1 . Backup real-time channels is a technique to provide uninterrupted service in the presence of router failures during real-time transmissions. This technique adds the notion of availability to the concept of QoS (Quality of Service), usually expressed in terms of guaranteed throughput and maximum delays only. Availability comes at the cost of increasing the required resource (bandwidth, buffers) reservations, due to the needs of a backup channel. However, this extra resources reservation is potentially wasted, since fault rates are very low. This paper proposes a systematic method for estimating and optimizing resource reservations. This approach is based on inaccurate failure detection in order to reduce latency is proposed. The cost of inaccurate failure detection is that the backup channel will be activated and utilized unnecessarily upon detection of "false failures". However, the paper shows, through simulations and using MPEG transmission traces, that the percentage of false failures is almost negligible.The availability requirement can be dealt with through the use of redundant channels. Redundancy can be managed using two extreme approaches. In the Multiple Copy Approach [1] all redundant channels are simultaneously active, while in the Backup Channels Approach [2] a backup is only activated upon failure detection. The backup approach is, in principle, more efficient, since it does not waste bandwidth in the absence of failures, but this paper shows that its performance strongly relies on the failure detection latency and channel setup time. Backup channels have the drawback of keeping some bandwidth reserved without being used. This resource waste can be improved by using the technique by Shin [3] based on reserving spare resources. However, two important issues concerning resource reservation remain unsolved: how to estimate the bandwidth reservation for the primary and backup channels and how does failure latency affects these estimations.The resource reservation for a given workload on a real-time channel can be estimated from its real-time requirements by properly defining that workload under the flow model of a particular resource reservation protocol. In this sense, it is worth noting that every protocol defines its own different flow model. This way, the IETF standard proposal for resource reservation, known as RSVP (Reservation Protocol) [4,5], defines a flow for a guaranteed service [6] based on the token bucket scheme while other proposals, like the Tenet Suite [7] or ATM [8] use different flow models.On the other hand, the real-time requirements for a particular workload under the backup channels scheme is determined by the failure detection scheme. This is due to the following observation: the longer is the failure latency time, the more demanding the delay requirements of the backup channel are, in order to maintain the desired QoS. According to this, the paper introduces a scheme for fault detection whose goal is to reduce failure latency (and, thus, resource reservations). The proposed t...
