1.6 WDM OPTICAL NETWORK ARCHITECTURES
There are three classes of WDM optical network architectures: broadcast-and-select networks, wavelength routed networks, and linear lightwave networks. We now explain each of these networks in detail.
1.6.1 Broadcast-and-Select Networks
A broadcast-and-select network consists of a passive star coupler connecting the nodes in the network as shown in Fig. 1.10(a). Each node is equipped with one or more fixed-tuned or tunable optical transmitters and one or more fixed-tuned or tunable optical receivers. Different nodes transmit messages on different wavelengths simultaneously. The star coupler combines all these messages and then broadcasts the combined message to all the nodes. A node selects a desired wavelength to receive the desired message by tuning its receiver to that wavelength. Note that the star coupler offers an optical equivalent to radio systems: each transmitter broadcasts its signal or message on a different wavelength and the receivers are
Figure 1.10 (a) Broadcast-and-select network. (b) Logical topology.
tuned to receive the desired signal. An N × N star coupler can be realized using a multistage interconnection network which has log2 N stages of 2 × 2 couplers with N/2 couplers per stage (assuming N is a power of 2) or directly in integrated optics form with a common coupling region. Integrated optics refers to integration of optical components with fiber interconnections onto a single optical substrate, similar to the way in which electrical components (such as resistors, capacitors, and inductors) are combined in an electronic integrated circuit.
In single-hop broadcast-and-select networks, a message, once transmitted as light, reaches its final destination directly, without being converted to electronic form in between. In order to support packet switching in these networks, we need to have optical transmitters and receivers that can tune rapidly. This is because, in a packet-switched network, a node must be able to transmit (receive) successive packets to (from) different nodes on different wavelengths. The main networking challenge in these networks is the coordination of transmissions between various nodes. In the absence of coordination or efficient medium access control (MAC) protocol, collisions occur when two or more nodes transmit on the same wavelength at the same time. Also, destination conflicts occur if two or more nodes transmit on different wavelengths to the same destination when the destination has only one tunable optical receiver. Moreover, the destination must know when to tune to the appropriate wavelength to receive a packet. Several MAC protocols have been proposed to prevent such collisions/conflicts for single-hop broadcast-and-select networks, assuming the availability of rapidly tunable transmitters and/or receivers . (See Problem 1.10.) To support packet switching efficiently in broadcast-and-select networks, a multihop approach, which avoids rapid tuning altogether, can be used. Each node has a small number of fixed-tuned optical transmitters and fixed-tuned optical receivers. Each transmitter is at a different wavelength. We can represent the network as a graph, wherein a node corresponds to a network node and an edge corresponds to a transmitterreceiver pair on the same wavelength. Thus we obtain a virtual or logical topology over the physical broadcast topology. Figure 1.10(a) shows a four-node broadcast-and-select network. Each node transmits at one fixed wavelength and receives on one fixed wavelength. For example, node 0 can transmit directly to node 1 using wavelength w0, but not to node 2. To transmit to node 2, node 0 sends a packet to node 1 on wavelength w0, which receives it, converts it to electronic
form, and retransmits it on wavelength w1. The packet then reaches node 2. The virtual topology of the network in Fig. 1.10(a) is shown in Fig. 1.10(b). In these networks, a packet may have to go through more than one hop before reaching its destination. This leads to an increase in propagation delay in addition to queueing delays at intermediate nodes, and wastage of network capacity.
The advantage of broadcast-and-select networks is in their simplicity and natural multicasting capability (ability to transmit a message to multiple destinations). However, they have severe limitations. (1) They require a large number of wavelengths, typically at least as many as there are nodes in the network, because there is no wavelength reuse in the network. Thus the networks are not scalable beyond the number of supported wavelengths. (2) They cannot span long distances since the transmitted power is split among various nodes and each node receives only a small fraction of the transmitted power, which becomes smaller as the number of nodes increases. For these reasons, the main application for broadcast-and-select is high-speed local area networks (LANs) and metropolitan area networks (MANs).
