610 AccEss NETWORKS bandwidth. Moreover, this wavelength-routed PON can also support broadcast ser- vices efficiently using the spectral slicing technique described earlier. Thus there is an upgrade path starting from a broadcast network with shared bandwidth to a broadcast network with dedicated bandwidth and eventually to a switched network with dedicated bandwidth. Summary Service providers, both telephone operators and cable companies, are actively look- ing to deploy broadband access networks to provide a variety of new services. Fiber-based services are now available for many businesses in metropolitan ar- eas. When it comes to residential access, however, fiber is yet to reach the home. SONET/SDH ring-based architectures have been deployed to support the needs of large business customers, but they are not as suited for supporting the needs of res- idential users and small business customers. The two main architectures for broad- band access networks are the hybrid fiber coax (HFC) architecture, which is based on evolving the current plant deployed by cable television operators, and the fiber to the curb (FTTC) architecture, or equivalently a passive optical network (PON) architecture. Compared to the HFC approach, FTTC has a higher initial cost, but provides bandwidth deeper in the network and may prove to be a better longer-term solution. Although FTTC refers to a simple broadcast TDM star PON architecture, we also explored several upgrade options of the PON approach that provide higher capacities by making clever use of wavelength division multiplexing techniques. A number of major telephone carriers and manufacturers in the world have gotten to- gether and defined the requirements for an FTTC-based architecture to enable them to deploy the full service access network [FSA98]. As of this writing, PONs based on this architecture are just beginning to be deployed. FTTC is attractive in places where coaxial cable is not already deployed, which is the case in many countries other than the United States. FTTC also makes sense for telephone companies who lack a cable infrastructure. Variants of FTTC have been around a long time, but deployment has been slow for several reasons. First, there is significant cost associated with building and deploying a new access network, which can take several years to pay back. Therefore there is a big barrier toward making the investment in the first place. Second, this is coupled with the uncertain outlook in terms of the revenue that can be generated from the investment. Third, optical component costs are only now starting to decline, with the development of components especially optimized for PON applications, such as low-cost, uncooled semiconductor lasers and transceivers. The HFC approach, on the other hand, is attractive in places where coaxial cable is already deployed to the home, such as the United States. It is the logical evolution Problems 611 choice for cable companies who have already deployed a simpler version of the HFC architecture to provide basic cable television service. As optical component costs come down and bandwidth needs increase, it is clear that optical fiber will play a major role in access networks the question is how close will it get to our homes? Further Reading There is a vast body of literature on access networks, and several conferences have sessions devoted to it. There is an informative Web page maintained by the DSL forum (http'//www.adsl.com). See also [Bha99] for a nice overview of the different types of DSL. The papers in [Fra98, Aar95, Kob94, KKS00, SKY89] describe plans for deploy- ing fiber in the access network and compare different architectural approaches. TDM PONs were first proposed in [Ste87]. At this time, a group of telephone carriers and manufacturers have gotten together and agreed upon a standard for a fiber-based ac- cess architecture, called the full service access network (FSAN) initiative. Details may be found on the Web at www.fsanet.net and in [FSA98, Qua98]. The International Telecommunications Union (ITU) has published a requirements standard based on FSAN [ITU98]. [FRI96, FHJ+98, VMVQ00] describe some possible evolutions of the basic TPON architecture by making clever use of WDM and optical amplifiers. A variety of WDM PONs are described in [WKR+88, WL88, Fri94, ZJS+95, IFD95, IRF96]. 11.1 Problems Do a power budget calculation for the different types of PON architectures consid- ered in this chapter and determine the number of ONUs that can be supported in each case, assuming the following parameters: Laser output power LED output power Transmit bit rate Receiver sensitivity Fiber loss, including connectors 1 x 8 wavelength router loss 1 • 32 wavelength router loss 1 • 64 wavelength router loss Excess splitter loss -3 dBm -20 dBm 155 Mb/s -40 dBm 10 dB 5 dB 9 dB 12 dB 1 dB 612 ACCESS NETWORKS 11.2 - [Aar95] [ Bha 9 91 [FHJ+98] [Fra98] [Fri94] [FRI96] [FSA98] [IFDSS] The normal wavelength router losses are indicated above. However, with spectral slicing, an additional loss is also incurred as only a small fraction of the spectrum is transmitted out of each port on the wavelength router. Assume that in addition to the standard loss, we