570 NETWORK SURVIVABILITY Figure 10.18 Comparison of (a) 1 + 10MS and (b) 1 + 10Ch protection schemes. 10.5.1 1 + 1 OMS Protection This is perhaps the simplest optical layer protection scheme and is shown in Fig- ure 10.18(a). Because of its simplicity, it has been implemented by several vendors in their OLTs. The composite WDM signal is bridged onto two diverse paths using an optical splitter. At the other end, an optical switch is used to select the better among the two signals, based primarily on detecting the presence or absence of light signals. The split incurs an additional 3 dB loss, and the switch also adds a small amount of loss (< 1 dB). An alternative implementation uses optical amplifiers on each of the fibers and a passive combiner to combine both directions at the receiver. At any time, one amplifier is turned on and the other is turned off. This has the advantage of avoiding a single point of failure in the system (the selector switch in other implementations), but may be more expensive to implement. 10.5 Optical Layer Protection Schemes 571 10.5.2 1 "1 OMS Protection This scheme is similar to the SONET 1:1 scheme discussed in Section 10.2.1 and the benefits are similar: support for low-priority traffic and also the ability to have N working systems share a single protection system. Compared to the 1 + 1 scheme of Figure 10.18(a), a typical implementation uses a switch at the transmitter, instead of a splitter, resulting in a somewhat lower total loss in the path. Just as in the SONET equivalent, an APS protocol is needed to provide coordination between the two ends of the link. 10.5.3 OMS-DPRing The OMS-DPRing (dedicated protection ring) is similar to a SONET UPSR, except that it operates at the OMS (or optical line) layer, whereas the UPSR operates in the SONET path layer. It can also be thought of as an optical unidirectional line-switched ring (ULSR). One possible implementation of an OMS-DPRing [Bat98] is shown in Fig- ure 10.19. Signals are coupled into and out of the ring via passive couplers. Each node transmits on both directions of the ring. Note that different nodes must transmit at different wavelengths; otherwise their transmissions would collide. Under normal operation, the ring functions as a bus, with one pair of amplifiers turned off on the entire ring and all the others turned on. If there is a link failure, the amplifiers next to the failed link are turned off and the ones that were originally inactive are now turned on to restore traffic. For example, in Figure 10.19(a), the amplifier pair to the right of node A is turned off under normal operation and the other amplifiers are turned on. In Figure 10.19(b), when link CD fails, the amplifier pair at C adjacent to the failed link is turned off, and the originally inactive amplifiers at node A are turned onto create a new bus and restore traffic. 10.5.4 OMS-SPRing The OMS-SPRing (shared protection ring) is analogous to a SONET BLSR/4 with some changes. A possible implementation of a four-fiber ring is shown in Fig- ure 10.20. Two of the fibers have WDM equipment deployed, and the remaining two fibers around the ring are used for protection purposes and do not have at- tached WDM equipment. In the event of a cut, the signal is either span switched or ring switched onto the protection fibers, as shown in Figure 10.21. In both cases, not having WDM equipment on the protection fibers not only saves cost but also pro- vides a relatively lower-loss path around the ring for the protection traffic. Optical amplifiers may be needed on the protection fibers depending on the link losses. 572 NETWORK SURVIVABILITY Figure 10.19 OMS-DPRing protection. (a) Normal operation. One pair of amplifiers is inactive (turned off) and the others are turned on, creating a bus. (b) After a failure, the currently inactive amplifiers are turned on and an amplifier pair adjacent to the failure is turned off to bring up the alternate path and restore traffic. 10.5 Optical Layer Protection Schemes 573 Figure 10.20 O MS-SPRing shown under normal operation. Only the working fibers are connected to optical add/drop multiplexers. The protection fibers are connected around the ring. A two-fiber version of OMS-SPRing can also be realized by dedicating half the wavelengths on each fiber for protection purposes. By making sure that protection wavelengths on one fiber correspond to the working wavelengths on the other fiber, the signals can be rerouted without requiring wavelength conversion. This scheme, however, requires the two groups of wavelengths to be demultiplexed and multi- plexed at each node, and thus is not strictly operating at the OMS layer. 10.5.5 I"N Transponder Protection The OMS layer schemes that we discussed above handle link failures and node failures but do not handle failures of the end equipment, particularly the transpon- ders. The transponders may be protected in a I:N configuration by having a spare transponder for every N working transponders. One problem to overcome is that transponders today operate at fixed wavelengths, and so the spare transponder will operate at a different wavelength than the working transponder. When the signal is switched over to the spare transponder, we also need to set up a new lightpath on the new wavelength through the network. Alternatively, we could use a tunable laser in the spare transponder. 574 NETWORK SURVIVABILITY Figure 10.21 OMS-SPRing after a failure. (a) Span switching. (b) Ring switching. 