End-to-end shared restoration in multi-domain networks / by Zhiying Gao.

Abstract

Emerging multi-service data applications require high-bandwidth high-quality connectivity across multiple network domains, each of which is generally controlled by an independent service provider. These applications necessitate the need for highly intelligent survivable routing mechanisms to compute end-to-end paths and to perform functions of protection and bandwidth management across multiple domains. On the other hand, current protection and restoration mechanisms focus on the network survivability inside a single domain network. Powerful dynamic protection and restoration algorithms have been developed for single-domain networks. The majority of these algorithms are based on the exchange of detailed link-state information among the nodes, which makes them less attractive to networks with multiple domains where link-state information needs to be abstracted within each domain for efficiency and scalability reasons. To address this problem, we present two network information abstraction models designed to aggregate link-state information within each domain and only to advertise the aggregated information to other domains. The first abstraction model is referred to as virtual path abstraction model, with which every domain is abstracted as a set of border-nodes interconnected by virtual paths. The multi-domain network is then topologically aggregated to become a single-domain network, called virtual path network, which consists of border-nodes interconnected internally by virtual paths and externally by inter-domain links. The second abstraction model is referred to as virtual node model, with which every domain is modeled as a virtual node with a certain internal minimum capacity that can be advertised to other domains. The multi-domain network is then topologically aggregated to become a single-domain network, called virtual node network, consisting of virtual nodes interconnected by inter-domain links. We have designed and developed three distributed end-to-end shared restoration schemes based on the information abstraction models presented above. These three schemes are referred to as Link Disjointed Virtual Path (LDVP) restoration. Domain Disjointed Virtual Path (DDVP) restoration, and Link Disjointed Virtual Node (LDVN) restoration. The LDVP and LDVN schemes are designed to provide link diversity between the primary and backup paths of each demand, whereas the DDVP scheme is designed to compute a pair of domain-disjointed paths for the demand. We show that the proposed schemes are more scalable than the existing restoration schemes because they require less amount of link-state information to be advertised between the domains. This will reduce the routing message overhead and make the proposed schemes to be scalable to large multi-domain networks. We also evaluate the performance o f the proposed schemes in terms of capacity usage and restoration time through simulation experiments on two multi-domain networks; one is based on the NSF (National Science Foundation) network, and the other is based on the European Optical Network. The simulation results show that the proposed schemes save the backup bandwidth significantly because of the sharing of backup resources among failure-disjointed connections. The simulation results also show that the restoration time achieved by the proposed restoration schemes (over the multi-domain network) is around or less than 60 ms, which is within the range accepted in today’s networks.