520 CONTROL AND MANAGEMENT 9.6.2 able to add additional wavelengths (up to a designed maximum number) without disrupting the operation of the existing wavelengths. Also, ideally the failure of one channel shouldn't affect other channels, and the failed channel should be capable of being serviced without affecting the other channels. An issue that comes up in this regard is the use of arrayed multiwavelength components versus separate com- ponents for individual wavelengths, such as multiwavelength laser arrays instead of individual lasers for each wavelength. Using arrayed components can reduce the cost and footprint of the equipment. However, if one element in the array fails, the entire array will have to be replaced. This reduces the system availability, as replacing the array will involve disrupting the operation of multiple channels, and not just a single channel. Using arrays also increases the replacement cost of the module. Therefore there is always a trade-off between obtaining reduced cost and footprint on one front against system availability and replacement cost on the other front. We may also want to start out by deploying the equipment in the form of a point-to-point link and later upgrade it to handle ring or other network configura- tions. We may also desire flexibility in associating specific port cards in the equipment with specific wavelengths. For example, it is better to have a system where we can choose the wavelength transmitted out of a port card independently of what slot it is located in. Another problem in WDM systems is the need to maintain an inventory of wavelength-specific spare cards. For example, each channel may be realized by using a card with a wavelength-specific laser in it. Thus you would need to stock spare cards for each wavelength. This can be avoided by using a wavelength-selectable (or tunable) laser on each card instead of a wavelength-specific laser; such devices are only now becoming commercially available at reasonable cost. Connection Management The optical network provides lightpaths, or more generally, circuit-switched connec- tions, to its user. Connection management deals with setting up connections, keeping track of them, and taking them down when they are not needed anymore. The traditional telecommunications way of providing this function is through a centralized management system, or rather a set of systems. However, this process has been extremely cumbersome and slow. The process usually involves configuring equipment from a variety of vendors, each with its own management system, and usually one network element at a time. Moreover, interoperability between manage- ment systems, while clearly feasible, has been difficult to achieve in practice. Finally, service providers in many cases deploy equipment only when needed. The net result of this process is that it can take months for a service provider to turn up a new connection in response to a user request. Given this fact, it is not surprising that 9.6 Configuration Management 521 once a connection is set up, it remains in effect for a fairly significant period of time, ranging from several months to years! As optical networks evolve, connections are getting more dynamic and networks are becoming bigger and more complex. Service providers would like to provide connections to their customers rapidly, ideally in seconds to minutes, and not impose long-term holding time commitments on these connections. In other words, users would dial up bandwidth as needed. Supporting all this requires carriers to predeploy equipment (and bandwidth) ahead of time in the network and having methods in place to be able to turn on the service rapidly when needed. This is becoming a significant competitive issue in differentiating one carrier from another. This method of operation also stimulates what is called bandwidth trading, where carriers trade their unused bandwidth with other carriers for increasingly shorter durations to improve the utilization of their networks and maximize their revenue. Due to the reasons above, we are seeing a trend toward a more distributed form of control for connection management. Distributed control protocols have been used in IP and ATM networks. They have also had a fair degree of success with respect to standardization and accomplishing interoperability across vendor boundaries. We can make use of similar protocols for performing these functions in the optical layer. Distributed connection control has several components to it: Topology management. Each node in the network maintains a database of the net- work topology and the current set of resources available as well as the resources used to support traffic. In the event of any changes in the network, for example, a link capacity change, the updated topology information needs to be propagated to all the network nodes. We can use the same techniques used in IP networks for this purpose. Nodes periodically, or in the event of changes, flood the up- dated information to all the network nodes. We can use an Internet routing and topology management protocol such as OSPF or IS-IS (see Section 6.3), with suit- able modifications to represent optical layer topology information, and update it automatically. At the time the network is brought up, or whenever there is a topology change (link/node addition, removal), nodes will need to automatically discover the network topology. This is done typically by having adjacent nodes exchange information to determine their local connectivity (to their neighbors) and then broadcasting this information to all the network nodes using the same procedure used to convey topology changes. Route computation. When a connection is requested from the network, the network needs to find a route and obtain resources along the route to support this connec- tion. This can be done by applying a routing algorithm on the topology database 522 CONTROL AND MANAGEMENT of the network. The routing algorithm needs to take into account the various constraints imposed by the network, such as wavelength conversion ability, and the capacity available on each link of the network. We studied this aspect in Section 8.2.2. In addition to computing