Broadcast-and-Select Network Demonstrators
A number of demonstrators based on the broadcast-and-select approach have been built . Bellcore's LAMBDANET (late 1980s and early 1990s), which employed 18 wavelengths separated by 2 nm, was one of the first demonstrators. Transmission at 1.5 Gb/s per wavelength over 57 km was also demonstrated. Nippon Telegraph and Telephone Corporation's (NTT's) test bed in the early 1990s used 100 wavelengths, spaced 10 GHz apart, each carrying data at 622 Mb/s. International Business Machines Corporation's (IBM's) RAINBOW-I and RAINBOW-II test beds were designed to support 32 wavelengths separated by 1 nm, in a star configuration. RAINBOW-I interconnected computers; after a connection was set up, transmission at 300 Mb/s on each wavelength was possible. In RAINBOW-II, an extension of RAINBOW-I, the transmission rate per wavelength was upgraded to 1 Gb/s, and the control protocol processing was improved. STARNET-I and II WDM computer networks, developed at Stanford University, resemble RAINBOW. STARNET-I had two data rates at each node, both transmitted on a unique wavelength using a single laser. These rates were a fast 2.5 Gb/s for establishing circuit-switched connections with other nodes as in RAINBOW and 100 Mb/s for establishing a logical ring network. In STARTNET-II, the high data rate was modified to 1.25 Gb/s. The European Research and Development for Advanced Communications in Europe (RACE) program developed a star network and used it for video applications in a BBC television studio test bed. Each node in the network, called a local routing center (LRC), interconnected up to 16 (electrical signal) sources and destinations operating at 155 Mb/s each. Up to 16 such LRCs were then interconnected using a star coupler. LIGHTNING is another test bed that is a tree hierarchical WDM network used for interconnecting processors which form the leaves of the tree. Another example is the supercomputer supernet, developed jointly by the University of California at Los Angeles, Jet Propulsion Laboratory, and the Aerospace Corporation, for connecting supercomputers separated geographically by several kilometers.
1.6.2 Wavelength Routed Networks
Wavelength routed WDM networks have the potential to avoid the three problemslack of wavelength reuse, power splitting loss, and scalability to wide area networks (WANs)of broadcast-and-select networks. A wavelength routed network consists of WXCs (routing nodes) interconnected by point-to-point fiber links in an arbitrary topology. Each end node (end user) is connected to a WXC via a fiber link. The combination of end node and its corresponding WXC is referred to as a (network) node. Each node is equipped with a set of transmitters and receivers, for sending data into the network and receiving data from the network, respectively, both of which may be wavelength-tunable.
In a wavelength routed network, a message is sent from one node to another node using a wavelength continuous route called a lightpath, without requiring any opticalelectronicoptical conversion and buffering at the intermediate nodes. This process is known as wavelength routing. Note that the intermediate nodes route the lightpath in the optical domain using their WXCs. The end nodes of the lightpath access the lightpath using transmitters/receivers that are tuned to the wavelength on which the lightpath operates. A lightpath is an all-optical communication path between two nodes, established by allocating the same wavelength throughout the route of the transmitted data. Thus it is a high-bandwidth pipe, carrying data up to several gigabits per second, and is uniquely identified by a physical path and a wavelength. The requirement that the same wavelength must be used on all the links along the selected route is known as the wavelength continuity constraint. Two lightpaths cannot be assigned the same wavelength on any fiber. This requirement is known as distinct wavelength assignment constraint. However, two lightpaths can use the same wavelength if they use disjoint sets of links. This property is known as wavelength reuse.
Example 1: Consider a wavelength routed network with five nodes and two wavelengths per fiber as shown in Fig. 1.11. Assume that lightpaths are to be established one for each of the node pairs < 0, 2 >, < 1, 3 >, < 2, 4 >, < 3, 0 >, and < 4, 1 >. Further assume that every node is equipped with one transmitter and one receiver. For the given set of node pairs, a node is a source for one lightpath and destination for one lightpath. There exists only one physical path between any node pair. Every fiber link in the network would carry physical paths corresponding to two lightpaths, if lightpaths were successfully established for all the given node pairs. Since two wavelengths are available on any fiber link, it should be possible to route all the lightpaths. However, due to the wavelength continuity constraint it is not possible to establish lightpaths for all five node pairs. The figure shows a possible way of routing four lightpaths p0, p1, p2, and p3, where pi is the lightpath emanating from node i. Since p0 uses wavelength w0, p1 can use only w1, as p0 and p1 share a link. Lightpath p2 can use only w0, as p1 and p2 share a link. Lightpath p3 can use only w1, as p2 and p3 share a link. As a consequence, wavelength w0 is free on link 4 → 0 and w1 is free on link 0 → 1. Therefore, a lightpath cannot be established from node 4 to node 1 even though bandwidth (wavelength) is available on links 4→ 0 and 0 → 1, a transmitter is available at node 4, and a receiver is available at node 1. As we will see later, this bandwidth loss caused by the wavelength continuity constraint can be overcome by using a wavelength converter. However, observe that both the wavelengths w0 and w1 are reused two times, thus helping increase the number of lightpaths established while employing a limited number of wavelengths.
