Experimental study of dynamic IP topology reconfiguration in IP/WDM networks

With the widespread deployment of IP over WDM networks, it becomes necessary to develop dynamic lightpath and topology reconfiguration mechanisms and reconfiguration triggers that can effectively exploit WDM reconfigurability. A reconfiguration model for overlay IP/WDM networks is introduced. An IP over WDM network testbed has been set up using Telcordia MONET WADMs and IP routers. The testbed network and software architecture is presented and the trafficengineering experiments and convergence measurements conducted over the testbed are reported. I. IP/WDM INTERNETWORKING ARCHITECTURE As the emergence of multi protocol lambda switching (MPλS) control technique and optical packet switching equipment, there are two IP over WDM networking architectures: IP over reconfigurable WDM and IP over switched WDM [1], [2]. The former is constructed upon a circuit-switched WDM network whereas the latter is based on a packet or label switched WDM network. Under the first architecture, established circuits form a WDM lightpath topology that is reconfigurable in response to traffic changes and/or network planning. In the second architecture, e.g. in the case of Optical Label Switching, there is a unified IP/WDM topology build upon the packet-switched WDM network. In a switched WDM network, optical headers are attached to the payload data and are processed at each network switch. We focus on IP over reconfigurable WDM networks in this paper. The reconfigurable IP/WDM network we envision consists of IP routers attached to a WDM core network, or interconnected OXCs that are capable of processing IP packets. The reconfigurable IP/WDM inter-networking architecture is defined essentially by the organization of the WDM control plane. Three network models have been proposed, overlay, peer, and augmented model [3]. Under the overlay network model, the IP routing, topology discovery and distribution, and signaling protocols are independent of the routing, topology discovery and distribution, and signaling at the WDM layer. There are two alternatives to interface between IP layer and WDM layer. First, the two layers can communicate through the WDM This research is partially funded by the US Government DARPA NGI SuperNet project under the contract F3060298-C-0202. network management system. There is no direct interface between IP network and WDM network. This is similar to IP over ATM Soft Permanent Circuits (SPVC). Second, the two networks can interface directly through Optical User Network Interface (O-UNI). O-UNI is an edge module for the WDM network to serve IP client requests. O-UNI only support interfaces of lightpath request, deletion, and query because of the limited information shared between the IP and WDM layer. With the peer network model, control in the IP/WDM network is facilitated by MPLS control plan. Each WDM node consists of an integrated IP router and an OXC. The interaction between the router and OXC is still being developed. Signaling among various nodes is achieved using CR-LDP and RSVP with proper extensions. Under this model, the IP layer nodes act as peers of the OXCs, a common addressing scheme, presumably IP, will be used for both IP and WDM networks, such that a single routing protocol instance runs over both the IP and the WDM networks. The augmented network model is located in between the overlay model and the peer model. Under the augmented model, there are actually separated routing instances in the IP and WDM domains, but control information from one routing instance is passed through to another routing instance. An example implementation of the augmented model is that each of the IP and WDM domain runs an IGP instance, and interacts with each another through an EGP. This paper reports a testbed implementation and experimentation over an overlay IP over WDM network. IP and WDM are interfaced through O-UNI. II. A RECONFIGURATION MODEL FOR IP OVER WDM NETWORKS To leverage on the flexibility provided by WDM reconfiguration, IP topology formed by lightpaths can be changed dynamically. Fault, restoration, network provisioning, or traffic engineering can trigger IP/WDM reconfiguration. However, reconfiguration causes network instability. During the process of topology migration, the ongoing IP traffic may be delayed or even lost. If reconfiguration takes considerable time to converge, the transit network may form forwarding loops and therefore wastes network resources. In this section, we detail the reconfiguration processes, introduce reconfiguration convergence, and discuss reconfiguration constraints. 