get only 1/2N of the transmitted power in each channel, where N is the number of ONUs. Consider the RITENET architecture shown in Figure 11.10. Suppose the laser speed at the CO is limited to 155 Mb/s. The network needs to support 20 ONUs and provide each ONU with 10 Mb/s bandwidth from the CO to the ONU and 2 Mb/s from the ONU to the CO. How could you modify the architecture to support this requirement? References R. Aaron, editor. IEEE Communications Magazine: Special Issue on Access to Broadband Services, volume 33, Aug. 1995. V. K. Bhagavath. Emerging high-speed xDSL access services: Architectures, issues, insights, and implications. IEEE Communications Magazine, 37( 11):106-114, Nov. 1999. R. D. Feldman, E. E. Harstead, S. Jiang, T. H. Wood, and M. Zirngibl. An evaluation of architectures incorporating wavelength division multiplexing for broad-band fiber access. IEEEIOSA Journal on Lightwave Technology, 16( 9):1546-1559, Sept. 1998. P. W. France, editor. BT Technology Journal-Special Issue on Local Access Technologies, volume 16, Oct. 1998. N. J. Frigo et al. A wavelength-division-multiplexed passive optical network with cost-shared components. IEEE Photonics Technology Letters, 6( 11):1365-1367, 1994. N. J. Frigo, K. C. Reichmann, and P. P. Iannone. WDM passive optical networks: A robust and flexible infrastructure for local access. In Proceedings of International Workshop on Photonic Networks and Technologies, pages 201-212, 1996. Full Services Access Network Requirements Specification, 1998. Available on the Web at www.fsanet.net. P. P. Iannone, N. J. Frigo, and T. E. Darcie. WDM passive optical network architecture with bidirectional optical spectral slicing. In OFC'9.5 Technical Digest, Dazes 51-53. 1995. PaDer TuK2. I" References 613 [IRF96] P.P. Iannone, K. C. Reichmann, and N. J. Frigo. Broadcast digital video delivered over WDM passive optical networks. IEEE Photonics Technology Letters, 8(7):930-932, 1996. [ITU98] ITU-T. Recommendation G.983" Broadband Optical Access Systems Based on Passive Optical Networks, 1998. [KKS00] D. Kettler, H. Kafka, and D. Spears. Driving fiber to the home. IEEE Communications Magazine, 38(11 ):106-110, Nov. 2000. [Kob94] I. Kobayashi, editor. IEEE Communications Magazine: Special Issue on Fiber-Optic Subscriber Loops, volume 32, Feb. 1994. [Qua98] J.A. Quayle et al. Achieving global consensus on the strategic broadband access network the full service access initiative. BT Technology Journal, 16(4):58-70, Oct. 1998. [SKY89] P.W. Shumate, O. Krumpholz, and K. Yamaguchi, editors. IEEE/OSA JLT/JSAC Special Issue on Subscriber Loop Technology, volume 7, Nov. 1989. [Ste87] J. Stern et al. Passive optical local networks for telephony applications. Electronics Letters, 23:1255-1257, 1987. [VMVQ00] I. Van de Voorde, C. M. Martin, J. Vandewege, and X. Z. Qiu. The superPON demonstrator: An exploration of possible evolution paths for optical access networks. IEEE Communications Magazine, 38(2):74-82, Feb. 2000. [WKR+88] S.S. Wagner, H. Kobrinski, T. J. Robe, H. L. Lemberg, and L. S. Smoot. Experimental demonstration of a passive optical subscriber loop architecture. Electronics Letters, 24:344-346, 1988. [WL88] S.S. Wagner and H. L. Lemberg. Technology and system issues for the WDM-based fiber loop architecture. IEEE/OSA Journal on Lightwave Technology, 7(11):1759-1768, 1988. [ZJS+95] M. Zirngibl, C. H. Joyner, L. W. Stulz, C. Dragone, H. M. Presby, and I. P. Kaminow. LARnet, a local access router network. IEEE Photonics Technology Letters, 7(2):1041-1135, Feb. 1995. This Page Intentionally Left Blank Photonic Packet Switching I N THIS CHAPTER, we study optical networks that are capable of providing packet-switched service at the optical layer. We call these networks photonic packet-switched (PPS) networks. Packet-switched services are provided today us- ing electronic switches by many networks, such as IP and ATM networks. Here, we are interested in networks where the packet-switching functions are performed optically. The goal of PPS networks is to provide the same services that electronic packet-switched networks provide, but at much higher speeds. The optical networks that we have studied so far provide circuit-switched services. These networks provide lightpaths, which can be established and taken down as needed. In these networks, the optical nodes do not switch signals on a packet-by-packet basis, but rather only switch at the time a circuit is established or taken down. Packet switching is done in the electronic domain by other equipment such as IP routers or ATM switches. These routers and ATM switches make use of lightpaths provided by the optical layer to establish links between themselves as needed. In addition to switching packets, routers and ATM switches make use of sophisticated software and hardware to perform the control functions needed in a packet-switched network. We will see in this chapter that all the building blocks needed for optical packet switching are in a fairly rudimentary state today and exist only in research laboratories~they are either difficult to realize, very bulky, or very expensive, even after a decade of research in this area. Moreover, it is likely that