A Brief Introduction to Optical Networks EE 122
A Brief Introduction to Optical Networks EE 122, UC Berkeley April 27 & 30, 2001 Gaurav Agarwal gaurav@eecs. berkeley. edu 4/27/2001 EE 122, UC Berkeley
What I hope you will learn n n Why Optical? Intro to Optical Hardware Three generations of Optical Various Switching Architectures n n Circuit, Packet and Burst Protection and Restoration 4/27/2001 EE 122, UC Berkeley 2
Outline n n Why Optical? (Any guesses? ? ? ) Intro to Optical Hardware Three generations of Optical Various Switching Architectures n n Circuit, Packet and Burst Protection and Restoration 4/27/2001 EE 122, UC Berkeley 3
Bandwidth: Lots of it n Usable band in a fiber n n n Link Speeds upto 40 Gbps per n n n OC-3 155 Mbps OC-768 40 Gbps becoming available Total link capacity n n 1. 30 m - 1. 65 m 40 THz spaced at 100 GHz 400 s per fiber 400 * 40 Gbps = 16 Tbps! Do we need all this bandwidth? 4/27/2001 EE 122, UC Berkeley 4
Other advantages n n Transparent to bit rates and modulation schemes Low bit error rates n n n 10 -9 as compared to 10 -5 for copper wires High speed transmission To make this possible, we need: n n 4/27/2001 All-Optical reconfigurable (within seconds) networks Definitely a difficult task EE 122, UC Berkeley 5
What a path will look like Lasers generate the signal All-Optical Switch* Optical receivers All-Optical Switch* Optical Amplifier * All-optical Switch with wavelength converters and optical buffers 4/27/2001 EE 122, UC Berkeley 6
Outline n n Why Optical? Intro to Optical Hardware Three generations of Optical Various Switching Architectures n n Circuit, Packet and Burst Protection and Restoration 4/27/2001 EE 122, UC Berkeley 7
Fiber & Lasers n Fiber n n n Larger transmission band Reduced dispersion, non linearity and attenuation loss Lasers n n Upto 40 Gbps Tunability emerging Reduced noise (both phase and intensity) Made from semiconductor or fiber 4/27/2001 EE 122, UC Berkeley 8
Optical Amplifiers n As opposed to regenerators n n Make possible long distance transmissions Transparent to bit rate and signal format Have large gain bandwidths (useful in WDM systems) Expensive (~$50 K) Now: Optical Amps Then: Regenerators 4/27/2001 EE 122, UC Berkeley 9
Optical Add-Drop Multiplexers n Optical Add-Drop Multiplexer (OADM) n n Allows transit traffic to bypass node optically New traffic stream can enter without affecting the existing streams 1 2 3 1 OADM ’ 3 3 4/27/2001 2 ’ 3 EE 122, UC Berkeley 10
Optical Switches n n n Route a channel from any I/P port to any O/P port Can be fixed, rearrangable, or with converters MEMS (Micro Electro Mechanical Systems) n n Thermo-Optic Switches n n Agilent (HP) LC (Liquid Crystal) Switches n n JDS Uniphase, Nanovation, Lucent Bubble Switches n n Lucent, Optical Micro Machines, Calient, Xros etc. Corning, Chorum Technologies Non-Linear Switches (still in the labs) 4/27/2001 EE 122, UC Berkeley 11
MEMS Switches 2 -D Optical Switches n n Crossbar architecture Simple Digital Control of mirrors Complexity O(N²) for full non blocking architecture Current port count limited to 32 x 32. 4/27/2001 EE 122, UC Berkeley 12
3 D MEMS Switch Architecture 3 -D Optical Switches n n Analog Control of Mirrors. Long beam paths (~1 m) require collimators. Complexity O(N) (Only 2 N mirrors required for a full non blocking Nx. N switch) Lucent Lambda Router : Port 256 x 256; each channel supports upto 320 Gbps. 4/27/2001 EE 122, UC Berkeley 13
Wavelength Converters n n n Improve utilization of available wavelengths on links All-optical WCs being developed Greatly reduce blocking probabilities 3 2 WC No converters 1 New request 1 3 4/27/2001 With converters 1 New request 1 3 EE 122, UC Berkeley 14
Optical Buffers n n Fiber delay lines are used To get a delay of 1 msec: n n 4/27/2001 Speed of Light = 3*108 m/sec Length of Fiber = 3*108 *10 -3 m = 300 km EE 122, UC Berkeley 15
