ASTEROID - An Analyzable, Resilient, Embedded Real-Time Operating System Design

操作系统（OS）在任何复杂的计算系统中都起着关键作用。当前支持软件集成内存管理和资源虚拟化的操作系统包含几个依赖于无错误硬件（HW）的核心功能。这些函数中未被检测到的错误迅速且不可逆地传播到整个系统，使得从函数故障中恢复几乎是不可能的。其他操作系统功能可以通过适当的机制从故障中恢复。这样的函数继承了使用它的应用程序的可靠性需求。该项目的思想是开发操作系统和硬件机制，利用多核系统的硬件和通信资源，有效地提供所需的可靠性。在项目的前两个阶段，我们确定了操作系统和处理器管道的关键核心功能。我们开发了故障检测和纠正机制，利用了最先进的和未来的多核和多核架构的固有冗余。因此，我们的系统能够支持使用冗余多线程的现代多线程应用程序，并为这些应用程序提供实时保证。我们将研究扩展到多处理器合作中产生的错误，例如通过片上网络（NoC）中的故障。我们确定了可靠计算基础，软件和硬件组件需要正确运行，以提供实时和可靠性保证，并调查了这些组件的故障如何影响系统。第三个项目阶段的目标是将前几个阶段开发的机制和方法集成到一个系统中，该系统可以实现对瞬态硬件故障的完全跨层保护。为此，我们将集成现有的结果，优化它们的覆盖范围、资源和性能开销，并研究在前一阶段未解决的剩余差距，包括NoC的端到端保护和设备驱动程序的os辅助复制。

The operating system (OS) plays a key role in any complex computing system. A current OS supporting software integration with memory management and resource virtualization contains several core functions that depend on error free hardware (HW). Undetected errors in these functions quickly and irreversibly propagate through the system making it virtually impossible to recover from a function failure. Other OS functions can recover from failures with appropriate mechanisms. Such functions inherit the dependability requirements of the applications using it. The project idea is to develop OS and HW mechanisms that utilize the HW and communication resources of a many-core system to efficiently provide the required dependability. In the first two project phases we identified critical core functionality of the operating system and the processor pipeline. We developed fault detection and correction mechanisms that exploit the inherent redundancy of state-of-the-art and future multi- and many-core architectures. As a result, our system is able to support modern multithreaded applications using redundant multithreading and provide real-time guarantees to these applications. We extended our research into errors arising in the cooperation of multiple processors, for instance through failures in the network-on-chip (NoC). We identified the Reliable Computing Base, software and hardware components that need to function correctly in order to provide real-time and reliability guarantees and investigated how failures in these components affect the system. The goal of the third project phase is to integrate the mechanisms and methods developed in the previous phases into a system that achieves full cross-layer protection against transient hardware faults. For this purpose we will integrate existing results, optimize their coverage, resource, and performance overheads and research the remaining gaps that were not addressed in the previous phases, including end-to-end protection of the NoC and OS-assisted replication for device drivers.