Evaluation of novel resilience schemes in dynamic optical transport networks

Monika Jaeger, Ralf Huelsermann, Dominic A. Schupke, Rene Sedlak, Shizhong Xie, Chunming Qiao, Yun Chur Chung
2002 Optical Networking II  
Today, Wavelength Division Multiplexing (WDM) transmission systems are deployed extensively in transport networks. They are used mainly for static point-to-point connections. With the availability of fast reconfigurable Optical Cross Connects (OXC) and the introduction of a control plane in the Optical Transport Network (OTN), optical channel based logical networks can be built for dynamic WDM networks. Resilience in current transport networks is mainly based on static SONET/SDH dedicated and
more » ... ared protection. Distributed control planes allow new, flexible protection mechanisms (e.g. GMPLS reroute and fast reroute). To evaluate future distributed control concepts and new resilience schemes in transport networks, we have implemented a dynamic OTN simulation model. Several case studies have been performed using different protection and restoration methods. Different failure scenarios (single or multiple link failures) were used. The paper evaluates the case studies in terms of scalability, recovery time criteria, capacity use (efficiency) and availability. It is shown that the new and flexible resilience schemes are a promising alternative to traditional statically preplanned protection in transport networks. Furthermore, they provide increased network availability in multiple failure cases. distributed routing and signaling functions needed for connection control [1] . With the availability of a control plane for the optical layer, fast and efficient MPLS-like restoration mechanisms based on optical channels can be introduced in OTNs. In the following, we investigate and evaluate new resilience schemes for future dynamic optical transport networks. II. ROUTING IN OPTICAL NETWORKS An optical channel is a wavelength based end-to-end connection through the optical layer of an OTN. There are different methods for calculating optical channel routes in optical networks. Other than in shortest path IP-routing schemes, in optical networks it is necessary to ensure that a chosen path provides enough resources, i.e. free wavelengths on all links. We call the shortest path with sufficient resources at the time of the setup request the 'shortest available path'. IP-routing protocols use distributed routing mechanisms, i.e. each packet will be routed hop-by-hop through the network, where the routers along the route only choose the best next hop. In optical networks, resource information has to be distributed in addition to topology information. Due to the above described resource availability requirement on 'shortest available paths' and due to the difference that wavelength switched optical networks are connection-oriented networks as opposed to packet-oriented IP networks, the distribution of routing decisions is not as simple as in packet based networks. In our simulation models we therefore use explicit source routing. Each router in the network has to have the same knowledge about the current network state, and the routers at the source compute the complete routes through the network to the destination, originating from itself. The most simple way to model consistent topology databases distributed over all routers is to implement a unique central resource database, to which all routers have simultaneous access. With this approach the problem of inconsistent databases is negligible. With the availability of topology information the routers are enabled to construct network graphs. The commonly used algorithm to compute shortest paths in graphs is the Dijkstra Algorithm. There are several alternatives when to calculate paths. First, using a pre-calculation the router calculates routes to a limited set of destinations or to all possible destinations during the initialization phase. In case of an incoming connection request the router has to check only, whether there are resources available along the pre-calculated route. If not, the connection gets blocked. This method reduces the processor load during the working phase, because the processing intensive route calculation has been done before. On the other hand, this scheme does not take alternative available routes based on the current network state into account. If the route calculation occurs online after the connection request has arrived, it is possible to manipulate the network graph with current resource information. A modification of the route pre-calculation is the computing of more than one route between two nodes. In this case a set of routes can be checked online for free resources. There are several algorithms for computing k-shortest paths in the literature [3] . We consider two types of OXCs: first, so-called opaque OXCs with an electrical switching backplane, and second, transparent OXCs which switch signals in the optical layer. In the latter, there is no opto-electronic conversion and the signals are switched continuing on the same wavelengths. Thus, routes in transparent all-optical networks have to satisfy the Routing and Wavelength Assignment (RWA) constraints, where available paths must be wavelength continuing. The transparency length of all-optical connections is limited through degradation effects. However, throughout this paper we assume all transparent optical channel connections are shorter than the maximum transparent path length. We call optical networks with opaque switching nodes opaque networks and optical networks with transparent switching nodes transparent networks. One method to compute routes in all three types of optical networks is the layered-graph scheme [ Figure 1 ]. Every layer reflects a certain wavelength. A link is represented by all edges (wavelengths) lying on top of each other in the graph. At opaque nodes additionally vertical edges are inserted, connecting the layers and, thus, enabling wavelength conversion. The size of the resulting adjacency matrix [ Figure 1 ] scales quadratically with the number of wavelengths and nodes (every new link/node increases the size of the topology by the order of # of wavelengths), but allows to obtain the wavelength specific shortest available path on all links within one calculation step, since the topology distinguishes different wavelengths, i.e. wavelength-specific routes are found. Currently, we set the costs for the vertical edges indicating wavelength conversion to zero. In future studies we will test if non-zero costs can help to reduce wavelength conversion costs.
doi:10.1117/12.482424 fatcat:mobfl7ps3jhrrgfzrivsgenw7q