OPTICAL NETWORKS Optical Packet Switching Optical Burst Switching A. Gençata İTÜ, Dept. Computer Engineering 2005
Optical Packet Switching OPS 2
Optical Packet Switching An optical cross-connect (OXC) can only switch whole wavelengths unless electrical grooming switch fabrics are used. The past evolution of the Internet has shown that a future-proof network needs: high-capacity circuit switching, high-performance switching with finer granularities. This argument is the main motivation for Optical Packet Switching, OPS in short. 3
OPS Basics Optical packet switching is optical switching with the finest granularity. Incoming packets are switched all-optically without being converted to electrical signal. It is the most flexible and also the most demanding switching scheme. There are two categories of optical packetswitched networks: slotted (synchronous) networks un-slotted (asynchronous) networks 4
Slotted Networks At the input ports of each node, packets can arrive at different times. The switch fabric can change its state incrementally, or set up one input-output connection at an arbitrary time jointly set up multiple input-output connections together at the same time. Then, it is possible to switch multiple time-aligned packets together or each packet individually on the fly. In both cases, bit-level synchronization and fast clock recovery are necessary for packet-header recognition and packet delineation. 5
Slotted Networks In general, all of the packets have the same size in a slotted network. In a variation, it is possible that packets are of variable length, but each packet s length is an integral multiple of a slot, and all packet transmissions start at a slot boundary. A fixed-size time slot contains both the payload and the header. The time slot has longer duration than the whole packet to provide some extra guard time. All the input packets arriving at the input ports need to be aligned in phase with one another before entering the switch fabric. 6
Generic Node Architecture for Slotted Networks 7
Packet Delay Variations The time for a packet to travel through a certain distance of the fiber depends on: the fiber length, the chromatic dispersion, and the temperature variation. Chromatic dispersion results in different propagation speed for packets transmitted on different wavelengths. For example: A typical fiber dispersion of 20 ps/nm/km (ps is the time unit for delay variation, nm the unit for wavelength difference, and km the unit for propagation distance), A wavelength variation of 30 nm, and A propagation distance of 100 km the propagation delay variation would be about 60 ns. 8
Packet Delay Variations The packet propagation speed also varies with temperature, with a typical figure of 40 ps/ 0 C/km. 100 km of fiber under temperature variation range of 0-25 0 C introduces 100 ns of delay variation. The delay each packet experiences inside a node depends on the switch fabric and contention-resolution scheme. Depending on the implementation of the switch fabric, a packet can take different paths with unequal lengths within the switch. The jitter induced by dispersion between different wavelengths and unequal optical paths varies from packet to packet at the output of the switch. (A fast output synchronization interface might be required.) Thermal effects within a node can be easily controlled. 9
Packet Delineation Packet delineation is essential for both header reading and switch configuration. During packet delineation, the incoming bits are locked in phase with the clock so that the node can read the header information. Packets enter a node from different links. For all the previously-stated reasons, they could arrive totally out of phase with one another. A typical synchronization stage consists of a series of switches and delay lines. 10
Packet Delineation Once the header processor recognizes the bit pattern and performs packet delineation, it identifies the packet start time, the control unit calculates the necessary delay and configures the correct path through these switched delay lines. The length of the delay lines are in an exponential sequence between the 2x2 switches, i.e., the first delay line is equal to ½ a time-slot duration, the second delay is equal to 1/4 a time-slot duration, etc. Such a packet-synchronization scheme introduces insertion loss and crosstalk due to the switches used. Cascading the switches will inevitably require optical amplification, which will result in degraded signal-to-noise ratio. The crosstalk accumulated through the switches may also increase the bit-error rate. In a multi-node network, the power penalty brought by all the synchronization stages may significantly impair the system performance. 11
Un-slotted Networks In an un-slotted network, the packets may or may not have the same size. Packets arrive and enter the switch without being aligned in time. The packet-by-packet switch operation could take place at any point in time. Obviously, the chance for contention is larger because the behavior of the packets is more unpredictable. On the other hand, un-slotted networks are more flexible compared with slotted networks, since they are better at accommodating packets with variable sizes. 12
Generic Node Architecture for Un-slotted Networks 13
Node Architecture The fixed-length fiber delay lines hold the packet when header processing and switch reconfiguration are taking place. There is no packet-alignment stage. All the packets experience the same amount of delay with the same relative position in which they arrived, provided there is no contention. Given the same traffic load, the network throughput is lower than that of the slotted networks because contention is more likely to occur. 14
Contention Resolution In an OPS network, contention occurs at a switching node whenever two or more packets try to leave the switch fabric on the same output port, on the same wavelength, at the same time. In electrical packet-switched networks, contention is resolved with the store-and-forward technique. The packets in contention are stored in a memory bank, and are sent out at a later time when the desired output port becomes available. Store-and-forward is possible because of the availability of electronic random-access memory (RAM). There is no equivalent optical RAM technology. Therefore, the optical packet switches need to adopt different approaches for contention resolution. 15
