| 摘要 |
In a packet switching network, a distributed architecture provides efficient computation of routes in Quality of Service (QoS)-based routing scenarios. Using a client-server model, only designated route servers store and maintain a databa se containing the entire network topology, so that each network node is not require d to store and maintain the network topology. Client nodes maintain a cache containing pre-computed routes so that they can often make routing decisions autonomously. A client contacts a designated route server only when the client cannot obtain from its local cache a route to a given destination that meets the performance requirements. A client cache may contain pre-computed routes with designated QoS profiles to all destinations or to a subset of destinations. Rout e servers may also contain caches, which may contain pre-computed routes to all destinations in the network with all QoS profiles, or may contain only a subset of such routes. Each client node may also be provided with intelligence to learn, maintain and adapt local information based on the statistical usage of the network. Clien t caches may learn statically, i.e, the cache contains routes based on a QoS profi le provided by the network service provider, or they may learn dynamically, i.e., routes are modified based on ongoing network usage statistics. The goal is to minimize the need to contact the route server as much as possible. Protocols are defined to maintain synchronization between the route server and its clients distributed across the network. These protocols need to be observed to ensure th at all nodes have the latest view of the network topology stored at the route serve r.
|
