Fault Recovery in a Real-Time Switched Ethernet Architecture

As distributed multimedia applications enter the mainstream, Quality-of-Service (QoS) becomes the central focus of state-of-the-art network devices. The QoS of a network connection is determined on an end-to-end basis by all the network devices through which it traverses. EtheReal is a real-time switched Ethernet architecture that aims to provide QoS guarantees over local area networks, thus offering the "last-mile" QoS link to the end hosts when coupled with QoS-capable IP routers. EtheReal's QoS support is innovative because it does not require any modifications to end hosts' operating systems and network interface cards. Because of its connection-oriented architecture, EtheReal switches maintain per-connection state, which complicates the fault recovery process when link/switch failures occur. This paper describes the fault handling techniques employed in EtheReal, specifically a propagation-order spanning tree algorithm and multiple-spanning-tree routing, to achieve fast and transparent network fault recovery. Based on measurements from a fully operational EtheReal switch network testbed, we show that single link failures can be recovered, at the EtheReal connection level, within 31 msec, which equates to dropping one frame in a 30-frame-per-second video connection.