Developing Mixed-Criticality Real-Time Systems: Analysis Methods and Tools

实时嵌入式计算系统监视并通常控制与物理世界交互的工件或其他系统。此类系统的示例包括航空电子系统、汽车系统和医疗机器人。这种实时嵌入式系统是安全关键的;它们的故障可能会对人或环境造成伤害。 我提出的计划的首要主题是避免或限制实时嵌入式计算系统的故障。失败可能是由于： * 设计和实施失败（硬件和软件层）; * 操作硬件/组件故障; * 当系统不能在时间界限或期限内响应时，表现为时间错误的资源分配问题。 我特别打算推进调度和时序相关问题以及硬件组件故障的检测、诊断和恢复方面的最新技术。我还将建立在近似计算的基础上，在满足时间约束更关键的时候，对输出质量进行权衡。 我的努力考虑到一些不同的压力，影响实时嵌入式系统的发展，如需要减少尺寸，重量和功耗（有时称为SWaP约束）的计算平台。这些压力导致不同关键性的计算活动在同一硬件平台上的共置（例如，在无人机上，拍摄照片的任务不如负责飞机稳定性的任务重要）。这种设计导致混合临界系统。然而，安全关键系统需要满足诸如IEEE 26262的安全标准中强调的要求。这些标准使安全认证成为可能，并指出了不同关键级别任务的可接受故障率。 近年来，混合临界实时调度领域的许多工作都集中在双临界系统上。任务的关键性可以分为高关键性和低关键性，大多数现有工作都将低关键性任务视为尽力而为的任务。换句话说，在之前的工作中，研究人员已经确保高关键性任务符合他们的最后期限，即使不必要，低关键性任务也会被放弃。在以前的工作中采取的方法也不符合安全标准，这些标准指示每个关键性级别的故障概率或故障率（因此，高关键性任务可能会失败，但与低关键性任务的失败率不同）。 通过这笔资金，我打算开发易于处理的概率分析方法，以及实时混合临界系统的运行时支持。在容错方面，已经开展了一些工作，审查混合临界系统，但在平台和分析方法方面还需要开展更多工作，以确保遵守安全标准和认证程序。

Real-time embedded computing systems monitor and, often, control artifacts or other systems that interact with the physical world. Examples of such systems include avionics systems, automotive systems and medical robotics. Such real-time embedded systems are safety critical; their failure may result in harm to people or their environment. The overarching theme of my proposed program is to avoid, or limit, failures of real-time embedded computing systems. Failures may be a result of: * design and implementation failures (at both the hardware and software layers); * operational hardware/component failures; * resource allocation problems that manifest as timing errors when the system is not able to respond within timing bounds or deadlines. I specifically intend to advance the state of the art in scheduling and timing-related problems as well as with detection, diagnosis and recovery from hardware component failures. I will also build upon work in approximate computing to make output quality tradeoffs when it is more critical to meet timing constraints. My effort takes into consideration some of the different pressures that influence the development of real-time embedded systems such as the need for reduced size, weight and power consumption (sometimes called SWaP constraints) of the computing platforms. These pressures result in the colocation of computing activities of different criticalities on the same hardware platform (e.g., on a UAV, a task that takes photographs is less critical than the task that is responsible for aircraft stability). Such designs result in mixed-criticality systems. Nevertheless, safety critical systems need to meet the requirements highlighted in safety standards such as IEEE 26262. These standards enable safety certification, and they indicate the acceptable failure rates for tasks at different criticality levels. Much of the recent work in the area of mixed-criticality real-time scheduling has focused on dual criticality systems. Tasks are either high or low criticality, and most existing work treats low criticality tasks as best-effort tasks. In other words, in prior work, researchers have ensured that high-criticality tasks meet their deadlines and low-criticality tasks are dropped even when this is not necessary. The approach taken in prior work does not also match with safety standards which indicate a failure probability or failure rate for each criticality level (so it is possible for a high-criticality task to fail, but not at the same rate as a low-criticality task). Via this grant, I intend to develop tractable probabilistic analysis methods as well runtime support for real-time mixed-criticality systems. In the area of fault tolerance, there has been some work that examines mixed criticality systems, but more work is needed in terms of platforms and analysis methods to ensure compliance with safety standards and certification processes.