A software-based approach to scalable fault-tolerant real-time ethernet for networked control systems

Network fault-tolerance and predictable timing behavior have been considered very stringent requirements for mission-critical network systems such as distributed process control applications. Various fault-tolerant and deterministic control networks have been developed, mostly with proprietary solutions that employ proprietary hardware and protocols targeted at specific applications. Today, the trend in the mission-critical control system industry is to replace proprietary networks with commercial-off-the-shelf (COTS) or open networks so as to reduce product development cycle time and cost as well as to achieve system interoperability. Due to its ubiquity, simplicity, and low cost, Ethernet becomes a de facto choice for developing open mission-critical network strategies. However, Ethernet was not originally designed to handle network faults. Furthermore, as the traditional Ethernet uses a shared communication network protocol (CSMA/CD), packet transmission delay is unpredictable. To address these issues, various research and development efforts are under way to add fault-tolerance and real-time capabilities to Ethernet-based mission-critical networks.; In this thesis, we present the design, implementation, and evaluation of end-host-based fault-tolerant and real-time software mechanisms. They provide four innovative contributions over Ethernet-based distributed control network systems. First, the fault-tolerant software enables utilization of purely COTS Ethernet devices with supporting application transparency. Second, a significant failure-switchover time improvement is made over the standard spanning tree algorithm (IEEE 802.1D) to meet the requirement of distributed control applications. Third, scalable fault-tolerant architectures are proposed to remove single-point-of-failure over multi-domain network systems using COTS redundancy protocols. Finally, a probabilistic admission control algorithm over switched Ethernet is proposed to support soft real-time control applications with heterogeneous periodic flows. It enables real-time application-to-application QoS management over switched Ethernet without sophisticated packet scheduling or resource reservation mechanisms in Ethernet switches.; We made a prototype fault-tolerant mechanism in the kernel of Windows NT operating system as an intermediate driver. The experimentation results show that the fault-tolerant software is stable and provides fast switchover time (less than 2 seconds) and the impact of the driver overhead is minor. We also implemented the probabilistic admission control algorithm on the application layer of the Windows 2000 operating system, and validated its efficiency through extensive experiments.