Exploiting the Benefits of Virtual Concatenation in Optical Transport Networks

We explore the benefits of virtual concatenation in SONET/SDH-based traffic-groomable mesh networks. Our results qualitatively and quantitively demonstrate that virtual concatenation can improve bandwidth efficiency, simplify network control, and balance network load. 1. Background Optical SONET/SDH networks are the dominant infrastructure to support data and voice traffic in backbone and metro-area networks. With the maturity of WDM switching technology, we can now deploy a multi-service optical WDM network employing intelligent optical crossconnects (OXCs). But SONET/SDH will still be very important as the framing layer to support efficient and intelligent OAM&P functionalities in backbone and metro networks. As data traffic continues to increase substantially, the inefficiency of transporting variable-length packet through fixed-length SONET frames emerges as a major concern when network operators try to optimize the usage of their bandwidth to support various services, such as IP, frame relay, Ethernet, etc. In traditional SONET/SDH multiplexing hierarchy, frames of multiple low-speed traffic streams (say, STS-1 frame, approx. 51.84 Mbps) are combined to form the frame of a high-speed stream. In order to support high-speed traffic from single client source, e.g., an ATM switch, N “contiguous” lower-order SONET containers are merged into one of greater capacity. This is called SONET/SDH concatenation technique. Usually, SONET/SDH concatenation is implemented at certain speeds, such as STS-3c, STS-12c, etc, which provide a tiered bandwidth-allocation mechanism for different client services. Unfortunately, although “contiguous” and “tiered” concatenation is simpler for implementation, it is not very flexible or efficient, especially in a multi-service network. From network node perspective, traffic streams from different client network equipment are to be discretely mapped into different tiers of SONET bandwidth trunks (data containers), which may result in huge capacity waste. For example, carrying a Gigabit Ethernet connection using a concatenated OC-48 pipe (approx. 2.5 Gbps) leads to 60% bandwidth wastage. From a network perspective, time-slot contiguous requirement imposes a constraint for traffic provisioning and may degrade network performance in a dynamic traffic environment where resources are easy to be fragmented. This constraint also makes it more difficult for operators to perform efficient traffic grooming, i.e., packing different low-speed traffic streams onto high-capacity wavelength channels. 2. SONET Virtual Concatenation Virtual concatenation (VC) helps a SONET/SDH-based network to carry traffic in a finer granularity, and utilize link capacity more efficiently. The basic principle of VC [1, 2] is that, a number of smaller containers, which are not necessarily contiguous, are concatenated, to create a bigger container. Depending on a network’s switching granularity, virtual concatenation is possible for small container size from VC-1.5 upto STS-3c. Figure 1 shows an example of how to support multiple services using a single OC-48 channel through VC. In Fig. 1, an OC-48 channel is used to carry two Gigabit Ethernet, one 200 Mbps Fibre Channel, and two STS-1 TDM voice streams. Through a STS-1 switch, traffic can be switched onto different OC-12 pipes, and these OC-12 pipes can be sent through the network over various routes. Figure 1 also illustrates the potential load-balancing benefit brought by VC. Also, note that, when traffic from one client is sent over different routes, VC mappers at destination node need to compensate for differential delay between bifurcated streams when they are reconstructed. Currently, a commercially available device may support upto 50 ms (+/-25 ms) differential delay with external RAM, which is equivalent to a 10,000 km difference in route length [3]. In general, a virtually-concatenated SONET channel made up of N×STS-1 is transported as individual STS1s across the network; at the receiver, the individual STS-1s are re-aligned and sorted to recreate the original payload. Figure 2 provides an overview of a multi-service SONET/SDH-based optical network employing VC. VC can be supported at the edge OXC (in port cards) or in a separate traffic-aggregation elements connecting client equipment and the OXC. In this study, we explore, from a network perspective, benefits of SONET virtual concatenation in an optical WDM network under dynamic traffic. 3. Benefits of Virtual Concatenation: a Network Perspective From a network perspective, VC can benefit optical networks in following aspects: • Relax time-slot alignment and contiguity constraints. Instead of aligning to particular time-slots and consisting of N contiguous STS-1 time-slots in a wavelength channel, a high-speed STS-N channel can (theoretically) be made up by any N STS-1 time-slots and carried by different wavelength channels. • More efficiently utilize channel capacity to support multiple types of data and voice services. Instead of mapping data traffic (packet/cell/frame) into SONET frames in a discrete tiered manner, optical networks can carry data traffic in a more resource-efficient way. Traffic granularity can be increased in the unit of 1.6 Mbps (VT1.5) in metro-area networks, and 51 Mbps (STS-1) or 150 Mbps (STS-3c) in backbone networks. • Bifurcate traffic streams to balance network load. With VC, it is possible to split a high-speed