Efficient Channel Reservation for Backup Paths in Optical Mesh Networks Somdip Datta* , Sudipta Sengupta ¶ , Subir Biswas ¶ , and Samir Datta ¶ *Electrical Engg., Princeton University, Princeton, NJ 08544, USA. ¶ Tellium, Inc., 2 Crescent Place, Oceanport, NJ 07757-0901, USA. Email: datta@princeton.edu, {sudipta, sbiswas, sdatta}@tellium.com Abstract- In an optical mesh network, backup channels are shared between multiple lightpaths to reduce restoration capacity overhead. The sharability of channels is usually constrained by the mandate to provide 100% recovery of all lightpaths affected by any single event failure in the network. This paper proposes a pool based channel reservation scheme that is optimal when the set of primary and backup paths (specified at link level without channel allocation) is given. In the online case, our simulations on representative network topologies show that this method improves over the existing (more restrictive) method of allocating shared backup channels using primary path diversity. I. INTRODUCTION Optical backbone networks with high capacity and reliability have fueled the internet capacity explosion in the last decade. As capacity requirements continue to grow rapidly and unpredictably, driven by diverse applications involving voice, video and data, mesh architecture [1] has emerged as the solution for rapid and efficient deployment of capacity in longhaul optical backbone networks. An optical mesh network consists of Optical Cross Connects (OXCs), interconnected by fiber links containing many optical channels. The basic service provided by the network is to setup a sequence of wavelength channels between two access points in the network, called a lightpath. The clients at the edge of the network (IP/ATM routers, Optical Add- Drop multiplexers [1]) use these lightpaths as high capacity pipes for data/voice traffic. The critical nature of these lightpaths implies that any fiber cut or equipment failure will be catastrophic unless a restoration mechanism is in place to reroute affected lightpaths along alternate routes. In a reactive restoration scheme [2], when a primary path fails, a search is initiated to find a backup path that does not involve the failed components. Though such a scheme tends to be very efficient in terms of capacity overhead required in the network, it does not guarantee successful recovery since the search for the backup might fail due to unavailability of resources. Furthermore, the process of searching for a suitable backup path introduces substantial delay in the restoration process, which may not be tolerable for many applications. In a proactive approach [2, 3], a backup path is found and resources reserved along it at the time of establishing the primary path itself. If the backup and primary are ensured to be link-diverse, this method yields a 100% restoration guarantee against any single link failure in the network. In a 1+1 scheme [2], all channels along the backup path are exclusively reserved for the lightpath. For a more efficient utilization of resources, any channel on a backup path may be used by two or more backup paths if their corresponding primary paths are link-diverse. Since such primary paths will not fail simultaneously, 100% restoration guarantee against single link failures is provides. Such a shared mesh restoration scheme [2, 3] significantly reduces the capacity overhead required for restoration. The implementation [2, 3] of the proactive shared mesh restoration scheme embeds the sharing concept in the search algorithm for a backup path as follows: If a certain channel has been already reserved for one or more backup paths, it can be reused for another backup path if and only if all of their primary paths use mutually disjoint sets of links. In Section II, some key definitions have been listed that help in defining this concept more formally in Section III.A. In Section III.B, we present a capacity model defining the cost minimization problem for a shared mesh network, subject to 100% restoration guarantee from any single link failures. With the help of a simple example in Section III.C, we show that the existing implementation is over restrictive with respect to the model in defining the sharability and leads to a capacity penalty. In Section IV, as the main contribution of this paper, we present a different approach for sharing backup paths, developed directly from the model in Section III.B, in which channels used for restoration in a particular link are placed into a common pool rather than being marked individually for particular (or set of) backup paths. Such an implementation has several advantages including less restoration capacity overhead, less data storage at nodes, and in addition, it provides easier access to control and monitoring of certain vital parameters related to service quality and protocol congestion delays during failure scenarios. In the Section V, we present the performance comparison of the two implementations in a simulation setup involving 2104 0-7803-7206-9/01/$17.00 © 2001 IEEE