we will need elec- tronics to perform the intelligent control functions for the foreseeable future. Optics can be used to switch the data through, but it does not yet have the computing 615 616 PHOTONIC PACKET SWITCHING capabilities to perform many of the control functions required, such as processing the packet header, determining the route for the packet, prioritizing packets based on class of service, maintaining topology information, and so on. However, there are a few motivations for researching optical packet switching. One is that optical packet switches hold the potential for realizing higher capacities than electronic routers (although this potential is yet to be demonstrated!). For in- stance, the capacity of the best routers today is less than i Tb/s, with the highest-speed interfaces being at 10 Gb/s. In contrast, optical switches are, for the most part, bit rate independent, so they can be used to switch tens to hundreds of Tb/s of traffic. At line rates of 80 Gb/s and beyond, electronic time division multiplexing (TDM) appears to be running out of gas, and optical time division multiplexing may be the way to go. Another motivation for studying optical packet switching is that it can improve the bandwidth utilization within the optical layer. The notion is that high-speed optical links between routers are still underutilized due to the bursty nature of traffic, and using an underlying optical packet layer instead of an optical circuit layer will help improve link utilizations. The question is whether having another high-speed packet-switched layer under an already existing packet-switched layer (say, IP) will provide sufficient improvement in statistical link utilization. The answer depends on the statistical properties of the traffic. The conventionally accepted wisdom is that as many lower-speed bursty traffic streams are multiplexed through many layers, the burstiness of the aggregate stream is lower than that of the individual streams. In this case, having an optical packet layer under an electrical packet layer may not help much because the traffic entering the optical layer is already smoothed out. However, it has been shown recently that with some types of bursty traffic, notably the so-called self-similar traffic, that the burstiness of a multiplexed stream is not less than that of its constituent individual streams [PF95, ENW96]. For such traffic, using an optical packet layer provides the potential to improve the link utilization. Figure 12.1 shows a generic example of a store-and-forward packet-switched network. In this network, the nodes A-F are the switching/routing nodes; the end nodes 1-6 are the sources and sinks of packet data. We will assume that all packets are of fixed length. Packets sent by an end node will, in general, traverse multiple links and hence multiple routing nodes, before they reach their destination end node. For example, if node 1 has to send a packet to node 6, there are several possible routes that it can take, all consisting of multiple links and routing nodes. If the route chosen for this packet is 1-A-B-D-F-6, this packet traverses the links I-A, A-B, B-D, D-F, and F-6. The routing nodes traversed are A, B, D, and F. Note that the route chosen may be specified by the packet itself, or the packet may simply specify only the destination node and leave the choice of route to the routing nodes in its path. In the remainder of the discussion, we will assume that the route is chosen Photonic Packet Switching 617 Figure 12.1 A generic store-and-forward network. Figure 12.2 A routing node in the network of Figure 12.1. by the routing nodes based on the packet destination that is carried in the packet header. Figure 12.1 is also the block diagram of a PPS network. The major difference is that the links run at very high speeds (hundreds of gigabits per second) and the signals are handled mostly optically within each routing node. Figure 12.2 shows a block diagram depicting many of the functions of a routing node, or router. In general, there is one input from, and one output to, each other 618 PHOTONIC PACKET SWITCHING routing node and end node that this routing node is connected to by a link. For example, in Figure 12.1, routing node A has three inputs and outputs: from/to routing node B, routing node C, and end node 1. Similarly, routing node C has five inputs and outputs. Routers perform the following functions (see Section 6.3 for a more detailed description of how these functions are performed by IP routers): Routing. Routers maintain up-to-date information of the network topology. This information is maintained in the form of a routing table stored at each node. Forwarding. For each incoming packet, a router processes the packet header and looks up its routing table to determine the output port for that packet. It may also make some changes to the header itself and reinsert the header at the output. For example, we studied in Chapter 6 that an ATM switch determines the virtual circuit identifier (VCI) in the header of the incoming cell and looks it up in its locally stored VCI table, which provides the corresponding output link and outgoing VCI. It then inserts the outgoing VCI into the