Demetrios Stamatelakis

Cycle-oriented distributed preconfiguration: ring-like speed with mesh-like capacity for self-planning network restoration

Cycle-oriented preconfiguration of spare capacity is a new idea for the design and operation of mesh-restorable networks. It offers a sought-after goal: to retain the capacity-efficiency of a mesh-restorable network, while approaching the speed of line-switched self-healing rings. We show that through a strategy of pre-failure cross-connection between the spare links of a mesh network, it is possible to achieve 100% restoration with little, if any, additional spare capacity than in a mesh network. In addition, we find that this strategy requires the operation of only two cross-connections per restoration path. Although spares are connected into cycles, the method is different than self-healing rings because each preconfigured cycle contributes to the restoration of more failure scenarios than can a ring. Additionally, two restoration paths may be obtained from each pre-formed cycle, whereas a ring only yields one restoration path for each failure it addresses. We give an optimal design formulation and results for preconfiguration of spare capacity and describe a distributed self-organizing protocol through which a network can continually approximate the optimal preconfiguration state.

IP layer restoration and network planning based on virtual protection cycles

We describe a novel restoration strategy called virtual protection cycles (p-cycles, patents pending) for extremely fast restoration in IP networks. Originally conceived for use in WDM and Sonet transport networks, we outline the adaption of the p-cycle concept to an IP environment. In an IP router-based network, p-cycles are implemented with virtual circuits techniques (such as an MPLS label switched path, or other means) to form closed logical loops that protect a number of IP links, or a node. In the event of failure, packets which would normally have been lost are encapsulated with a p-cycle IP address and reenter the routing table, which diverts them onto a protection cycle. They travel by normal forwarding or label switching along the p-cycle until they reach a node where the continuing route cost to the original destination is lower than that at the p-cycle entry node. Diverted packets are deencapsulated (dropped from the p-cycle) at that node and follow a normal (existing) route from there to their destination. Conventional routing protocols such as OSPF remain in place and operate as they do today, to develop a longer term global update to routing tables. Diversionary flows on the p-cycle inherently cease when the global routing update takes effect in response to the failed link or node. The p-cycle thus provides an immediate real-time detour, preventing packet loss, until conventional global routing reconvergence occurs. The aim of the paper is to explain the basic p-cycle concept and its adaptation to both link and node restoration in the IP transport layer, and to outline certain initial results on the problem of optimized design of p-cycle based IP networks.

New options and insights for survivable transport networks

This article is devoted to a selection of recent topics in survivable networking. New ideas in capacity design and ring-to-mesh evolution are given, as well as a systematic comparison of the capacity requirements of several mesh-based schemes showing how they perform over a range of network graph connectivity. The work provides new options and insights to address the following questions. How does one evolve from an existing ring-based network to a future mesh network? If the facilities graph is very sparse, how can mesh efficiency be much better than rings? How do the options for mesh protection or restoration rank in capacity requirements? How much is efficiency increased if we enrich our network connectivity? We also outline p-cycles, showing this new concept can realize ring-like speed with meshlike efficiency. The scope is limited to conveying basic ideas with an understanding that they could be further adapted for use in IP or DWDM layers with GMPLS-type protocols or a centralized control plane.