traffic stream into multiple low-speed streams and route them separately through the network. This enables traffic to be distributed in the network more evenly, and hence improves network performance. Our investigation is based on discrete-event simulations. Connections with different bandwidth granularities come and leave network, one at a time, following Poisson arrival process and negativeexponential-distribution holding time. Two traffic patterns are studied (Pattern I and II), one consisting of five service classes and the other has ten service classes. Capacity of each wavelength is OC-192. In Pattern I, data rates for each class are approximately 51 Mbps, 153 Mbps, 622 Mbps, 2.5 Gbps, and 10 Gbps, which can be perfectly mapped into the tiered SONET container, i.e., OC-1, OC-3c, OC-12c, OC-48c, and OC-192c. Table 1 shows Pattern II’s service classes, service rates, and corresponding SONET containers with or without VC. When traffic bifurcation is needed, a simple route computation and traffic bifurcation heuristic is applied to a connection request. (We have proved that the problem of finding the set of minimal-cost routes with enough aggregated capacity for a request in a network is NP-complete [5].) The heuristic works as follows: • Step 1: A shortest path is computed according to link cost for the request. • Step 2: Bandwidth of the route is calculated. Bandwidth of the route is constrained by the link along the route with minimal free capacity. Then, update the available capacity of the links along the route. • Step 3: Remove the link without free capacity and repeat Steps 1 and 2 until connection can be carried by the set of routes computed or no more routes exist for the connection. Note that, depending on implementation, VC mappers at receiver node may only be able to handle certain number of routes (denoted by t) for a single connection, in which case the connection will be blocked after t routes have been examined, and connection has not been fully provisioned. 4. Illustrative Numerical Examples Figure 3 illustrates network performance in terms of bandwidth blocking percentage (BBP) as a function of offered load in Erlangs. BBP is used as measurement metric since connections from different classes have different bandwidth requirements. Network offered load is normalized to the unit of OC-192. Each fiber is assumed to support 8 wavelengths. Figure 3(a) shows network performance for Pattern I with or without VC. Two network configurations are examined, i.e., all nodes are either equipped with STS-1 full-grooming switches or partial-grooming switches. Note that, in a partial-grooming switch, only a limited number of wavelengths (6 in our simulations) can be switched to a separate grooming switch (or grooming fabric within an OXC) to perform traffic grooming. Observe that there is 5-10 percent performance gain by using VC. In Pattern I, every class can be perfectly mapped into one of tiered SONET containers, and no bifurcation is assumed in this example. Therefore, performance improvement shown in Fig. 3(a) comes solely from eliminating time-slot alignment and contiguity constraints provided by VC. This observation is similar to the effect of wavelength conversion in a wavelength-routed WDM network, where each connection requires full wavelength. Figure 3(b) shows how VC can improve performance when the network needs to support data-oriented services with different bandwidth requirements, assuming full-grooming OXCs everywhere. Now, BBP is significantly reduced by employing VC. Meanwhile, more improvement can be achieved by allowing a simple traffic bifurcation scheme. In our study, a connection will be bifurcated only if no single route with enough capacity exists. The results for different values of t have been examined and some are shown in Fig. 3(b), i.e., 4, 8, and unlimited. It is expected that more advanced load-balancing and traffic-bifurcation approaches can further improve network throughput. Additional results based on different network configurations (e.g., different number of wavelengths, different network topologies, different OXC switching capabilities) are not included here due to space limitation. Besides the three benefits we have quantitively demonstrated, SONET/SDH VC can also benefit an optical network on network compatibility, resiliency, and control and management. VC works across legacy networks. Only end nodes of the network are aware of containers being virtually concatenated. For resiliency, since individual members of a virtually-concatenated channel may be carried through different routes, a network failure may only affect partial bandwidth of a connection. Hence, a connection may still get service under reduced bandwidth before failures are fixed (best-effort services) or protection/restoration schemes are activated (priority services). With VC, link-state information can be maintained at an aggregated level since time-slot alignment and container-continuity constraints are handled by end nodes. Moreover, built on virtual concatenation, the link-capacity adjustment scheme (LCAS, approved by ITU-T as G.7042) allows network operators to adjust pipe capacity while in use (on the fly). This increases the possibility for on-demand traffic provisioning and on-line traffic grooming/re-grooming and makes SONET/SDH-based optical WDM network more data friendly. However, more intelligent algorithms and mechanisms need to be explored in order to fully utilize the benefits provided by this technique.