10.5.6 1 + 1 OCh Dedicated Protection In 1 + 10Ch protection, two lightpaths on disjoint routes are set up for each client connection. As shown in Figure 10.18(b), the client signal is split at the input and the destination selects the better of the two lightpaths. As with SONET and SDH, no signaling is required. This approach works in point-to-point, ring, and mesh configurations. In the context of a ring, the scheme is also called OCh-DPRing (OCh dedicated protection ring) or optical UPSR. Like SONET UPSRs, this approach is bandwidth inefficient in that the protection bandwidth is not shared among multiple client connections. However, it is one of the simplest protection schemes and therefore has been implemented by several vendors in optical add/drop multiplexers and crossconnects. Figure 10.22 shows another possible implementation of the bridge and select functions within a node. Here, the signal entering the optical layer is split and sent to two transponders, and then diversely routed across the network. At the receiving end, the signal is terminated in two transponders, and the better signal is selected afterwards to be sent to the client. In Figure 10.18, the client signal is passed through 10.5 Optical Layer Protection Schemes 575 Figure 10.22 Another implementation of 1 + 10Ch protection. The signal from the client equipment is split and sent to two transponders for transmission over diverse paths, and at the destination the better copy is selected by an optical switch at the output of the transponders. a transponder and split afterwards. At the receiving end, one of the two signals is selected by an optical switch before it is sent into a transponder and then onwards to the client. This uses half as many transponders as the previous option but does not protect against a transponder failure. Aside from this aspect, there are several other subtleties that affect the choice of one implementation versus the other, such as the criteria for switching from one path to another, and potential restoration time differences between the two approaches. 10.5.7 OCh-SPRing The OCh-SPRing (shared protection ring) is somewhat similar to a SONET BLSR/4. However, the BLSR operates at the line (multiplex section) layer, whereas this scheme operates at the optical channel layer and not the optical multiplex section layer. Working lightpaths are set up on the shortest path along the ring. When a working lightpath fails, it is restored either using a span switch or a ring switch, just as in a SONET BLSR/4. Nonoverlapping lightpaths in the ring can share a single wavelength around the ring for protection, and this spatial reuse allows the OCh-SPRing to be more efficient than an OCh-DPRing for distributed traffic. The operation of the OCh-SPRing is essentially the same as that shown in Figures 10.5-10.7, where the fibers now correspond to wavelengths and the connections correspond to lightpaths. 576 NETWORK SURVIVABILITY Just as with a BLSR, fast coordination between the ring nodes is needed in order to support node failures or low-priority traffic. 10.5.8 OCh-Mesh Protection Ring architectures are inherently suitable for sparse physical topologies and in situa- tions where most of the traffic is confined within the ring. Many backbone networks tend to be somewhat more densely connected than rings and are essentially meshed, with traffic being fairly distributed. A typical North American long-haul carrier's backbone network may have, say, 50 nodes, with an average node having 3-4 ad- jacent nodes, with some nodes having as many as 5-10 adjacent nodes. For such networks, mesh protection schemes offer more bandwidth-efficient protection than rings. The bandwidth efficiency of a mesh relative to a ring depends on several factors, including the network topology, the traffic pattern, and the type of mesh protection scheme used. In general, the more dense or meshed the topology, the greater the benefit of mesh protection. Also, if traffic in the network is primarily localized, then rings can do a good job. In contrast, if traffic in the network is distributed, then rings are inefficientnmany lightpaths will need to be partitioned into multiple rings, and multiple rings need to be interconnected and protected to support these lightpaths. Efficiency improvements ranging from 20% to 60% have been reported for mesh protection schemes relative to ring protection schemes [RM99a, RM99b]. Here we provide a simple example to illustrate the efficiency of mesh protection relative to ring protection. We will look at a more realistic detailed example in Section 13.2.6. Example 10.1 Consider the network shown in Figure 10.23(a), with three lightpaths to be supported. Assume that all these lightpaths need to be protected. Each lightpath uses 1 unit of capacity on each link that it traverses. First suppose we use 1 + 10Ch dedicated protection. We would then set up dedicated protection lightpaths as shown in Figure 10.23(b). In this case, a total of eight units of protection capacity is needed in the network. Next let us consider a configuration that uses shared ring protection (OCh-SPRing). Here we have an interesting problem of how to configure the rings themselves. One solution is to configure the rings as shown in Figure 10.23(c). In this case, lightpaths X and Y each share the same bandwidth for protection, while lightpath Z has a separate ring for protection. This configuration requires a total of eight units of capacity for protection, which is the same as for dedicated protection above. Note, however, that the protection capacity can be reduced to six units by having lightpaths X and Y share a ring but using dedicated protection for lightpath Z. Another way to look at this is that by using the eight units of 10.5 Optical Layer Protection Schemes 577 Figure 10.23 Example to illustrate the bandwidth efficiency of mesh protection rela- tive to ring protection. (a) A mesh network with three lightpaths present. (b) Protect- ing the lightpaths using 1 + 1 dedicated protection. (c) Protecting the lightpaths using OCh-SPRing protection. (d) Protecting the lightpaths using OCh-mesh protection. capacity, we can support additional lightpaths that can share the ring used to protect lightpath Z. We now consider the case of shared mesh protection. Our mesh protection scheme works as follows. We will use the same routes used by the 1 + 1 scheme for routing the protection lightpaths. The big difference is that the protection lightpaths are not set up ahead of time, but are only set up when there is a failure. 