Figure 1.11 (a) Wavelength routed network. (b) Logical topology.
Wavelength reuse (an important feature that refers to simultaneous transmission of messages on the same wavelength over fiber-link-disjoint lightpaths) in wavelength routed networks makes them more scalable than broadcast-and-select networks. Another important characteristic which enables wavelength routed networks to span long distances is that the transmitted power invested in the lightpath is not split to irrelevant destinations. Given a WDM network, the problem of routing and assigning wavelengths to lightpaths is of paramount importance in these networks, and clever algorithms are needed in order to ensure this function (routing and wavelength assignment) is performed using a minimum number of wavelengths. The number of available wavelengths in a fiber link plays a major role, in these networks, which currently varies between 4 and 32, but is expected to increase (with announcements of over 100 wavelengths already made).
Packet switching in wavelength routed networks can be supported by using either a single-hop or a multi-hop approach, in a way similar to broadcast-and-select networks. In the multi-hop approach, a virtual topology (a set of lightpaths or optical layer) is imposed over the physical topology (which is not broadcast here) by setting the WXCs in the nodes. Over this virtual topology, a packet from one node may have to be routed through some intermediate nodes before reaching its final destination. At each intermediate node, the packet is converted to electronic form and retransmitted on another wavelength. A virtual topology, formed by lightpaths p0 through p3, corresponding to the physical network shown in Fig. 1.11(a), is given in Fig. 1.11(b). A packet from node 3 (source) is routed through intermediate node 0 undergoing opticalelectronicoptical conversion at this node before reaching node 2 (destination).
Existing Internet backbone networks consist of high-capacity IP (Internet Protocoldeveloped for providing connectionless transfer of packets across an internetwork) routers interconnected by point-to-point fiber links. Traffic is transported between routers through high-speed gigabit links. These links are realized by SONET or ATM-over-SONET technology. The backbone routers use IP-over-SONET or IP-over-ATM-over-SONET technology to route IP tra3c in the backbone network. Most of the SONET-based backbone transport networks provide data interface at the rate of OC-3 and OC-12. The traffic demand is growing at a faster rate and a point has been reached where data interfaces at the rate of OC-48 and more are required. Upgrading the existing SONET transport infrastructures to handle these high-capacity interface rates is not desirable, as it is impractical to go for upgrading every time the interface rate increases. Also, such upgrading is not economical. A viable and cost-effective solution is to use WDM technology in backbone transport networks. In such (for example, IP-over-WDM) networks, network nodes are interconnected by WDM fiber links (where each link is capable of carrying multiple signals simultaneously, each on a different wavelength), and the nodes employ WXCs and electronic processing elements. Figure 1.12 shows a typical WDM backbone network. The electronic processing element can be an IP router, ATM switch, or a SONET system.
Figure 1.12 A WDM backbone network.
Any two IP routers in this network can be connected together by a lightpath. Two nodes that are not connected directly by a lightpath communicate using multihop approach, i.e., by using electronic packet switching at the intermediate nodes. This electronic packet switching can be provided by IP routers, ATM switches, or SONET equipment, leading to an IP-over-WDM or an ATM-over-WDM, or a SONET-over-WDM network, respectively.
