Optically Transparent Integrated Metro-Access Network

Transcription

1 Springer (2011). This is an author s copy of the work. It is posted here by permission of Springer for your personal use.not for redistribution. The definitive version is published in Telecommunication Systems Journal in The final publication is available at DOI /s ly Transparent Integrated Metro-Access Network Slaviša Aleksić Vienna University of Technology, Vienna University of Technology, Institute of Telecommunications, Favoritenstr. 9/389, 1040 Vienna, Austria Phone: ; Fax: ; slavisa.aleksic@tuwien.ac.at Abstract High-capacity optical transmission technologies have made possible very high data rates and a large number of wavelength channels. Further, optical network functionality has made progress from simple point-to-point WDM links to automatically switched optical networks. In the future, dynamic burst-switched and packet-switched photonic networks may be expected. This paper describes a novel architecture of optically-transparent integrated WDM metropolitan area network (MAN) that is capable of switching on both packet-by-packet and burst-by-burst basis, thereby having the potential to achieve high throughput efficiency. The optically transparent MAN also includes a large part of the access network infrastructure. It is scalable, flexible, easy upgradeable and able to support heterogeneous network traffic. Some results of a preliminary study on network performance are shown. Keywords networks, metropolitan area networks, network and node architecture, media access control protocol, performance evaluation. 1 Introduction Nowadays, nobody can imagine a wide area, a regional or a metropolitan area network without optical transmission systems. networks make use of wavelength-division multiplexing (WDM) technology and optical amplifiers to transmit a large number of wavelength channels over optical fibers. Thus, optical technologies are able to provide large transmission capacities that, in turn, can be used by higher network layers to provide high bandwidth to advanced network applications and services. On the one hand, bursty traffic is becoming predominant in the current data centric networks. On the other hand, network operators have to be able to provide users with high-quality services and high bandwidth on demand exploiting the potential for statistical multiplexing in the time domain. Future telecommunications applications will require a dynamic, high-capacity optical network with the capability to carry heterogeneous network traffic. Therefore, a finer bandwidth granularity and a faster response to traffic variations than in optical circuit-switched WDM networks is needed, which can be achieved by employing optical packet switching (OPS) [1-4] or optical burst switching (OBS) [5-7]. The main advantage of those techniques in comparison to pure optical circuit-switched networks is an improvement of bandwidth utilization through limiting the wavelength occupation time by a single connection, thereby gaining increased wavelength utilization from statistical multiplexing directly in the optical layer. In the last few years, a large penetration of optical transmission technology into access and local networks can be observed, especially in the urban areas. The capacity of current local area networks has been rapidly increased by employing advanced network technologies such as Gigabit Ethernet (GbE) and 10 GbE. The access bottleneck can be reduced by introducing various fiber to the home/curb/subscriber implementations in conjunction with wireless access services such as universal mobile telecommunications system (UMTS) and wireless local area networks (WLANs). Storage-based technologies and protocols such as Fibre Channel, ESCON, FICON, FCIP, and iscsi have also to be considered when talking about heterogeneity of data traffic and protocols that need to be supported by optical backbones. The role of metropolitan area networks (MANs) is to interconnect high-speed WDM backbone networks with access networks in order to provide a high bandwidth to heterogeneous customers and to support business to business solutions (e.g. distributed storage and enterprise networks). 1

2 The future MANs have to be able to provide high capacity while maintaining efficiency and flexibility that are required to handle heterogeneous traffic. Besides that, they should be upgradeable in a modular way, transparent to multiple legacy services/protocols, and highly reliable. A significant research effort has been put recently into development of new architectures and protocols for optical metropolitan area networks. Several different ring-based architectures for advanced optical metro networks have been proposed and investigated [8-17]. Wavelength allocation and access to the media range from static wavelength-routed optical network (WRON) over static slotted ring (SR) architectures to dynamic optical burst switched and optical packet switched networks. Dynamic optical burst switched (OBS) networks can implement either a one-way resource reservation