【作者】 高志刚；

【导师】 吴朝晖；

【作者基本信息】 浙江大学 ， 计算机科学与技术， 2008， 博士

【摘要】 随着信息技术的快速发展,嵌入式系统已经渗透到生产和生活的方方面面。汽车电子是嵌入式系统应用的一个典型领域。汽车电子技术是当今汽车技术发展的主要动力。汽车电子控制系统是汽车电子系统的核心,对于提高车辆的性能、安全性和可靠性具有重要的作用。汽车电子控制软件具有严格的强实时要求并受到系统的资源限制和能量限制。随着汽车电子控制系统的日益复杂化以及行业内部竞争的加剧,目前的软件开发方法越来越难以满足汽车电子控制软件高复杂度、高安全性、低成本、短开发周期等要求。基于模型的软件开发可以有效地控制软件的复杂度,提高软件的开发效率,降低软件的开发和维护成本,提高软件的正确性,从而成为目前嵌入式软件开发的重要研究领域之一。模型综合是基于模型嵌入式软件开发的重要阶段,该阶段对于满足嵌入式软件的非功能约束,生成正确的嵌入式软件具有重要的意义。然而,目前的模型综合方法存在以下的不足:1)不支持汽车电子控制软件中复杂的事务模型;2)不支持OSEK标准所定义的混合调度方式和分组调度方式;3)不支持复杂事务模型在混合调度方式和分组调度方式下的定时验证;4)未考虑节能问题,从而难以满足基于模型的汽车电子控制软件开发中模型综合的要求。本文以汽车电子控制系统作为实现模型的硬件平台,以符合汽车电子行业公认的OSEK标准的操作系统作为实现模型的软件平台,研究在基于模型的汽车电子软件开发中使用复杂事务模型时的模型综合问题。重点研究了单处理器环境中时间优化的模型综合方法、分布式环境中多资源约束的模型综合方法和能量节省的模型综合方法,为汽车电子软件开发中模型综合的自动化提供了理论和方法支持。本文的主要工作包括以下四个方面:1)提出了汽车电子软件模型综合的理论框架。本文介绍了基于模型汽车电子软件开发环境SmartOSEK IDE V3.5(目前处于研究和开发中)中模型综合的理论框架——MOSAES以及与MOSAES相关的模型:结构模型、平台模型、实现模型和约束模型等,为汽车电子软件的模型综合提供了明晰的上下文。2)研究了单处理器环境中的模型综合问题,提出了以优化临界缩放因子为目标的模型综合方法。本文扩展了HKL算法,提出了满足1-M和M-1共享关系时混合调度方式下事务的定时验证方法,以及满足M-1共享关系时分组调度方式下事务的定时验证方法。提出了单处理器环境中混合调度和分组调度方式下模型综合的流程。在模型综合时,使用定时验证保证了生成的实现模型满足定时约束;使用模拟退火算法以探索优先级的分配,优化临界缩放因子;并根据任务之间的干涉关系合并任务,在保持定时验证结果的情况下减小系统的开销。3)研究了分布式环境中的模型综合问题,提出了满足多资源约束的模型综合方法。该方法包括构件分配和实现模型生成两个阶段。首先,在满足处理器的计算资源、内存资源约束和保持计算密度平衡的条件下把构件分配到处理器上,然后以满足定时约束为目标生成实现模型。构件分配和实现模型生成是一个回溯的过程。此外,还提出了混合调度方式和分组调度方式下,任务段之间不连续时,事务的定时验证方法,以满足分布式环境中验证定时约束的需要。4)研究了模型综合中的节能问题,提出了能量节省的模型综合方法。本文首先提出了固定优先级线性事务模型中实现节能的HDVS算法。然后在HDVS算法节能原理的基础上,分析了结构模型的特点和调度方式对于事务类型划分和任务执行的影响,扩展了单处理器中的模型综合方法,从而集成节能特性到单处理器环境的模型综合过程中。更多还原

