Multilayer network survivability models and application

This abstract is submitted in support of the INFORMS Telecommunications Dissertation Award [1]. Internet backbone networks are evolving to a layered architecture where IP with Multi-Protocol Label Switching (MPLS) network is on top of Wavelength Division Multiplexing (WDM) optical networks. Providing survivability in such networks presents several challenging problems. The requirements for rapid restoration and the conflicts between the distributed routing in MPLS and cost-driven centralized network management have suggested to precalculate backup paths and spare capacity before failure happens [2]. This decision is also driven by the increasing requests for bandwidth-on-demand services. These requests have fostered an open market to trade network capacity. How to improve the cost performance ratio of spare resource has recently become one of research focuses and many algorithms have been proposed [3], [4], [5], [6], [7], [8], [9], [10]. There are two similarities among these algorithms. Firstly, they utilize shortest path algorithm to preplan backup paths. By finding backup paths for each flow, the original spare capacity allocation problem, as a multicommodity flow problem, is partitioned into multiple smaller and easier ones. Secondly, these algorithms share link spare capacity protecting different failure cases. The spare capacity sharing happens at the overlapped segments on backup paths of two or more traffic demands whose working paths are disjoint from any failure case (not affected by the same failure case simultaneously). One of the benefits of using the shortest path algorithm is to be implemented in distributed routing protocols. The spare capacity sharing scheme reduces network redundancy significantly but its requirement of saving per-flow state information becomes a bottleneck. Resource aggregation for Fault Tolerance (RAFT) uses only hop count [3] to route backup paths. Sharing with partial information (SPI) algorithm avoids this bottleneck by using link metrics calculated on an estimation of the spare capacity sharing information [4], [5]. Recent works show that only a small amount of aggre-