scheme without acknowledgement or an end-to-end reservation scheme with acknowledgement. The first option typically uses either the just-in-time (JIT) or the just-enoughtime (JET) resource reservation technique. Here, a burst can be dropped at a traversed node due to contention. Consequently, wavelength conversion is essential in order to decrease the probability of contention. In an OBS network with reservation acknowledgment, reservation of an end-to-end lightpath has to be acknowledged before transmission of the data burst. Since the end-to-end resources are reserved a-priory, lossless transport and end-to-end delay can be guaranteed. An alternative to OBS networks using switching in the wavelength domain is to use switching in the time domain. In this approach, named time sliced optical burst switching (TSOBS), wavelength channels can be organized into a series of frames, each of which is sub-divided into fixed length timeslots [18]. The equidistant time slots at a fixed position within successive frames form time-division multiplexed (TDM) channels, which are then used to transmit data bursts over a number of frames. The key building blocks of TSOBS routers are so-called optical time slot interchangers (OTSI), which provide the required time-domain switching for all wavelengths. Due to the fact that switching is done in the time domain rather than in the wavelength domain, there is no need for wavelength converters. This is the main advantage of the TSOBS approach. Switching packets directly in the optical domain is an attractive solution for future high-capacity, dynamic metro networks combining the benefits of packet flexibility and optical transparency. packets provide a good way to efficiently integrate different layers and to reduce the network operating cost. However, enabling technologies for ultrahigh-speed all-optical packet switching is either not commercially available today or very complex and expensive. Therefore, a combination of optical packet and burst switching based on commercially available components could be a good choice for a dynamic and efficient high-capacity optical MAN. In this paper, a WDM based metropolitan area network with a double counter-rotating ring structure and optical packet/burst switching is presented. The network comprises edge nodes fully transparent to data rate and format of access traffic. The network nodes can be implemented using commercially available components only. Because of its transparency and time-slotted wavelength channels, the network is flexible and can be easily upgraded step-by-step. The end-to-end delay through the entire network can be guaranteed by a-priory reserving a number of slots within a frame. For isochronous time-critical multimedia data-streams, equidistant slots can be reserved by the master node during the connection setup. 2 Network and Node Architecture 2.1 Network Architecture The network uses a central dual counter-rotating WDM ring to interconnect M metro edge nodes that aggregate traffic form concentration units (CUs), which are located in the access area, i.e., represent the optical line terminals (see Fig. 1). Each fiber ring carries W time-slotted wavelength channels. One of the channels is dedicated to transmission of control and configuration information as well as to serve synchronization of all edge nodes to a central slot clock generated by the master node. Additional to the generation of the slot clock, the master node is also responsible for frame creation, (re-) configuration and slot reservation for isochronous data transmission. Note that every metro edge node is able to act as a master node in the case of a failure in the current master node. In this case, other edge nodes will not receive the slot clock from neither ring and a procedure for selection of a new master node will automatically start. 2

3 Frame q Frame q + n Control Channel Datal Channels Metro Edge Node 2 Slots CU CU Metro Edge Node 1 (Master node) Packet/Burst Switched WDM Ring - MAN Metro Edge Node 3 CU PON CU CU CU: Concentration Unit PON: Passive Network MAN: Metropolitan Area Network Metro Edge Node M Fig. 1. packet/burst switched metro network based on a double counter-rotating slottedring topology. Edge nodes are basically nothing more than transparent optical switches. Only the control channel is terminated and processed by every edge node, while data channels are switched and forwarded to the concentration units that are capable of transmitting and receiving data slots. The CUs comprise either tunable or fixed transceivers. Low-cost terminals equipped with fixed transceivers are capable of receiving/transmitting on a single wavelength only, and consequently, they are disadvantaged in comparison to the high performance units equipped with tunable transceivers. If needed, low-cost concentration units can be updated from the single-wavelength to a multiwavelength version without any significant change in the network. New configuration data for this particular unit have only to be broadcasted to all edge nodes. The concentration