Outline n n Why Optical? Intro to Optical Hardware Three generations of Optical Various Switching Architectures n n Circuit, Packet and Burst Protection and Restoration 4/27/2001 EE 122, UC Berkeley 16
Generation I n n n Point-to-point optical links used simply as a transmission medium Fiber connected by Electronic routers/switches with O -E-O conversion Regenerators used for long haul Electronic data as the signal E-O Switch Signal received as electronic O-E-O Switch O-E Switch Regenerators 4/27/2001 EE 122, UC Berkeley 17
Generation II n n n Static paths in the core of the network All-Optical Switches (may not be intelligent) Circuit-switched Configurable (but in the order of minutes/hours) Soft of here 4/27/2001 EE 122, UC Berkeley 18
Gen II: IP-over-Optical Subnet IP Router Network NNI UNI Optical Subnet Light Path Optical Subnet IP Router Network End-to-end path 4/27/2001 EE 122, UC Berkeley 19
Peer Model n n n IP and optical networks are treated as a single integrated network OXCs are treated as IP routers with assigned IP addresses No distinction between UNI and NNI Single routing protocol instance runs over both domains Topology and link state info maintained by both IP and optical routers is identical 4/27/2001 EE 122, UC Berkeley 20
Overlay Model n n IP network routing and signaling protocols are independent of the corresponding optical networking protocols IP Client & Optical network Server Static/Signaled overlay versions Similar to IP-over-ATM 4/27/2001 EE 122, UC Berkeley 21
Integrated Model n n Leverages “best-of-both-worlds” by interdomain separation while still reusing MPLS framework Separate routing instances in IP and ON domains Information from one routing instance can be passed through the other routing instance BGP may be adapted for this information exchange 4/27/2001 EE 122, UC Berkeley 22
Generation III n n n An All-Optical network Optical switches reconfigurable in milliseconds Intelligent and dynamic wavelength asignment, path calculation, protection built into the network Possibly packet-switched Dream of the Optical World 4/27/2001 EE 122, UC Berkeley 23
Generation III (contd. ) n n n Optical “routers” perform L 3 routing No differentiation between optical and electrical IP domains Routing decision for each packet made at each hop Statistical sharing of link bandwidth Complete utilization of link resources 4/27/2001 EE 122, UC Berkeley 24
Outline n n Why Optical? Intro to Optical Hardware Three generations of Optical Various Switching Architectures n n Circuit, Packet and Burst Protection and Restoration 4/27/2001 EE 122, UC Berkeley 25
State of the World Today Electronic Network O/E/O E/O O/E/O E/O Electronic Network 4/27/2001 E/O Optical Core EE 122, UC Berkeley Electronic Network 26
View of a E/O node Input Port 1 Optical Link 1 Input Port 2 Input Port 3 Electrical Optical Link 2 Input Port 1 Input Port 2 Input Port 3 OP 1 OP 2 OP 3 OP 4 Optical Link 3 Input Port 4 Physical View 4/27/2001 Input Port 4 Logical View EE 122, UC Berkeley O P N-1 OPN 27
Optical Circuit Switching Electronic Network O/E/O OS E/O O/E/O OS E/O Electronic Network 4/27/2001 E/O Optical Core EE 122, UC Berkeley Electronic Network 28
Optical Circuit Switching Electronic Network O/E/O OS E/O O/E/O OS O/E/O WC OS E/O Electronic Network 4/27/2001 E/O Optical Core EE 122, UC Berkeley Electronic Network 29
Optical Circuit Switching n n n n A circuit or ‘lightpath’ is set up through a network of optical switches Path setup takes at least one RTT Need not do O/E/O conversion at every node No optical buffers since path is pre-set Need to choose path Need to assign wavelengths to paths Hope for easy and efficient reconfiguration 4/27/2001 EE 122, UC Berkeley 30