【Abstract】 With the rapid development of information technology, embedded systems have permeated into every aspect of industries and lives. Automotive electronics is a typical field in which embedded systems are applied, and it has become the main motivation for the development of automotive techniques. As the core of automotive electronic systems, the automotive electronic control system plays an important role in improving the performance, safety, and reliability.Software for automotive electronic control system has stringent hard real-time requirements, and is constrained by system resources and energy. With the increasing complication of automotive electronic control systems and the pressure of shortening time to market, the current software development methodology become more and more difficult in fulfilling the automotive electronic control software’s requirements for high complexity, high safety, low cost, short development periods. Model-based software development can efficiently control complexity, improve productivity, reduce the cost of developing and maintaining, and improve correctness, which make it become one of the important research fields in current embedded software development. Model synthesis, as a critical phase in model-based embedded software development, has important influence on fulfilling the non-functional constraints of embedded software and generating the correct embedded software. However, the existing model synthesis methodologies has the following problems: 1) they do not support complex transaction models; 2) they do not support the hybrid scheduling mode and the group-based scheduling mode; 3) they do not support the timing verification of complex transaction models under the hybrid scheduling mode and the group-based scheduling mode; 4) they do not consider the energy-saving problem. Therefore, it is difficult for them to fulfill the requirements of model synthesis in the model-based development for automotive electronic control software.This thesis researches the model synthesis problem when using automotive electronic control systems as the hardware platforms of implementation models, operating systems compatible with OSEK standard, a wildly accepted standard in automotive electronic field as software platforms, complex transaction model as task models. This thesis focuses on the model synthesis problems in uniprocessor environment, distributed environment and energy-efficient model synthesis. This research provides supports for automatic model synthesis in the development of automotive electronics software.The main contents and contributions of this thesis are as follows.First, it presents the MOSAES, a theoretic framework of model synthesis for the development of model-based automotive electronics software. This thesis describes the relationship between MOSAES and SmartOSEK IDE V3.5 (which is ongoing), and defines the models used in MOSAES, such as the structural model, the platform model, the implementation model, and the constraint model, which provide a clear context for model synthesis.Second, it presents the model synthesis methodology for uniprocessor environment with the goal of optimizing Critical Scaling Factors. This thesis extends the HKL algorithm and presents the method of timing verification when using 1-M and M-1 sharing under the hybrid scheduling mode, and the method of timing verification when using M-1 sharing under the group-based scheduling mode. It also presents the workflow of model synthesis under the hybrid scheduling mode and the group-based scheduling mode for uniprocessor environment. During model synthesis, we use timing analysis to verify the timing constraints, use the Simulated Annealing algorithm to explore priority assignment and optimizing Critical Scaling Factors, merge tasks according to the interference relations among tasks in order to reduce system overheads while keeping the results of timing verification.Third, it presents the model synthesis methodology for distributed environment with the goal of meeting multiple resource constraints. This methodology includes two steps: component assignment and implementation model generation. It first assigns components into processors under the conditions that satisfy computation resource and memory resource constraints, and keep the balance of computation density among processors. And then it generates implementation models while satisfying timing constraints. It is a backtracking process between component assignment and implementation model generation. Morever, this thesis presents the timing verification method for non-consecutive task segments under the hybrid scheduling mode and the group-based scheduling mode to meet the requirements of timing verification in distributed environment.Last, it presents the energy-efficient model synthesis methodology. It first presents the HDVS algorithm under the Linear Transactions with Fixed Priorities model. And then, on the basis of energy-saving theory, this thesis analyzes the characteristics of structural models and the influence of scheduling modes on the classification of transaction types and tasks execution. This thesis extends the model synthesis methodology for uniprocessor environment, and integrates energy efficiency into MOSAES for uniprocessor environment.This thesis was supported in part by the 863 National High Technology Program of China.更多还原