Certain classes of network applications (e.g. scientific visualization, real-time digital audio and video) require stringent guaranteed performance where there is a priori agreement to the quality of service expected from the network and protection enforced between unrelated connections. Supporting guaranteed service over packet and cell-switched networks requires a connection-oriented approach with admission control and resource reservation. Some researchers argue that current approaches to supplying performance guarantees in the WAN multicast environment cannot grow to support large-scale communication and many sessions. This research will characterize the scaling limitations hard state approaches of the guaranteed performance service model and evaluate several designs alternatives proposed to make the service more scalable. The findings can be used to address the question of whether this service is only suitable for private enterprise network deployment, or if supporting this service is possible and practical on public wide area networks.; In this research, we use Suite 2 of the Tenet Protocols as our reference implementation. Suite 2 is a group of network protocols, designed and built at U.C. Berkeley, that support guaranteed performance service in a multicast framework. Using Suite 2, we determine the scaling limitations that are inherent to hard state multicast protocols supplying guaranteed performance service. Our investigation primarily focuses on the signaling component of the protocols that is responsible for connection administration (e.g. setup, teardown and modification), although when necessary, we broaden the scope to consider other factors such as data delivery and routing. As we vary the size and number of connections, we evaluate the required protocol overhead in terms of size and number of messages, time-to-completion and success rate for establishment actions. Using this data, we suggest mechanisms and design options to make it more scalable, and evaluate the effects that these changes impose on the delivered service.; In final analysis, we will discover if the connection-oriented approach, required to provide a guaranteed performance service model, critically limits the scalability of the protocols. This analysis will allow us to show if this service model is practical for deployment over public WANs.
展开▼