A WDM-based transport network can be decomposed broadly into three layers, a physical media layer, an optical layer, and a client layer, as shown in Fig. 1.13. Application of WDM technology has introduced the optical layer between the lower
Figure 1.13 Possible layers in a WDM optical transport network.
physical media layer and upper client layer. A set of lightpaths constitutes the optical layer (virtual topology). The optical layer provides client-independent or protocol-transparent circuit-switched service to a variety of clients that constitute the client layer. This is possible because the lightpaths can carry messages at a variety of bit rates and protocols. Thus the optical layer can support a variety of clients concurrently. For example, some lightpaths could carry SONET data, whereas others could carry IP packets/datagrams or ATM cells. A network with an optical layer can be configured such that in the event of failures, lightpaths can be rerouted over alternate paths automatically. This provides a high degree of reliability in the network. According to International Telecommunications UnionTelecommunication Standardization Sector (ITU-T) Recommendation G.872, an optical layer can be further decomposed into three sublayers: an optical channel layer, an optical multiplex section layer, and an optical transmission section layer. The functionality of the optical channel layer is to provide end-to-end networking of optical channels (lightpaths) for transparently conveying the client data. The optical multiplex section layer concerns networking of aggregate multiwavelength optical signals. The optical transmission section layer concerns the transmission of optical signals on different kinds of optical media such as single-mode and multimode transmission.
These attractive featureswavelength reuse, protocol transparency, and reliabilitymake wavelength routed networks suitable for WANs. This book concentrates on wavelength routed (wide area) optical networks, mainly addressing the key issuesdesign, reconfiguration, and failure recovery in the optical layer, and architectures and technologies which will enable the realization of wavelength routed future optical Internet networks to transport high-speed IP traffic. Designing an optical layer to meet the traffic demand is an important problem in order to use the wavelength and fiber resources efficiently and to provide quality service to the users. Reconfiguring the optical layer is necessitated by the changing traffic demand. Since a huge amount of traffic is carried by the optical layer, rapid service recovery in case of network component failures is critically important.
Wavelength Routed Network Demonstrators
One of the most comprehensive field deployment trials is the Defense Advanced Research Projects Agency (DARPA)-sponsored Multiwavelength Optical Network (MONET) program. It is a wavelength routed test bed that uses eight wavelengths spaced 200 GHz apart, with a data transmission rate of 2.5 Gb/s per wavelength. The field trial consists of a local ring network with several WADMs in it and a WXC to interconnect the local ring network to a long-distance network. The program has demonstrated 10-Gb/s data transmission over 2,000 km. The All-Optical Network (AON) test bed, deployed in Boston between MIT's Lincoln Laboratory and Digital Equipment Corporation, uses 20 wavelengths spaced by 50 GHz. Each node in the test bed has a tunable transmitter (distributed Bragg grating laser with 10 ns [1 ns is 10-9 second] tuning time) and a tunable filter. NTT's test bed consists of a unidirectional ring with a central hub node and many access nodes. It uses six wavelengths spaced by 100 GHz with a distance of 40 km between nodes at 622 Mb/s, with a hub node and two access nodes. The ring employs a failure protection mechanism. Another ring architecture was demonstrated by the Optical Networks Technology Consortium (ONTC). The ONTC test bed consists of two unidirectional rings, with two access nodes per ring, connected by a 2 x 2 WXC. It has four wavelengths spaced by 4 nm, with a data transmission rate of 155 Mb/s over a total distance of 150 km. MultiWavelength Transport Network (MWTN) was one of the first wavelength routed test beds, developed by the European RACE program, to demonstrate successfully the concept of wavelength routing in optical networks. The test bed uses four wavelengths spaced 4 nm apart, with each wavelength carrying SDH data at 622 Mb/s or 2.5 Gb/s in a Stockholm field trial.
1.6.3 Linear Lightwave Networks
A usable portion of the optical spectrum (for example, the 1.55-micron band) can be partitioned into a number of either wavelengths or wavebands as shown in Fig. 1.14 . Observe that in Fig. 1.14(b) each waveband is further subdivided into a number of wavelengths. Note that sufficient spacing or guard bands have to be placed between any two wavelengths to allow for imprecision and drift in laser transmitter tuning and to make it possible to separate adjacent signals at the receivers. Wavelength routed networks use wavelength (one-level) partitioning and in these networks several wavelengths are multiplexed on a fiber link. Linear lightwave networks, on the other hand, use waveband (two-level) partitioning, and in these networks several wavebands are multiplexed on a fiber and several wavelengths are multiplexed on each waveband. In a wavelength routed network, routing nodes demultiplex, switch, and multiplex wavelengths, whereas in a linear lightwave network, routing nodes demultiplex, switch, and multiplex wavebands, but not wavelengths within a waveband. Thus the hardware requirements at the nodes, by grouping a set of wavelengths into a waveband, in linear lightwave networks get simplified because the number of optical switches required in a node is equal to the number of wavebands, not the number of wavelengths. Since a linear lightwave network as a whole does not distinguish between wavelengths within a waveband, individual wavelengths within a waveband are separated from each other at the end nodes (optical receivers).