routes for carrying the working traffic, the algorithm may also have to compute protection routes for the connection, which are used in the event of failures. Signaling protocol. Once routes are computed, the connection needs to be set up. This process involves reserving the resources required for the connection and setting the actual switches inside the network to set up the connection. The process requires nodes to exchange messages with other nodes. Typically, the destination or source of the connection signals to each of the nodes along the connection path to perform this function. Protocols based on MPLS Internet signaling protocols such as RSVP or CR-LDP (see Section 6.3) can be used for this purpose. The same protocols can also be used to take down connections when they are no longer needed. The process of setting up or taking down a connection must be executed carefully. For example, if the connection is simply taken down by the source and destination, then the intermediate nodes may sense the loss of light on the connection as a failure condition and trigger unwanted alarms and protection switching. This can be avoided by suitable coordination among the nodes along the route of the lightpath. Signaling network. Nodes need a signaling channel to exchange control information with other nodes. We described the many options available to realize this in Section 9.5.5. Interaction with Other Layers One important aspect of the connection management protocols is in how they interact with the client layers of the optical layer. With IP routers emerging as the dominant clients of the optical layer, and because the optical layer control protocols are based on Internet protocols, the issue of how these protocols interact in particular with the IP layer becomes a crucial issue. Different types of interactions are likely needed for different scenarios, such as metro versus long-haul networks, incumbent versus new service providers, mul- tiservice versus IP service-centric providers, and facility ownership versus leasing providers. There are many schools of thought with respect to this interaction, ranging from the so-called overlay model to a peer model. Figure 9.9 shows a variety of models being considered today. 9.6 Configuration Management 523 Figure 9.9 Different control plane models for interconnecting client layers with the optical layer. (a) Overlay model, (b) overlay+ model, (c) peer model, and (d) augmented model. Figure 9.9(a) shows the overlay model. In this model, the optical layer has its own control plane, and the higher layers have their own independent control planes. The optical layer provides a user network interface (UNI), through which higher (client) layers can request connections from the optical layer. Within the optical layer, different subnetworks can interoperate through a standardized network-to-network interface (NNI). This approach allows the connection control software for the optical layer to be tailored specifically to the optical layer without having to worry about developing a single unified piece of control software. It also allows the optical layer and client layers to scale and evolve independently. Details of the optical network topology can be hidden from the client layer through the UNI. We can use this model to interconnect a variety of clients, including IP, ATM, Ethernet, and SONET/SDH clients, with the optical layer. The model is also appropriate for supporting private line lightpath service, transport bandwidth brokering, carrier's carrier trunking, and optical virtual private networks. Finally, this model can be applied to incumbent or new multiservice carriers who either own or lease their transport facilities. 524 CONTROL AND MANAGEMENT An enhanced version of the overlay model is the overlay plus model, shown in Figure 9.9(b), which allows closer interaction between the layers. In this case, there is a trusted intermediate intelligent controller between the two layers that has available to it a suitably abstracted version of specific client and optical layer topology and status information. The controller can use this information to request and release lightpaths based on specific policies, such as specific service level agreements made between the client and optical layers. These requests can be rapidly invoked to avoid network abnormalities such as congestion and failures, increase infrastructure utilization, coordinate protection and restoration options, and automate engineering by rebalancing the network and forecasting needed resource (such as node and link capacity) upgrades for both the IP and optical layers. Figure 9.9(c) shows the peer model, where IP routers and optical layer elements, such as OXCs and OADMs, run the same control plane software. This would allow routers to look at OXCs as if they were routers, effectively treating the IP layer and optical layer as peers. An OXC would simply be a special type of router, analogous to a label-switched router (LSR). Routers would have full topology awareness of the optical layer and could therefore control optical layer connections directly. While this is an elegant approach, it is made complicated by the fact that optical layer elements impose significantly different constraints with respect to routing and protection of connections, compared to the IP layer. In this case, we need to find a way to suitably abstract optical layer routing constraints into a form that can be used by route computation engines residing on IP routers. Figure 9.9(d) shows another enhanced version of the overlay model, called an augmented model, where the IP layer has access to summarized routing, addressing, and topology information of the optical layer, but still operates as a separate control plane from the optical layer. The models in Figure 9.9(c) and (d) tend to apply mainly to new IP-centric providers or IP-centric business units within established carriers who own their transport facilities. These models allow (or require) significantly more trust and closer coupling between the IP and optical layers, compared to the overlay models of Figure 9.9(a) and (b). All these models are being pursued today, but the overlay approach is likely to be the first one implemented. It has also been adopted for standardization by the ITU. 9.6.3 Adaptation Management Adaptation management is the function of taking the client signals and converting them to a form that can be used inside the optical layer. This function includes the following: 9.6 Configuration Management 525 9 Converting the signal to the appropriate wavelength, optical power level, and other optical parameters associated with the optical layer. This is done through the use of transponders, which convert the signal to electrical form and retransmit the signal using a WDM-specific laser. In the other direction, the WDM signal is received and converted into a standardized signal, such as a short-reach SONET signal. 