76 0-7803-7206-9/01/$17.00 © 2001 IEEE In our work, for reconfiguration and routing purposes, we model non-wavelength interchange WDM switch as a group of nodes, each of which interconnects a bi-directional wavelength channel (see Figure 1). The logical node can still add, drop, or pass through the traffic. Fig. 1. WDM node modeling. A. Reconfiguration Processes In an overlay IP/WDM network, there are two tasks associated with IP topology reconfiguration, WDM reconfiguration and IP reconfiguration, respectively. WDM reconfiguration instructs the OXC and OADM to set up the desired lightpath topology and has the following components. • Lightpath routing tlr: if the detail hops of a lightpath are not given in the reconfiguration trigger, the end-to-end path has to be computed dynamically. An example approach for wavelength routing and assignment is to use Dijkstra SPF algorithm subject to constraints. Constraints that have to be considered include wavelength availability and wavelength continuity. • Lightpath topology setup tsetup: this includes a distributed signaling procedure and switches setup. Depending on the implementation, signaling may be responsible for local lambda selection as exploited in MPLS. Switch setup may require a reset operation before adding a new connection to the fabric. • Routing convergence twdm-c: this represents the time for the WDM routing information base to resynchronize after update. If a link state protocol is used in wavelength routing, this is the time for the link state database to converge. If WDM network uses a single and centralized connection manager to compute lightpaths, this represents the connection database update time once changes occurred. WDM reconfiguration time Twdm can be defined as ( ) c wdm c wdm setup lr t t t t − − × − + + β , 1 0 ≤ ≤ β , where β represents the overlapping factor between lightpath computation and setup time and WDM convergence time. IP reconfiguration alters IP interfaces status and address if necessary, and then waits for the routing protocol to converge. Hereafter, we use OSPF as the IP routing protocol as the link state protocol not only supports multiple metrics but also promises a faster convergence time. IP reconfiguration, Tip, includes the following components. • Interface reconfiguration tif: the time to change the IP interfaces as specified in the new topology. • Routing protocol convergence tip-c: the OSPF convergence time. This includes the time for detection, propagation, and SPF recalculation. The number of calculations that must be performed given n link-state packets is proportional to nlogn in a modern SPF algorithm. OSPF convergence time is related to the size and the type of the network, e.g., the number of routers within each area, the number of neighbors for any one router, the number of areas supported by any one router, and the designated router selection. Tip can be written as ( ) c ip c ip if t t t − − × − + γ , 1 0 ≤ ≤ γ , where γ represents the overlapping factor between interface configuration and OSPF convergence. B. Reconfiguration Convergence When IP network topology changes, IP traffic must reroute quickly based on the new lightpath topology. IP convergence time describes the time it takes to an IP router to start using a new route after a topology changes. Reconfiguration convergence refers to the time that an IP/WDM reconfiguration has completed and IP and WDM network has converged. I.e. after a reconfiguration time interval, the new IP/WDM network is ready for another reconfiguration. Reconfiguration convergence time Tr can be written as ( ) wdm ip T T α − + 1 , 1 0 ≤ ≤ α , where α is the overlapping factor between IP reconfiguration and WDM reconfiguration. To minimize Tr , IP and WDM reconfiguration should be conducted in parallel. However, migration schedule may require serialization between certain IP and WDM reconfiguration processes to reduce instability and/or avoid traffic lost. Application impact due to reconfiguration is within the interval of Tr twdm-c since applications do not require WDM network convergence. C. Reconfiguration Constraints There are two groups of WDM reconfiguration constraints: client-imposed and network-oriented reconfiguration constraints. Clients may prefer specific path or have special requirement for certain reasons. Client imposed constraints include • Source and sink end points: these are usually given in terms of switch names. The network control and management system (NC&M) will also accept port names. • Connection directionality: bi-directional or unidirectional. • Disjoint path: there are many scenarios that client may demand a disjoint path. For example, client asks for two bidirectional connections between the same source and sink switch with disjoint paths. Another example is requesting a bi-directional wavelength channel with disjoint fiber path. • Preferred lambda: different lambda on the same fiber may have different signal quality. λ Interchange Node λ Non-Interchange Node Traffic Add/Drop Port λ Channel 1