units need to be synchronized to the ring slot clock in order to be able to transmit data in a particular time slot. The slots are grouped into frames of identical length in both control and data channels. The occupation of a slot is indicated in the frame q of the control channel that precedes the data frame containing the particular slot (frame q + n). An edge node can reserve a free data slot by indicating it in the corresponding slot of the control channel. The occupation is done after collecting transmission requests from all CUs connected to the particular edge node. Please note that not only a single slot but also a number of slots within a frame can be reserved in order to transmit a data packet or a burst that exceeds the size of a single slot. Unlike the TDM approach in TSOBS, a single data burst can not occupy slots in more than one frame. The time between the slot reservation in frame q of the control channel and the actual data transmission in the corresponding slot of frame q + n on the according data channel is needed to inform the requesting CU about the successful slot occupation and to give the CU enough time to prepare timely accurate sending of its data in the correct slot also by taking into account the signal propagation delay in the fiber connecting the CU and the edge node. Because of the time-slotted data transmission and the centralized media access control in the upstream direction, a TDMA-based passive optical network (PON) can be directly attached to a port of an edge node. A WDM PON access infrastructure can also be integrated. That is, not only point-to-point but also point-to-multipoint connections can be established between an edge node and concentration units, which makes the network architecture even more flexible. 2.2 Node Architecture The architecture of edge nodes for the proposed optical MAN is shown in Fig. 2. It consists of a transparent optical packet/burst switch and two units for synchronization and processing of control channels of the two counter-rotating rings. The control channels are terminated at every node. They are dropped by using fibre Bragg gratings (FBGs) and circulators, then received by fixed receivers, modified at the node, and finally, forwarded to neighbouring nodes by using fixed transmitters and WDM couplers. The slot clock is extracted from both control channels and used to synchronize all the concentration units (CUs) attached to this node. The node indicates slot occupation in a control frame according to the transmission requests from CUs and sends the modified control frame to the next edge node. This is the main task of the metropolitan area part of 3

4 the media access control unit (Metro MAC). The variable optical delay, τ, has two functions. The first one is to compensate for the control channel processing latency at the node. The second one is to ensure that the frames from both rings arrive at the same time at the node switch. From CU 1 Slot Clock Recovery and Control Channel Rx/Tx Access MAC Metro MAC From Ring 2 To Ring 1 Circulator FBG Coupler τ From CU N N To / From Access Layer Packet/Burst Switch To CU 1 N τ To CU N Access MAC Metro MAC Slot Clock Recovery and Control Channel Rx/Tx Coupler FBG Circulator FBG: Fiber Bragg Grating MAC: Media Access Control To Ring 2 From Ring 1 Fig. 2: Architecture of the metro edge node. The access MAC module is responsible for synchronization and coordination of data transmission to and from concentration units (CUs). The whole communication between the access MAC and CUs is done over an access control channel, which also has a slotted structure directly corresponding to the structure of the control channel on the metro rings. for data transmission in frame q + n form the CUs, and then, after the slot reservation in the ring control channel has been done by the Metro MAC module, it forwards the results of the slot occupation via the access control channel to the CUs, thereby acknowledging the requests. CUs aggregate the traffic received from a number of users, perform burst assembly and send optical bursts or individual optical packets to the metro edge node after receiving a confirmation of successful resource reservation from the edge node. The Access MAC module is also responsible for collecting and forwarding the configuration data to and from CUs. The CUs need to be synchronized to the ring slot clock. For this purpose, every CU needs, in addition to the extraction of the slot clock form incoming control frames, also to measure the time needed for signal propagation from the edge node to the CU and for processing of the control frame. These time delays need to be taken into account when calculating the exact timing for data transmission that must fit in the specific slot on the metro ring. Note that metro edge nodes extract the slot clock from the control channel electronically. There is no need for all-optical clock recovery. The first two input/output ports of the optical packet/burst switch are connected to the two counterrotating metro rings, while the residual N ports are used to