9 Adding and removing appropriate overheads to enable the signal to be managed inside the optical layer. This could include one or more of the overhead techniques that we studied in Section 9.5.7. 9 Policing the client signal to make sure that the client signal stays within bound- aries that have been agreed upon as part of the service agreement. We discussed this in Section 9.5. The WDM network must support different types of interfaces to accommodate a variety of different users requiring different functions. Figure 9.10 shows the different possible adaptation interfaces. 1. Compliant wavelength interface: One interface might be to allow the client to send in light at a wavelength that is supported in the network. In this case, the user would be expected to comply with a variety of criteria set by the network, such as the signal wavelength, power, modulation type, and so on. These wavelengths may be regarded as compliant wavelengths. In this case, the interface might be a purely optical interface, with no optoelectronic conversions required (a significant cost savings). For example, you might envision that SONET or IP equipment must incorporate WDM-capable lasers at wavelengths suitable for the WDM network. Likewise, it would be possible to directly send a wavelength from the WDM network into SONET equipment. Here the user complies to the requirements imposed by the network. 2. Noncompliant wavelength interface: This is the most common interface and encompasses a variety of different types of attached client equipment that use optical transmitters and/or receivers not compatible with the signals used in- side the WDM network. For example, this would include SONET equipment using 1.3 #m lasers. Here until all-optical wavelength conversion (and perhaps all-optical regeneration) becomes feasible, optoelectronic conversion must be used, along with possibly regeneration, to convert the signal to a form suitable for the WDM network. This is likely to be the interface as well when we need to interconnect WDM equipment from different vendors adhering to different specifications, as we discussed in Section 9.4. 3. Subrate multiplexing: Additional adaptation functions include time division mul- tiplexing of lower-speed streams into a higher-speed stream within the WDM 526 CONTROL AND MANAGEMENT Figure 9.10 clients. Different types of interfaces between a WDM optical network and its equipment prior to transmission. For example, the WDM equipment could in- clude multiplexing of SONET OC-48 streams into OC-192 streams. This could reduce costs by eliminating the separate equipment that would normally be needed to perform this function. The level of transparency offered by the network also affects the type of adap- tation performed at the edges of the network. The network needs to be capable of transporting multiple bit rates. In general the optical path can be engineered to support signals up to a specified maximum bit rate. The adaptation devices and re- generators used within the network need to be capable of supporting a variety of bit rates as well. An important enabler for this purpose is a programmable clock data recovery chip that can be set to work at a variety of bit rates. The chips available today are capable of handling integral multiples of bit rates (for example, 155 Mb/s, 622 Mb/s, 1.25 Gb/s, and 2.5 Gb/s). They are also capable of handling a narrow range of bit rates around a mean value. For example, a single chip could deal with SONET OC-24 signals or with Gigabit Ethernet signals, which are both around 1.25 Gb/s but not exactly at the same rate. Finally, using a digital wrapper to en- capsulate the client signal allows the network to transport multiple data rates and protocol formats in a supervised way. 9.7 Optical Safety The semiconductor lasers used in optical communication systems are relatively low-power devices; nevertheless, their emissions can cause serious damage to the human eye, including permanent blindness and burns. The closer the laser wave- length is to the visible range, the more damage it can do, since the cornea is more 9.7 Optical Safety 527 transparent to these wavelengths. For this reason, systems with lasers must obey cer- tain safety standards. Systems with lasers are classified according to their emission levels, and the relevant classes for communication systems are described next. These safety issues in some cases can limit the allowable optical power used in the system. A Class I system cannot emit damaging radiation. The laser itself may be a high-power laser, but it is prevented from causing damage by enclosing it in a suitably interlocking enclosure. The maximum power limit in a fiber for a Class I system is about 10 mW (10 dBm) at 1.55 #m and 1 mW (0 dBm) at 1.3 #m. Moreover, the power must not exceed this level even under a single failure condition within the equipment. A typical home CD player, for example, is a Class I system. A Class IIIa system allows higher emission powersmup to 17 dBm in the 1.55 #m wavelength rangembut access must be restricted to trained service personnel. Class IIIa laser emissions are generally safe unless the laser beam is collected or focused onto the human eye. A Class IIIb system permits even higher emission powers, and the radiation can cause eye damage even if not focused or collected. Under normal operation, optical communication systems are completely "en- closed" systems~laser radiation is confined to within the system