、方法以及实现，在反馈控制多处理器实时调度理论和实现中取得了一定的创新性成果。本文提出了一种新的反馈控制实时调度算法。其结合了实时调度算法和分布式反馈控制算法，可以在满足同构和异构的多处理器实时系统的性能要求下同时保证系统的实时特性。同时本文还提出了温度感知的反馈控制实时调度算法，在系统工作负载变化是使系统的温度保持一定水平。同时本文建立了相应的仿真平台，对所提出的方法而且进行了大量的仿真实验。通过仿真实验的分析表明，这些算法满足各自的性能指标要求。 |
| Abstract | Modern multiprocessor real-time systems are designed to support large scale real-time applications. For the important role that it plays in our information technology, it is not surprised that research interests of real-time multiprocessor systems gradually increases in recent years. Real-time scheduling theory, the corner of the real-time systems is critical for real-time systems in which the correctness of system output does not only depend on the results but also the time generating the results. Traditional real-time scheduling theory can only analyze and ensure the temporal properties of a small class of applications, for examples, aviation electronics, radar system and so on. Nevertheless, it is not validated when employing in modern multiprocessor or distributed real-time systems in which the workloads are often fluctuated and unpredictable. Feedback Control Real-time Scheduling(FCRS) is a promising method which concentrates to deal with dynamic and unpredictable workloads. FCRS combines traditional real-time scheduling theory and classic feedback control theory to generate a novel paradigm for real-time scheduling. FCRS can ensure the system performance even the fluctuate workload. And a lot of applications can be benefitted from it. FCRS is developed on a solid theory by combining both real-time scheduling and control theory. Moreover, FCRS can predict the whole systems performance, especially dynamic performance, even before it is completed. Another important property of FCRS is that it can produce efficient algorithm for the controller is essentially a numerical analysis algorithm for which it is easy to find the very efficient implementation. In this thesis, several new FCRS method on multiprocessor systems are proposed. First, we develop Distributed Utilization Control with Load Balancing(DUCLB) for homogeneous multiprocessor real-time systems with load balancing. Second, based on DUCLB, we propose HDUCLB(Heterogenous Distributed Utilization Control with Load Balancing) which can apply on heterogenous multiprocessor systems in which the capability and connection pattern is different from one processor to another. We also develop a temperature aware feedback control real-time scheduling algorithm which can maintain the temperature of real-time multiprocessor systems. Finally we develop a simulation framework for FCRS on real-time multiprocessor systems. Based on this simulation framework, we perform the simulations of the feedback control real-time scheduling methods proposed in this thesis and validate the effectiveness of them. |
| Pages | 147 |
| Language | 中文 |
| Content Type | 学位论文 |
| URI | http://ir.iscas.ac.cn/handle/311060/6278 |
| Collection | 中科院软件所_中科院软件所 |
| Recommended Citation GB/T 7714 | 傅勇. 多处理器反馈控制实时调度研究[D]. 软件研究所. 中国科学院软件研究所,2008. |
| Files in This Item: | ||||||
| File Name/Size | DocType | Version | Access | License | ||
| 10001_20041801502904（1983KB） | 限制开放 | -- | Application Full Text | |||
Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.
Edit Comment