Figure 1.14 (a) Wavelength partitioning. (b) Waveband partitioning.
Two constraintswavelength continuity and distinct wavelength assignmenton optical connections applicable to wavelength routed networks also apply to linear lightwave networks. Further, there are two routing constraints unique to linear lightwave networks: inseparability, that is, channels belonging to the same waveband when combined on a single fiber cannot be separated within the network; and distinct source combining, that is, on any fiber, only signals from distinct sources are allowed to be combined.
Figure 1.15 illustrates the inseparability constraint. Here nodes 0 through 5 are end nodes, while nodes A through H are routing nodes. The figure also shows two
Figure 1.15 Linear lightwave network.
connections, one between nodes 0 and 2, and the other between nodes 1 and 4. The notation < x, y > is used to denote a connection from node x to node y. < 0, 2 > passes through nodes A, B, E, F , and H, while < 1, 4 > passes through nodes A, B, D, and G. Since the two connections share the fiber A → B, they have to use different wavelengths (because of the distinct wavelength assignment constraint). Suppose wavelength w0 is used for < 0, 2 > and wavelength w1 for < 1, 4 >. The two signals, though on different wavelengths, may be in the same waveband. The two signals (that is, the power from both sources, nodes 0 and 1) are combined at node A. At node B, however, the two signals cannot be separated since they belong to the same waveband. Therefore, to route the two connections to their destinations, the combined signal power is split equally between the output ports of node B, one leading to node E and the other to routing node D. Only the end nodes (destinations), 2 and 4, filter out w0 and w1, respectively, rejecting the other wavelength. Note that some unintended destinations appearnode 2 in the case of connection < 1, 4 >, and node 4 in the case of connection < 0, 2 >. These unintended destinations are called fortuitous destinations; they tend to waste fiber resources and power, and are therefore to be dispensed with, if possible. For example, in this case the fortuitous destinations could have been avoided by rerouting connection < 1, 4 > on the path 1ACDG4.
The inseparability constraint can lead to a color clash. Suppose connections < 0, 2 > and < 1, 4 > routed on paths 0ABEF H2 and 1AC-DG4, respectively, use the same wavelength w0. Now a third connection < 5, 3 >, routed on the path 5CDGF H3 using the same waveband is assigned wavelength w1, which is different from w0 because connections < 1, 4 > and < 5, 3 > share the fiber C → D. Note that, due to inseparability, all three signals (generated at sources 0, 1, and 5) are carried on fiber F → H. Because signals generated at sources 0 and 1 are now on the same fiber F → H, using the same wavelength w0, they produce a color clash.
Distinct Source Combining
Distinct source combining disallows an optical signal from splitting at a node, taking multiple paths in the network, and then recombining with itself. We now illustrate how a correct but poor routing algorithm can violate the distinct source combining constraint. Consider the network shown in Fig. 1.15. Suppose that when two connections 0ABEF H2 and 1ABDG4, on wavelengths w0 and w1, respectively, are in progress, a new connection < 5, 3 > is routed along the path 5 CDGF H3 using a wavelength w2. All three connections are assumed to be in the same waveband. Due to inseparability, signal generated at source 5 carries with it (fortuitously) portions of signals generated at sources 0 and 1 after combining with them on fiber D→ G. This causes the signal generated at source 0 (which split at node B) to recombine with itself on fiber F → H, violating the distinct source combining constraint. Thus the new connection < 5, 3 > must not be routed along the path 5CDGF H3. However, this problem could have easily been avoided had the choice of routes been more prudent. Routing connection < 1, 4 > via A, C, D, and G, or routing connection < 5, 3 > via nodes C, D, B, E, F , and H would have made it possible for all three connections to be routed successfully.
Setting up connections with the above routing constraints in a linear lightwave network is significantly more complicated when wavebands contain more than one wavelength. Apart from this, the combining and splitting losses present at the nodes is preventing linear lightwave networks from becoming practical. Since at each node the power at an output port is a linear combination of the powers at the input ports, these networks are called linear lightwave networks.