and not seen out- side. The problem arises during servicing or installation, or when there is a fiber cut, in which case the system is no longer completely enclosed and emission powers must be kept below the levels recommended for that particular system class. Communi- cation systems deployed in the enterprise world must generally conform to Class I standards since untrained users are likely to be using them. Systems deployed within carrier networks, on the other hand, may likely be Class IIIa systems, since access to these systems is typically restricted to trained service personnel. The safety issue thus limits the maximum power that can be launched into a fiber. For single-channel systems without optical power amplifiers using semiconductor lasers, the emission levels are small enough (-3 to 0 dBm typically) that we do not have to worry much about laser safety. However, with WDM systems, or with systems using optical power amplifiers, we must be careful to regulate the total power into the fiber at all times. Simple safety mechanisms use shuttered optical connectors on the network equip- ment. This takes care of regulating emissions if a connector is removed from the equipment, but cannot prevent emissions on a cut fiber further away from the equip- ment. This is taken care of by a variety of automatic shutdown mechanisms that are designed into the network equipment. These mechanisms detect open connec- tions and turn off lasers and/or optical amplifiers (the spontaneous emission from amplifiers may itself be large enough to cause damage). Several techniques are used to perform this function. If an amplifier senses a loss of signal at its input, it turns off its pump lasers to prevent any output downstream. There is some handshaking needed between the two ends of a failed link to handle unidirectional cuts. If one end 528 CONTROL AND MANAGEMENT Figure 9.11 Open fiber control protocol in the Fibre Channel standard. senses a loss of signal, it turns off its transmitter or amplifier in the other direction. This in turn allows the other end to detect a loss of signal and turn off its transmitter or amplifier. Another technique is to look at the back-reflected light. In the event of a fiber cut, the back-reflection increases and can be used to trigger a shutdown mechanism. After the failure is repaired, the system can be brought up manually. More sophisticated open fiber control mechanisms allow the link to be brought back up automatically once the failure is repaired. These mechanisms typically pulse the link periodically to determine if the link has been repaired. The pulse power is maintained below the levels specified for the safety class. Here, we describe a particular protocol that has been chosen for the Fibre Channel standard. 9.7.1 Open Fiber Control Protocol Figure 9.11 shows a block diagram of a system with two nodes A and B using the OFC protocol. Figure 9.12 shows the finite-state machine of the protocol. The protocol works as follows: 1. Under normal operating conditions, A and B are in the ACTIVE state. If the link from A to B fails, receiver B detects a loss of light and turns off laser B, and B enters the DISCONNECT state. Receiver A subsequently detects a loss of light and turns off its laser and also enters the DISCONNECT state. Similarly, if the link from B to A fails, or if both links fail simultaneously, A and B both enter the DISCONNECT state. 2. In the DISCONNECT state, A transmits a pulse of duration r every T seconds. B does the same. If A detects light while it is transmitting a pulse, it enters the STOP state and is called the master. If A detects light while it is not transmitting 9.7 Optical Safety 529 ACTIVE LOL I " DISCONNECTp I [.,Jr [~ Light detected within 1:~ No light detected/7 ~/ N~ I ][ RECONNECT] i LOL within "c' Light detected J'l STOP LOL = loss of light Figure 9.12 State machine run by each node for the open fiber control protocol in the Fibre Channel standard. a pulse, it transmits a pulse for r seconds and then enters the STOP state and is called the slave; likewise for B. 3. Upon entering the STOP state, the node turns off its laser for a period of ~' seconds. It remains in this state until a loss of light condition is detected on the incoming link. If this happens within the r ~ seconds, it moves into the RECONNECT state. Otherwise, it moves back into the DISCONNECT state. 4. Upon entering the RECONNECT state, if the node is the master, it sends out a pulse of duration r. If light is detected on the incoming link within this time period, the node enters the ACTIVE state. Otherwise, it shuts off its transmitter and enters the DISCONNECT state. If the node is the slave, it monitors the link for a period of ~ seconds, and if light is detected on the incoming link within this period, it turns on its laser and enters the ACTIVE state. Otherwise, it goes back to the DISCONNECT state. This is a fairly complex protocol. A simpler version of this protocol would not have the STOP and RECONNECT states. Instead, the nodes would directly enter the ACTIVE state from the DISCONNECT state upon detecting light. The reason for having the other states is to try to ensure that both nodes have functioning safety circuitry. If one of the nodes does not turn off its laser during the STOP period, it is assumed that the safety circuitry is not working and the other node goes back to the DISCONNECT state. In order for the protocol to work, ~, ~', and T must be chosen carefully. In the DISCONNECT state, the average power transmitted is "rP/T, where P is the transmitted power when the laser is turned on. This must be less than the allowed . arrayed multiwavelength components versus separate com- ponents for individual wavelengths, such as multiwavelength laser arrays instead of individual lasers for each wavelength. Using arrayed. to be capable of supporting a variety of bit rates as well. An important enabler for this purpose is a programmable clock data recovery chip that can be set to work at a variety of bit rates chips available today are capable of handling integral multiples of bit rates (for example, 155 Mb/s, 622 Mb/s, 1.25 Gb/s, and 2.5 Gb/s). They are also capable of handling a narrow range of