attach concentration units to the edge node and to send/receive the access control channel. Similarly to the ring control channels, also the access control channel is separated from data channels before entering the switch by means of FBGs and circulators (see Fig. 3). The access control channel is added to the output ports connected to CUs by using WDM couplers. The ring ports have a higher priority than access ports. That is, the connections through the switch for a particular time slot are first determined for ring data traffic and after that for access data traffic upon availability. If a connection for transmitting access data in a time slot is not possible, a negative acknowledgment is sent back to the requesting CU, which can again request for data transmission in the next reservation cycle. Thus, due to the fact that data is transmitted only if the result of the distributed resource reservation procedure is positive, there is no packet or burst loss in the network. However, if a packet is lost because of overflow in a CU's transmitting buffer, the re-transmission will be provided by higher layers. In order to support high-priority isochronous data transmission, the master node can reserve a number of equally spaced time slots in each frame for a single end-to-end connection after a previous request during the connection setup. 4

5 The structure of an optical packet/burst switch is shown in Fig. 3. It consists of 2 (N+2) passive optical couplers/splitters and (N+2) 2 wavelength selective modules (WSMs). In each WSM, the incoming WDM signal is divided into W single channel paths by using a WDM de-multiplexer. The wavelength channels are selected using semiconductor optical amplifiers (SOAs). Only the channels, for which SOAs are set in the ON state by applying a current to them, pass the device. The selected channels are then collected at the output of the WSM by using a WDM multiplexer. We consider the deployment of gain-clamped SOAs (GC-SOAs) because of their large input dynamic range in comparison to conventional SOAs. They can also be used to amplify the signal, thereby compensating for losses in optical couplers/splitters and WDM multiplexers/demultiplexers. The switch is basically implemented as a broadcast-and-select architecture, so that wavelength channels from any input port can be forwarded to any output port. A single wavelength channel from a particular input port can also be selected at more than one output port simultaneously, thus, multicasting by copy&forward can easily be sup-ported directly in the optical domain. Switch Control Input Port 1 From Ring 1 WSM 1,1 WSM 1,2 WSM 1,3 WSM 1,N+2 Output Port 1 To Ring 1 Wavelength Selective Module (WSM) GC-SOA 1 GC-SOA 2 Input Port 2 From Ring 2 WSM 2,1 WSM 2,2 WSM 2,3 WSM 2,N+2 Output Port 2 To Ring 2 WDM Demux GC-SOA W WDM Mux Input Port 3 FBG From CU 1 To Access MAC WSM 3,1 WSM 3,2 WSM 3,3 WSM 3,N+2 Output Port 3 To From CU 1 Access MAC Input Port N + 2 FBG From CU N To Access MAC WSM N+2,1 WSM N+2,2 WSM N+2,3 WSM N+2,N+2 Output Port N + 2 From Access MAC To CU N Fig. 3: A broadcast-and-select optical packet/burst switch with N + 2 ports, each of which carrying W wavelength channels (CU: Concentration Unit, GC-SOA: Gain-Clamped Semiconductor Amplifier). 3 MAC Protocol The media access control (MAC) protocol for the proposed network concept comprises two ports, namely CU-MAC (Access MAC) and Edge-MAC (Metro MAC). Edge nodes comprise both sub- MACs, while in CUs, only the CU-MAC has to be implemented. Within edge nodes, both sub- MACs have to be synchronized by means of slot and frame clock. Fig. 4 shows two diagrams depicting the functionality of the CU-MAC and the Edge-MAC within edge nodes. 5

6 a) b) Receive frame q form the ring control channel read DAs and SAs for slot i in control frame q and store them in memory compare DAs with addresses of assigned CUs and store results in memory forward data carried in slot i of data frame q if SA in control frame q n = address of an assigned CU read SOT for all channels and slots within control frame q (for data frame q+n) Receive frame q form the access control channel read requests within slot i and store them in memory (request table) look the updated SOT up for frame q+n send acknowledges to CUs in case of successful slot reservation delete acknowledged requests form the request table look the CU request table up accommodate CU requests if possible (slot reservation) update SOT of control frame q for data frame q+n create and send updated control frame q on the ring control channel Fig. 4. Description of a) Edge-MAC and b) CU-MAC procedure during receiving of control frame q (DA: Destination Address, SA: Source Address, CU: Concentration Unit, SOT: Slot Occupation Table). The Edge-MAC is responsible for the reception and processing of the control channel on the metropolitan rings as well as for forwarding data from the rings destined to the CUs that are directly connected to this particular edge node (assigned CUs). It also reserves resources on the ring for transmission of data