Contention Resolution Mechanisms Wavelength conversion Offers effective contention resolution without relying on buffer memory. Wavelength converters can convert wavelengths of packets which are contending for the same wavelength of the same output port. It is a powerful and the most preferred contention-resolution scheme that does not cause extra packet latency, jitter, and resequencing problems. 16
Contention Resolution Mechanisms Optical delay line Provides sequential buffering. It is a close imitation of the RAM in electrical routers, although it offers fixed and finite amount of delay. Many previously-proposed architectures employ optical delay lines to resolve contentions. Optical delay lines rely on the propagation delay of the optical signal in silica to buffer the packet in time. This provides sequential access: They have more limitations than the electrical RAM. To implement large buffer capacity, the switch needs to include a large number of delay lines. 17
Contention Resolution Mechanisms Space deflection It is a multiple-path routing technique. Packets that lose the contention are routed to nodes other than their preferred next-hop nodes. They are expected to be eventually routed to their destinations. The effectiveness of deflection routing depends heavily on the network topology and the offered traffic pattern. 18
Contention Resolution Both wavelength conversion and optical buffering require extra hardware and control software. wavelength converters and lasers for wavelength conversion; fibers and additional switch ports for optical buffering. Deflection routing can be implemented with extra control software only. Due to the lack of viable optical RAM technologies, alloptical networks find it difficult to provide packet-level synchronization, which is required in slotted networks. In addition, it is preferred that a network can accommodate natural IP packets with variable lengths. 19
Node Architecture 20
Node Architecture 21
Summary Optical packet switching is promising to offer large capacity and data transparency. However, after many years of research, this technology has not yet been applied in actual products, because of: The lack of deep and fast optical memories The poor level of integration. These issues will need to be overcome, not only through technical breakthroughs but also through clever network design, making optimal use of optics and electronics, wherever they fit best. In the near future, the developments in OPS seem to lead to integration of optical and electronic networks and the use of optical burst switching (OBS). 22
Optical Burst Switching OBS 23
Introduction Optical Burst Switching (OBS) has been proposed as a compromise between optical circuit switching and optical packet switching, while avoiding their shortcomings. In 1983, the concept of burst switching was first proposed, which was applied to transfer voice and data traffic over TDM links. At that time, it was lacking attention because it failed to achieve better performance over ATM. In 1990, the concept of Optical Burst Switching (OBS) was proposed in project Highball at University of Delaware. OBS refers to a broad class of sub-wavelength switching architectures, which assemble optical bursts at the network edge and transparently forward the bursts through the core network. 24
OBS Network 25
OBS Network The ingress edge routers aggregate client data (IP packets) into large bursts. Each burst is associated with a control packet. It contains the related control information such as burst length and routing information. Compared to OPS, the most important characteristic of OBS is to separate the burst data from the control header by the interval of offset time. This characteristic is helpful to overcome the infancy of optical hardware logic in OBS today. The control packet is processed electronically at each intermediate node to make routing decisions. outgoing interface and wavelength. The OXCs are configured to switch the data burst entirely in the optical domain, thereby removing the electronic bottleneck in the end-to-end data path. 26
OPS vs. OBS 27
Burst Assembly and Scheduling Burst assembly is the procedure for aggregating data packets from various sources into bursts at the edge of an OBS network. Assembly algorithms can be classified as: timer-based, threshold-based, mixed timer/threshold-based. Timer-based scheme: A timer starts at the beginning of each new assembly cycle. After a fixed time, packets that arrived in this period are assembled into a burst. Threshold-based scheme: A threshold as a limiting parameter determines when to generate a burst and send the burst into the optical network. The threshold specifies the number of packets to be aggregated into a burst, or the burst length. Until the threshold condition is met, the incoming packets are stored in prioritized packet queues in the ingress node. Once the threshold is reached, a burst is created and sent into the network. 28
Signaling A signaling protocol is the procedure through which services are provisioned. In a signaling protocol, service provisioning includes: switching-path establishment, deletion, modification. By using a signaling protocol, a control packet can reserve resources for the corresponding data burst by guiding it through a routing path. In an optical network, there are one-way or two-way reservation signaling protocols. 29
Routing Usually, routing algorithms for wavelengthrouted networks are directly adopted in OBS networks. However, it may be necessary to develop OBSspecific routing algorithms. Recently, some researchers have put more efforts to develop routing protocols for the OBS environment. Dynamic routing with preplanned congestion avoidance for survivable OBS networks is proposed. 30
Contention Resolution In an OBS network, the bandwidth reservation is a one-way process, i.e., a burst starts its transmission without waiting for a reservation acknowledgement. This requires an OBS network to resolve possible contention, which arises if two or more bursts simultaneously compete for the same output fiber on the same wavelength at the same time. In an optical network, contention resolution is usually solved in time dimension, space dimension, or wavelength dimension. 31
Current Trends 32