Time: October 1, 2001 (Monday), 9am-11am
Room: SIS Building 503, 135 N Bellefield Ave.
Network survivability technique is a countermeasure to the vulnerability of communication networks in the face of failures and malicious attacks. With the expansion of network capacity and increasing concentration of services in backbone networks, reducing response time and improving cost efficiency to provide survivable network services become increasingly important.
In this dissertation, I study a critical component of network survivability technique, i.e. spare capacity allocation (SCA) problem on mesh networks. The SCA problem is to decide how much spare capacity should be reserved on links and where to pre-plan backup paths to protect traffic from a set of failures.
I formulate the SCA problem using a novel matrix-based network flow model. I provide bounds on the network redundancy of this NP-complete optimization problem. I develop an approximation algorithm called successive survivable routing (SSR). Compared with other algorithms in numerical results, SSR achieves a very good trade-off between near-optimality and computational speed. Moreover, the matrix model and SSR algorithm are extended to the cases of different path restoration schemes, arbitrary failure scenarios, and objectives with non-linear link costs.
Another contribution of this dissertation is to extend the above model to topology layout and spare capacity design in multi-layer networks. To make an upper layer network resilient from a set of lower layer failures, I use a matrix-based layout model to find an optimal topology. Such solution is then used in the matrix-based SCA formulations and the SSR algorithm for multi-layer networks is developed. Applications in fault-tolerant virtual private network (VPN) design are provided.
I also propose two alternative SCA algorithms. One is to randomly scale the path sets for the branch and bound algorithm and the other is a simulated annealing algorithm. Both of them take longer execution time than SSR but also find near optimal solutions.
In short, I give a two-pronged attack to SCA problem - a matrix-based model to disclose the problem structure, and a routing-based algorithm to find near-optimal solutions. These results are further extended to multi-layer survivability problems with important applications.
Dissertation electronic version: yuliu.pdf
Related publications are kept on my vita page.
Slides of the dissertation defense talk: yuliu.ppt(1.4MB), an updated version for INFORMS Telecom 2002: BocaTE01A.ppt(1.5MB).
Related talks include Boca Multilayer Survivability talk, Globecom'01 talk, INFOCOM'01, IEEE ComSoc PGH seminar, and several other presentations.