【作者】 秦承刚；

【导师】 于东；

【作者基本信息】 中国科学院研究生院（沈阳计算技术研究所） ， 计算机应用技术， 2012， 博士

【摘要】 实时操作系统是开放式数控系统的核心部件，是确保数控系统高速、高精、体系结构开放以及智能化的关键技术之一。国内外的相关研究机构均开展了面向开放式数控系统的实时操作系统相关技术研究。在“高档数控机床与基础制造装备”国家科技重大专项的若干课题中，将基于开源软件的实时操作系统相关研究列为重要研究内容。但目前实时操作系统的现状是注重提升系统的平均实时性，不能完全满足开放式数控系统的需求。因此，针对开放式数控系统的共性需求以及各类开放式数控系统的特殊需求，开展面向开放式数控系统的实时操作系统优化技术的研究与应用，从功能优化与性能优化两方面增强实时操作系统的兼容性、可扩展性、自适应性以及对多核处理器的支持是需要解决的重要问题。围绕上述问题，本论文开展的研究工作如下：1.面向开放式数控系统的实时操作系统功能优化。现有的实时操作系统没有考虑到数控系统在开放性与性能评估上的特殊需求，难以支撑开放式数控系统的可扩展性与可移植性。为此，本论文对现有的开源实时操作系统进行了功能优化，提出实时操作系统接口抽象层、开放式调度框架以及实时系统性能评估模块，并将其应用于开放式数控系统的开发，以解决各类开放式数控系统对实时操作系统的共性需求。2.面向数控系统的数据流反馈调度研究与应用。开放式数控系统的核心功能由基于数据流的软件模型实现。在硬件资源紧张的标准型数控系统中，数据流有可能发生中断。数据流中断会降低数控系统的加工速度与加工精度。针对开放式数控系统的软件结构，本论文提出一种数据流反馈调度模型。通过动态监测系统中各个缓存的数据量，利用马尔可夫模型预测缓存中数据量的变化，动态调节相关任务的执行频率与优先级，以降低数据流中断的发生次数。本论文在标准型开放式数控系统中应用了该模型，并验证了该模型的有效性。3.基于勒贝格采样的自适应反馈调度研究与应用。在标准型数控系统中，实时负载的动态变化会导致人机界面等非实时任务长期得不到执行，从而降低系统的整体性能。针对此问题，本论文提出了能够均衡系统负载的自适应反馈调度模型，并利用勒贝格采样方法降低了反馈调度模型的系统开销。通过在标准型数控系统中的应用，验证了自适应反馈调度模型的有效性。4.面向双核处理器的Cache感知硬实时调度算法的研究。在使用双核处理器处理数控系统中的硬实时任务集时，并发的实时任务会通过共享Cache彼此干扰，严重制约了系统的实时性，并使得数控系统的加工质量无法得到保证。多核处理器上的Cache感知实时调度算法是目前实时调度领域的热点研究问题，这类算法能够解决并发任务彼此干扰的问题。为了提高双核处理器上硬实时任务集的实时性，本论文开展了Cache感知实时调度算法的研究工作，提出一种面向双核处理器的Cache感知硬实时调度算法，并且给出了相应的可调度性判断条件。通过相关实验验证了该算法的有效性。5.多核数控系统中实时性优化方法的研究与应用。在使用多核处理器处理开放式数控系统的混合任务集时，非实时任务会通过共享Cache干扰实时任务的实时性，软实时任务也会通过共享Cache干扰硬实时任务的实时性，从而制约了数控系统的整体性能。针对此问题，本论文提出了一种能够实现共享Cache分区的实时内存管理器，以及基于共享Cache分区的Cache感知多级调度框架。实时内存管理器为不同种类的任务划分独立的共享Cache分区，以排除非实时任务对实时任务的干扰。多级调度框架中的Cache感知软实时调度算法通过启发式方法进一步降低软实时任务集通过共享Cache的相互干扰。通过在多核数控系统中应用实时内存管理器与Cache感知多级调度框架，验证了它们的有效性。更多还原