originating from the assigned CUs and establishes communication between CUs and the corresponding edge node. CUs send requests for transmission of data to the edge node on the control channel; while edge nodes may acknowledge a request by indicating it in the corresponding field of a slot in the access control channel (see also Fig 6b for the structure of the access control channel). The payload data is then sent on one of the W 1 data channels. The acknowledged requests presume a successful reservation of resources in a future data frame, which can be done by indicating it in advance on the ring control channel. The structure of the data and control channels for both ring and access parts are described in the next Section. Please note that information about slot occupation of the incoming frame is carried in the slot occupation table (SOT) for this frame, which is transmitted within the first slot. Thus, already after reading the first slot, the Edge-MAC has the information about the occupation of all slots in the incoming frame and can immediately start with the slot reservation procedure. Destination and source addresses (DA and SA) for data packets and bursts in frame q have already been received with frame q n and stored in memory, so that edge nodes can directly forward packets and burst addressed to assigned CUs. 4 Frame Format Both data and control frames are divided into j slots as shown in Figs. 5 and 6. The start and the end of a frame are indicated by a frame delimiter in the control channel. The time slots are separated by slot delimiters, which can be used to extract the slot clock. Connections through the switch can be rearranged at the start of every slot. There are guard intervals inserted immediately after slot delimiters in order to provide enough time for a clear change of the switch configuration between any two time slots, i.e., to avoid signal degradation caused by transitional effects in the switches and by the finite response time of the control circuitry. This makes possible switching of single packets or data bursts. Packets smaller than or equal to the slot length are easily transmitted by occupying a single slot. Long packets or data bursts are transmitted as a continuous data block by occupying a number of slots. In this case, there is no need for guard periods between slots during the packet/burst transmission, which improves significantly the transmission efficiency. The occupation of data slots on the metro ring is indicated in the control frame (see Fig. 5). Each slot of a control frame q is divided into W fields of which the first field is used for configuration as well as for exchange of the operation, administration and maintenance (OAM) information, while 6

7 the residual W 1 fields are used to indicate the status of the corresponding time slot in the frame q + n on the W 1 data channels. That is, the control frame precedes the corresponding data frame for a number of frames, n, which is chosen such that edge nodes have enough time to inform the concentration units about successful slot occupations and to allow the CUs enough time to timely transmit their data in the specified time slot. The W 1 channel occupation fields contain destination address (DA), source address (SA) and the length in slots of a data packet or burst to be transmitted over this channel. The reserved part of the channel occupation fields can be used for communication between edge nodes and the master node, for example to request reservation of resources for isochronous transmission. There is an additional field of the length of four Bytes in every of the W 1 channel occupation fields of the first slot that carries a slot occupation table (SOT) for the corresponding channel. Each of the SOT bits indicated occupancy of a slot in the particular channel, i.e., logical 1 means that the particular slot is occupied and logical 0 that it is free. The complete SOT indicating status of all slots at W 1 data channels is thus received during the first slot, which allows a significant reduction of the control channel processing delay. Configuration Channel Control Frame q DA SA Length Ch4 SOT Reserved Ch 1 Ch 2 Ch 3 Ch 4 Ch W 1 DA: Destination Address SA: Source Address Ch: Channel SOT: Slot Occupation Table Slot 1 Slot 2 Slot 3 Slot 4 Slot j Slot Delimiter Frame Delimiter A Free Slot An Occupied Slot Data Packet/Burst Slot 1 Slot 2 Slot 3 Slot 4 Slot j Header Payload An Occupied Slot Guard Interval Slot 1 Slot 2 Slot 3 Slot 4 Slot j Control Channel Data Channel 1 Data Channel W 1 Data Frame q + n Fig. 5. Format of data and control channel frames. A CU can request a single slot or a number of slots for transmission of a packet or a data burst in data frame q + n by indicating it in the corresponding slot of access control frame q (see Fig. 6a). For this purpose, there are W 1 Req. fields in every slot of an access control frame. The Req. fields contain source and destination addresses of the requested transmission as well as a Req. Length field and a Req. # field. These fields are contained in every slot of the control frame. a) Control Frame q Configuration Channel Req. DA Req. SA Req. Length Req. # Reserved Req 1 Req. 2 Req. 3 Req. 4 Req. W 1 Slot 1 Slot 2 Slot 3 Slot 4 Slot j Control Channel b) Control Frame q Configuration Channel DA SA Length Ack. # Reserved Ch 1 Ch 2 Ch 3 Ch 4 Ch W 1 DA: Destination Address SA: Source Address Ch: Channel Control Channel Fig. 6. Structure of the access control channel: a) from CU to metro edge node and b) from metro edge nodes to CUs. Fig. 6b shows the format of the control channel in the access area that is transmitted from a metro edge node to the CUs attached to this particular edge node. The structure of the access control 7