578 NETWORK SURVIVABILITY As long as two lightpaths don't fail simultaneously, we can have them share the same protection capacity in the network. In this case, only a single lightpath fails at any given time, assuming we have to deal only with link failures. Therefore we only need to provide sufficient protection bandwidth to protect one lightpath at a time. We leave it to the reader to verify that the four units of capacity shown in Figure 10.23(d) are sufficient. Mesh protection schemes are not new. They were used in the 1980s in networks with digital crossconnects. However, these protection schemes were centralized and operated rather slowly, taking minutes to hours to restore traffic after a failure. Also the protection was complex to manage, and there were no applicable standards. After the standardization of SONET/SDH and due to the fast 60 ms ring protection offered by SONET/SDH, these mesh-based restoration schemes were largely abandoned. Today, we are seeing a resurrection of mesh protection schemes in the optical layer of the network for several reasons. 9 The processing power available to implement mesh protection has dramatically increased over the past few decades, to the point where computationally inten- sive functions such as determining new routes can be performed rapidly. The communication bandwidth available for network control purposes has also gone up dramatically. To protect a network providing terabits/second of capacity, it is quite reasonable to dedicate several 2 Mb/s or 45 Mb/s lines in the network for control traffic. This was not the case earlier, where this amount of band- width would have been considered large, relative to the actual traffic within the network. 9 Optical crossconnects and other optical layer equipment protect bandwidth at much larger granularities (lightpaths) than digital crossconnects that operate at DS1 or DS3 speeds. As a result, they have fewer entities to manage and protect. However, this situation will change as traffic grows. 9 Relatively fast signaling and routing protocols have been developed for other forms of data networks, such as IP and ATM networks, and many of these protocols can be adapted for use in the optical layer. 9 The 60 ms protection time requirement is not a hard number. Many carriers interested in protecting data traffic will be satisfied with protection times on the order of a few hundred milliseconds, making it easier to implement more complex protection schemes. A variety of mesh protection schemes have been proposed, and many are cur- rently being implemented by optical crossconnect vendors. In addition to the factors 10.5 Optical Layer Protection Schemes 579 discussed above, the mesh protection schemes will have to overcome some key issues in order to facilitate widespread deployment: 9 Part of the reason that SONET/SDH protection has been so successful is that the protection schemes were standardized. This is yet to happen with mesh protection schemes. 9 One of the advantages of ring-based schemes is that the network is partitioned into multiple domains and each domain is protected independently. Thus one part of the network does not affect the other parts. This implies that the network can handle simultaneous multiple failures as long as they occur in different domains. Moreover, one part of the network can be serviced without impacting the protection scheme in the other parts. In order to get the full benefit of mesh protection, we will need to treat the network in its entirety as a single domain. Breaking up the network into smaller domains reduces the bandwidth efficiency unless the individual domains are reasonably large. Another dimension to this is the effect of software bugs or operator errors. In ring-based networks, such problems are localized, whereas in mesh networks, these problems can have a networkwide impact. 9 Mesh protection schemes are considerably more complex to manage than ring protection schemes. In order to make them successful, vendors will need to provide carriers with the appropriate management tools to hide the complexity from the network operators. For instance, this could mean providing automated tools to plan and compute primary and protection routes in the network, which are otherwise fairly complex operations. On the plus side, however, interconnecting rings is fairly complex, and mesh protection allows for more flexible planning of capacity in the network capacity does not have to be nailed down up front; instead it can be provisioned as needed across the network. 9 The more efficient mesh protection schemes will require rapid networkwide sig- naling mechanisms to be implemented to propagate information related to fail- ures and to reroute lightpaths that are affected by a failure. This in turn implies that the nodes performing the protection switching will have to be designed carefully to minimize processing latencies. 9 The more efficient mesh protection schemes require that protection routing tables be maintained at the nodes. These routing tables provide information about the network topology and protection paths in the network. The tables need to be updated when lightpaths, links, or nodes are added or removed from the network. Most importantly, these tables need to be consistent across all the nodes in the network. . connected than rings and are essentially meshed, with traffic being fairly distributed. A typical North American long-haul carrier's backbone network may have, say, 50 nodes, with an average node. the currently inactive amplifiers are turned on and an amplifier pair adjacent to the failure is turned off to bring up the alternate path and restore traffic. 10.5 Optical Layer Protection. optical layer equipment protect bandwidth at much larger granularities (lightpaths) than digital crossconnects that operate at DS1 or DS3 speeds. As a result, they have fewer entities to manage