【Abstract】 The Real-time Operating System (RTOS) is one of the core components of OpenComputerized Numerical Controller (CNC) and the key technology to ensure the CNCsystem’s characteristics of high-speed, high-accuracy, openness and intelligence.Therefore, many domestic and foreign researching organizations have been researchingthe related technologies of RTOS for the open CNC system. Several research topics ofNational Science and Technology Major Project on “High-end CNC machine tools andbasic manufacturing equipment” have proposed that RTOS based on the open sourcesoftware is one of the important research subjects. However, the existing RTOS studiespay too much attention to the average real-time capability, without thoroughlyconsidering the special requirements of the open CNC system. Thence, it is the keyresearch problem to optimize the existing RTOS subject to the demand of the majoritykinds of open CNC systems. By improving the compatibility, scalability, adaptabilityand the ability of supporting the multicore processors, RTOS will become the basis ofthe software platform of open CNC system. This dissertation focuses on the researchand application on optimization technologies for RTOS in CNC systems. The maincontributions of this dissertation are summarized as follows:1. Research and implementation on the function optimizations for RTOS. Theexisting RTOS cannot satisfy the specific requirements on openness and performanceevaluation of open CNC system. They cannot support the open CNC system’sinteroperability, scalability and portability. In order to solve these common requirements,3function extensions are presented in this dissertation. The RTOS API Abstract Layerdynamically transforms the API functions between the different API systems. Differentkinds of open CNC system will use different real-time scheduling algorithms. An openscheduling scheme is presented in this dissertation to solve the above problem. Areal-time system performance evaluation tool which could acquire the hardware levelperformance data and the OS level performance data is designed to evaluate theperformance of RTOS and open CNC system.2. Research and implementation on the data flow feedback scheduling model foropen CNC System. The core functions of the open CNC system’s software are based onthe data flow model. In the standard open CNC system whose hardware resources arenot sufficient, the data flow would be interrupted in certain specific scenarios, which degrades the velocity and accuracy of the machine process. A data flow feedbackscheduling model is presented in this dissertation to solve this problem. An experimentin standard open CNC system verified the validity of the model.3. Research and implementation on the adaptive feedback scheduling modelbased on Lebesgue sampling. In the standard open CNC system, the dynamic change ofreal-time workload might cause some non-real-time tasks is long-term lack of execution.In this case, the overall performance of the open CNC system can no longer be ensured.This dissertation presents an adaptive feedback scheduling model that can automaticallybalance the workload of ONC system. The overhead of the model itself is decreased bya Lebesgue sampling method. This model is verified by an experiment in a standardopen CNC system.4. Research and implementation on Cache-aware hard real-time schedulingalgorithm for dual-core processor. While a hard real-time task set is executed by adual-core processor, the concurrent tasks would disturb each other by competing for theshared Cache. The interference would degrade the predictability and the machinequality of the open CNC system. In order to optimize the predictability of the hardreal-time task set in the dual-core processor, a cache-aware hard real-time schedulingalgorithm and a judgment condition for the scheduability are presented in thisdissertation. An experiment verified the validity of the algorithm.5. Research and implementation on the real-time optimization method for theopen CNC system based on multi-core processor. While the open CNC system isimplemented by a single multi-core processor, the non-real-time tasks would disturb theexecution of the real-time tasks through the shared Cache. This interference restricts thereal-time performance of the open CNC system. This dissertation proposes a real-timememory manager and a Cache-aware multi-level scheduling scheme. The real-timememory manager can partition the shared Cache. On the basis of Cache partition, aCache-aware soft real-time scheduling algorithm in the multi-level scheduling schemefurther decrease Cache access conflict by a heuristic method. The real-time memorymanager and the Cache-aware multi-level scheduling scheme are verified by anexperiment in a CNC system based on a quad-core processor.更多还原