8 frame is similar to that of the metro ring, but instead of the SOT field there is an Ack. # field, which can be used to acknowledge a CU s request to send data in frame q + n of the corresponding data channel. The Ack. # field is contained in every slot of the control frame. The format of all address fields is identical. The first bit (bit 0) is an I/G flag showing if the address is an individual (unicast) or a group address (multicast). The bits 1 to 8 carry a metro edge node address. By using the 8 bit addresses it is possible to address up to 256 edge nodes. The residual 15 bits (bits 9 to 23) are used to address a particular CU. The CU address space is large and can support up to 32,768 CUs. The format of address fields is depicted in Fig I/G Flag Metro Edge Node Address CU Address Fig. 7. Format of address fields. The frame length, slot size and number of wavelength channels are design issues. However, the frame should be equal or larger than the max. burst size of the system. The guard intervals should be larger than 10 ns, in order to allow switching transients to die away. 5 Transmission Performance As already mentioned, the proposed network is transparent with respect to data rate and modulation format. However, optical signals at different data rates and using different modulation schemes will be unequal affected by various effects during their propagation through the network elements. In order to determine the scalability of the optical packet/burst switch, which is the main part of the metro edge node, a simulation model of a metro edge node has been developed. In the simulation setup shown in Fig. 8, dense wavelength-division multiplexed (DWDM) signals at 10 Gbit/s and 40 Gbit/s were generated, then amplified in an optical amplifier, and finally injected into an edge node. The separation of wavelength channels was chosen to be 100 GHz according to the ITU-T grid with the central channel at λ center = nm. All data channels were generated by using non-return-to-zero (NRZ) modulated pseudorandom bit sequences (PRBS). A single signal path through the switch was modeled by taking into account all relevant effects such as losses, noise sources, nonlinear effects, and crosstalk originating from neighboring channels. The model of an edge node comprise an optical circulator, a fiber Bragg grating, an 1 N optical splitter, an N 1 optical coupler, WDM multiplexers and demultiplexers, and gain-clamped semiconductor optical amplifiers. The main parameters used in the simulations are listed in Table 1. For a more detailed description of the simulation setup and the models used please refer to [19]. λ 1 λ M NRZ PRBS DWDM Transmiitter Amplifier OA Circulator FBG WDM Demux M GC-SOA 3 M WDM Mux 1 2 Amplifier WDM Coupler OA Receiver λ C 1 N Splitter λ C N 1 Coupler Packet/Burst Switch Fig. 8. Simulation setup (NRZ: Non-Return-to-Zero, PRBS: Pseudo-Random Bit Sequences, DWDM: Dense Wavelength-Division Multiplexing, OA: Amplifier, FBG: Fiber Bragg Grating, SOA: Semiconductor Amplifier). Fig. 9 shows the results of the scalability study regarding both the number of wavelength channels and the number of switch ports. It can be seen form this figure that eye closure penalties (ECP) below 0.5 db were obtained for both 10 Gbit/s and 40 Gbit/s non-return-to-zero (NRZ) modulated signals transmitted on up to 32 wavelength channels and for switch sizes of up to For both 10 Gbit/s and 40 Gbit/s signals the ECP remains below 0.8 db even up to 128 ports and 64 8

9 channels. Consequently, 126 CUs and more than 8,000 users can be connected to a single edge node. Thus, the proposed architecture of metro edge nodes features a good scalability regarding both the number of CU ports and the number of WDM channels, which is large enough for application in the metropolitan area. Table 1: The main parameters used in simulations (PRBS: Pseudo-Random Bit Sequences, DWDM: Dense Wavelength-Division Multiplexing, PIN: Positive-Intrinsic-Negative Photodiode, GC-SOA: Gain-Clamped Semiconductor Amplifier, LPF: Low-Pass Filter). Parameter Value Unit DWDM Transmitter PRBS mark probability 0.5 laser power 5 mw laser linewidth 10 MHz no. of wavelength channels 1-64 channel spacing 100 GHz GC-SOA active region length 600 µm active region width 2.5 µm active region thickness 0.04 µm DBR sections length 100 µm bias current 150 ma group effective index 3.7 Receiver PIN responsivity 0.8 A/W LPF type Bessel LPF bandwidth 0.75 bitrate Hz bias current 150 ma a) ECP [db] Gbit/s NRZ 40 Gbit/s NRZ b) ECP [db] port count of the optical packet/burst switch Gbit/s NRZ Gbit/s NRZ number of wavelength channels Figure 9: Results on scalability of the metro edge nodes: eye closure penalty (ECP) for a) various switch sizes from 8 8 to and b) for different numbers of channels from 1 to 64 (NRZ: Non-Return-to-Zero). 9

