Design and analysis of survivable WDM optical network

Abstract

Optical networks with Wavelength Division Multiplexing (WDM) technology provide huge bandwidth to meet the ever-increasing traffic demand of the next generation Internet. But the high-speed nature of WDM networks also makes the network more vulnerable to failures. Even a single network failure for a very short duration can result in enormous loss of data. In this thesis, we concentrate on designing a survivable WDM network. In essence, survivability concerns two important aspects, fast fault detection and localization, and fast fault recovery.

We first study fast fault detection and localization in WDM optical networks. Our work is based on the notion of monitoring cycle (m-cycle). Compared with other fault detection schemes, an m-cycle based fault detection scheme provides fast fault detection and requires less number of expensive monitors. Aiming at further cutting down the implementation cost, we propose the notion of super monitor. Instead of having a dedicated monitor for each m-cycle, a single super monitor can be placed at the junction of a set of overlapped m-cycles. In this thesis, we formulate and solve the monitor placement problem.

We then focus on enhancing the capacity efficiency of fast fault recovery schemes. Shared backup path protection (SBPP) schemes can provide 100% protection against any single link failure. This is achieved by establishing a pair of link-disjoint active and backup paths upon each call arrival. The bandwidth on different backup paths can be shared for protecting different calls. In this thesis, a new SBPP scheme is designed based on a two-step routing approach, where the active and backup paths are sequentially optimized with different objectives in mind. We then shift our focus to design fast protection scheme for multicast/broadcast communications. To this end, we refine the existing concept of blue/red tree. Blue/red tree is a pair of spanning trees where the connectivity between the root and any destination node is ensured upon a network failure. In particular, two efficient integer linear programs (ILPs) are formulated for finding the optimal blue/red trees.

Last but not the least, we investigate the survivability in IP networks. We notice that existing efforts on IP fast reroute (IPFRR) are effective in enhancing the IP resilience. But the impact of IPFRR on the end-to-end TCP performance is ignored. Notably, path rerouting can interfere with the TCP congestion control mechanism and thus cause severe throughput degradation. To address this problem, we propose a duplicate acknowledgement (ACK) suppression scheme. The key idea is to detect whether an out-of-order packet arrival event is due to IPFRR or not. If it is due to IPFRR, duplicate ACKs triggered will be suppressed by the TCP receiver so as not to cause unnecessary slow down at the TCP sender.