cell header. Switching. Switching is the actual process of switching the incoming packet to the appropriate output port determined by the forwarding process. In the electronic domain, the forwarding and switching functions are usually treated together as a single function, but it will prove to be useful to separate them in PPS networks. Buffering. There are many reasons why buffering is needed in a router. Perhaps the most important one in this context is to deal with destination conflicts. Multiple packets arrive simultaneously at different inputs of a router. Several of these may have to be switched to the same output port. However, at any given time, only one packet can be switched to any given output port. Thus the router will have to buffer the other packets until they get their turn. Buffers are also used to separate packets based on their priorities or class of service. Figure 12.2 shows buffers at the input as well as the output. We will explore the trade-offs between input and output buffering in Section 12.4. We will see that buffers are difficult to realize in the case of photonic packet switches, and most switch proposals therefore use only a small amount of buffering, usually integrated with the switch. Multiplexing. Routers multiplex many lower-speed streams into a higher-speed stream. They also perform the reverse demultiplexing operation. Synchronization. Synchronization can be broadly defined as the process of aligning two signal streams in time. In PPS networks, it refers either to the alignment of an incoming pulse stream and a locally available clock pulse stream or to the relative alignment of two incoming pulse streams. The first situation occurs during multiplexing and demultiplexing, and the second occurs at the inputs of 12.1 Optical Time Division Multiplexing 619 the router where the different packet streams need to be aligned to obtain good switching performance. PPS networks will have to perform all the functions described above. Some of these functions involve a fair amount of sophisticated logic and processing and are still best handled in the electrical domain. The routing and forwarding functions, in particular, fit into this category. Most PPS proposals to date assume that the packet header is transmitted separately from the data at a lower speed and process the header electronically. We will, however, study some of the approaches toward providing at least rudimentary header processing in the optical domain. Due to technological constraints, it is quite difficult to perform even the re- maining functions of switching, buffering, multiplexing, and synchronization in the optical domain. This will become clearer as we explore the different techniques for performing these functions. Therefore, PPS networks are at this time still in research laboratories and have not yet entered the commercial marketplace. To simplify the implementation, especially the control functions, most PPS proposals also assume the use of fixed-size packets, and we will make the same assumption in this chapter. Of course in reality we have to deal with varying packet sizes. If a fixed packet size is used inside the network, then the longer packets will have to be segmented at the network inputs and reassembled together at the end. Alternatively, we could design the PPS nodes to switch variable-sized packets, a more complex proposition. The outline of this chapter is as follows. We start by describing techniques for multiplexing and demultiplexing optical signals in the time domain, followed by methods of doing synchronization in the optical domain. Synchronization requires delaying one stream with respect to the other if they are misaligned in time. In this context, we will also study how tunable optical delays can be realized. We then discuss various solutions for dealing with the buffering problem. We conclude the chapter by discussing burst switching, a variant of PPS, and some of the experimental work that has been carried out to demonstrate the various aspects of PPS. 12.1 Optical Time Division Multiplexing At the inputs to the network, lower-speed data streams are multiplexed optically into a higher-speed stream, and at the outputs of the network, the lower-speed streams must be extracted from the higher-speed stream optically by means of a demultiplex- ing function. Functionally, optical TDM (OTDM) is identical to electronic TDM. The only difference is that the multiplexing and demultiplexing operations are per- formed entirely optically at high speeds. The typical aggregate rate in OTDM systems is on the order of 100 Gb/s, as we will see in Section 12.6. . packet is 1 -A- B-D-F-6, this packet traverses the links I -A, A- B, B-D, D-F, and F-6. The routing nodes traversed are A, B, D, and F. Note that the route chosen may be specified by the packet. through many layers, the burstiness of the aggregate stream is lower than that of the individual streams. In this case, having an optical packet layer under an electrical packet layer may not. [Ste87]. At this time, a group of telephone carriers and manufacturers have gotten together and agreed upon a standard for a fiber-based ac- cess architecture, called the full service access