10 6 Network Performance In this section, a preliminary performance study for the optical metropolitan area network is described. Let us consider a ring network consisting of N M metro edge nodes and N C concentration units connected to every single metro edge node as shown in Fig km 10 km 1 80 km km N M 2 1 N C Fig. 10. Considered example of a dual counter-rotating ring network 6.1 Average Waiting Time Since there is a slotted structure on two independent rings, the network can be described by two independent M/D/1 systems. According to the M/D/1 formulas, one can find the average waiting time in frames, T W, for which a burst of length B S in slots needs to wait in the CU s transmitting buffer for different network loads, ρ, as: T W = 1 ρ (1 ρ) N C ( N M 4 W 1) B j S, (1) where W is the number of wavelength channels and j is the frame length in slots. As it is mentioned in the previous Sections, CUs are allowed to transmit bursts that exceed the length of an individual slot. For the reason of simplicity, in this preliminary study we assume that all bursts existing on the ring at a particular time occupy the same number of slots. Consequently, the slotted nature of the network is preserved and the M/D/1 formula remains valid for any burst length. The influence of different burst lengths on the average waiting time is included through the quotient Bs/j in (1). Thus, an increase in average burst length leads to a lower number of bursts that can be transmitted within a frame, which is equivalent to a decrease of number of slots by keeping the frame length in bits constant. 6.2 Average Packet/Burst Transfer Delay The average time needed to send a burst form a CU to another CU in the network can be calculated as given in (2). Here we take into account all the processing latencies, the waiting time in the transmitting buffer, T W, as well as the delays caused by the transmission of optical signals in interconnecting fibers. 10

11 D = T req. + T reception_ of _ the _ first _ slot _( SOT ) + T slot _ reservation + 1 ρ N (1 ρ) C ( N M 4 W 1) B S T slot + (2) + T ack. + 2 T transmission _ access + T transmission _ metro The calculated average burst transfer delay through the network for different burst lengths is shown in Fig. 11. Average Burst Transfer Delay [µs] 650 B S = 30 Slots 625 B S = 15 Slots 600 B S = 10 Slots 575 B S = 1 Slot B S = 5 Slots B S = 2 Slots Network Load Fig. 11. Average burst transfer time trough the network vs. offered network load for various burst lengths (B S ). For the calculations, we assumed a data rate of 10 Gbit/s, N M = 15, N C = 30, W = 40 and j = 30. It can be seen from Fig. 11 that the average transfer delay remains at a very low level even for a high load situation when short bursts occupying only one or two slots are sent. Even up to a burst length of 15 slots, the average transmission time remains relatively low for network loads up to Very large data bursts occupying the whole frame (B S = 30) should be avoided in the case of heavy network load because of the long waiting time in the transmitting buffer, which can cause overflow of the CU s transmitting buffer, and consequently, loss of data. Please note that the effect of resource sharing between bursts of different lengths is not shown in Fig. 11. The coexistence of different burst lengths on the ring at the same time would probably lead to increased waiting times for longer bursts. 6.3 Throughput Efficiency Since data packets and bursts can generally be of any length and the resources for their transmission through the network are reserved in multiples of the slot length, it would be of particular interest to see how a change of slot length affects the achievable throughput efficiency. We can estimate throughput efficiency, E T [%], using the following simple expression: E T j j j Sb ( Db + GIb ) Bb Sb BS BS BS = 100 = j S b Db + GIb + Bb Sb BS = Sb BS [%] (3) In Equation (2), S b, D b, GI b, and B b denote the lengths in bits of the slot, delimiter, guard interval, and burst, respectively. The throughput efficiency for three different slot sizes of 64 Bytes,

12 Bytes and 9000 Bytes is plotted in Fig. 12 versus burst length. The slot length of 1518 Bytes is chosen because it is the maximum length of the payload field of an Ethernet frame, while 9000 Bytes correspond to the length of the jumbo Ethernet frames Throughput Efficiency [%] Bytes Slots 1518 Bytes Slots 9000 Bytes Slots 0 0 5,000 10,000 15,000 20,000 25,000 30,000 35,000 40,000 Burst Lenght [Bytes] Fig. 12. Throughput efficiency vs. burst length for three different slot lengths. We notice form Fig. 12 that when using small slots, high efficiencies of more than 90% can already be achieved for small bursts. Efficiencies above 90% can be achieved for longer bursts too, but in this case, a large number of slots are needed for transmission of a single burst. Additionally, smaller slots mean higher frequency of the slot clock and larger number of slot delimiters. Therefore the maximum achievable efficiency is limited to about 95% in our example. On the other hand, large slots are extremely inefficient if many small packets are transmitted. For large busts, the efficiency is increased and can reach almost 100% for some specific burst lengths, i.e., for burst lengths either corresponding to the slot length or to an exact multiple of the slot length. However, a small change in the burst length can cause a large difference in the efficiency. Therefore, choosing smaller slots may be more appropriate for variable burst lengths. 7 Conclusion In conclusion, a novel architecture of optical packet/burst-switched metro network is introduced in this paper. The main part of the network is an optically transparent metro edge node. It enables a dynamic, reliable, high-speed, and low-delay data transmission over the entire network. This paper describes the network architecture and some details of the appropriate media access control (MAC) protocol. Some preliminarily results of a study on edge node scalability are presented. The node features a good scalability for application in the metropolitan area. Also preliminary results of a transmission and network performance study are presented. Through merging the metropolitan area network with a large part of the access network infrastructure it becomes possible to move the data processing complexity from metro edge nodes to concentration nodes that are located close to the end users. This architectural model has a large potential to improve reliability, flexibility, and throughput efficiency of both access and metropolitan parts of the network. References [1] N.Wada, W. Chujo, and K. Kitayama, 1.28 Tbit/s (160 Gbit/s x 8 Wavelengths) Throughput Variable Length Packet Switching Using Code Based Label Switch, ECOC 2001, Amsterdam, Netherlands, Vol. 6, No. 2, (2001), pp [2] B. Meagher, et al., Design and Implementation of Ultra-Low Latency Label Switching for Packet-Switched WDM Networks IEEE JLT, Vol. 18, No. 12, (2000), pp [3] M. White, et al., The Architecture of HORNET: A Packet-Over-WDM Multiple-Access Metropolitan Area Ring Network, ELSEVIER Computer Networks, Vol. 32, No. 5, (2000), pp

13 [4] S. Aleksić, Packet-Switched OTDM Networks Employing the Packet Compression/ Expansion Technique, SPRINGER Photonic Network Communications, Vol. 5, No. 3, (2003), pp [5] Z. Rosberg, M. Zukerman, and J. White, Performance Analyses of Burst- Switching Networks, IEEE JSAC, Vol. 21, No. 7, (2003), pp [6] S. Bjørnstad, H. Øverby, Quality of Service Differentiation in Packet/Burst Switching. A Performance and Reliability Perspective, in ICTON2005, Barcelona, Spain, Vol. 1, (2005), pp [7] K. Dolzer, C. Gauger, J. Späth and S. Bodamer, Evaluation of Reservation Mechanisms for Burst Switching, Intern. Journal of Electronics and Communications (AE), Vol. 55, No. 1, (2001), pp [8] S. Yao, at al., All- Packet Switching for Metropolitan Area Networks: Opportunities and Challenges, IEEE Comm. Magazine, Vol. 39, No. 3, (2001), pp [9] L. G., Kazovsky at al., High Capacity Metropolitan Area Networks for the Next Generation Internet, 35th Asilomar Conf. Signals, Systems, and Comp., Vol. 1, Pacific Grove, CA (2001), pp [10] M. Scheutzow at al., Wavelength Reuse for Efficient Packet-Switched Transport in an AWG-Based Metro WDM Network, IEEE JLT, Vol. 21, No. 6 (2003), pp [11] A. S. Acampora, A High Capacity Metropolitan Area Network Using Lightwave Transmission and Time-Multiplexed Switching, IEEE Transactions on Comm., Vol. 38, No. 10 (1990), pp [12] A. Zapata at al., Next Generation 100-Gigabit Metro Ethernet (100 GbME) Using Multiwavelength Rings, IEEE JLT, Vol. 22, No. 11 (2004), pp [13] S. Aleksić, Design Considerations for a High-Speed Metro Network using All- Packet Processing, ICTON2006, Vol. 3 (2006), pp [14] A. Carena et al.: RingO, An Experimental WDM Packet Network for Metro Applications, IEEE JSAC, Vol. 22, No. 8, (2004), pp [15] M. Herzog, M. Maier and M. Reisslein, Metropolitan Area Packet-Switched WDM Networks: A Survey on Ring Systems, IEEE Comm. Surveys, Vol. 6, No. 2 (2004), pp [16] D. Dey, A. van Bochove, A Koonen, D. Geuzebroek, and M. Salvador, FLAMINGO: A Packet-Switched IP-over-WDM All-optical MAN, ECOC 2001, Amsterdam, Netherlands, Sept. 30 Oct. 4, 2001, pp [17] I. Chlamtac, V. Elek, A. Fumagalli, and C. Szabó, Scalable WDM Access Network Architecture Based on Photonic Slot Routing, IEEE/ACM Transactions on Networking, Vol. 7, No. 1, Feb. 1999, pp [18] J. Ramamirtham, J. Turner, Time Sliced Burst Switching, IEEE INFOCOM 2003, March/April 2003, San Francisco, CA, USA, Vol 3, (2003), pp [19] S. Aleksić, Transmission Performance of ly Transparent Metro Edge Nodes, ICTON2007, Rome, Italy, Vol. 3, (2007), pp