Problems n n Need to set up lightpath from source to destination Data transmission initiated after reception of acknowledgement (two way reservation) Poor utilization if subsequent transmission has small duration relative to set-up time. (Not suited for bursty traffic) Protection / fault recovery cannot be done efficiently Example : Network with N switches, D setup time per switch, T interhop delay. Circuit Setup time = 2. (N-1). T + N. D If N = 10, T = 10 ms, D = 5 ms, setup time = 230 ms. At 20 Gbps, equivalent to 575 MB (1 CD) worth of data ! 4/27/2001 EE 122, UC Berkeley 31
Optical Packet Switching n n n Internet works with packets Data transmitted as packets (fixed/variable length) Routing decision for each packet made at each hop by the router/switch Statistical sharing of link bandwidth leads to better link utilization Traffic grooming at the edges? Optical header? 4/27/2001 EE 122, UC Berkeley 32
Problems n n n Requires intelligence in the optical layer Or O/E/O conversion of header at each hop Packets are small Fast switching (nsec) Need store-and-forward at nodes or Deflection Routing. Also store packet during header processing Buffers are extremely hard to implement Fiber delay lines n n n 1 pkt = 12 kbits @ 10 Gbps requires 1. 2 s of delay => 360 m of fiber) Delay is quantized How about Qo. S? 4/27/2001 EE 122, UC Berkeley 33
Multiprotocol Lambda Switching n n D. Awduche et. al. , “Requirements for Traffic Engineering Over MPLS, ” RFC 2702 Problem decomposition by decoupling the Control plane from the Data plane n n 4/27/2001 Exploit recent advances in MPLS traffic engineering control plane All optical data plane Use as a “label” The on incoming port determines the output port and outgoing EE 122, UC Berkeley 34
OXCs and LSRs n n Electrical Network – Label Switched Routers (LSR) Optical Network – Optical Cross Connects Both electrical and optical nodes are IP addressable Distinctions n n n 4/27/2001 No merging No push and pop No packet-level processing in data plane EE 122, UC Berkeley 35
Optical Burst Switching n n n Lies in-between Circuit and Packet Switching One-way notification of burst (not reservation) – can have collisions and lost packets Header (control packet) is transmitted on a wavelength different from that of the payload The control packet is processed at each node electronically for resource allocation Variable length packets (bursts) do not undergo O/E/O conversions The burst is not buffered within the ON 4/27/2001 EE 122, UC Berkeley 36
Various OBSs n n n The schemes differ in the way bandwidth release is triggered. In-band-terminator (IBT) – header carries the routing information, then the payload followed by silence (needs to be done optically). Tell-and-go (TAG) – a control packet is sent out to reserve resources and then the burst is sent without waiting for acknowledgement. Refresh packets are sent to keep the path alive. 4/27/2001 EE 122, UC Berkeley 37
Offset-time schemes n n n Reserve-a-fixed-duration (RFD) Just Enough Time (JET) Bandwidth is reserved for a fixed duration (specified by the control packet) at each switch Control packet asks for a delayed reservation that is activated at the time of burst arrival OBS can provide a convenient way for Qo. S by providing extra offset time 4/27/2001 EE 122, UC Berkeley 38
Qo. S using Offset-Times Assume two classes of service Class 1 has higher priority Class 2 has zero offset time t o 1 i Time ta 1 ta 2(= ts 2) ts 1 ta 2(= ts 2) t s 1+ l 1 t o 1 i Time ta 2(= ts 2) tai = arrival time for class i request tsi = service time for class i request 4/27/2001 ta 1 t s 2+ l 2 ts 1 t s 1+ l 1 toi = offset time for class i request li = burst length for class i request EE 122, UC Berkeley 39
Comparison 4/27/2001 EE 122, UC Berkeley 40
Hierarchical Optical Network E/O E/O Optical MAN All O E/O OS OS All O OS E/O OS OS WC E/O E/O All O Optical MAN E/O Optical MAN Optical Core E/O E/O 4/27/2001 EE 122, UC Berkeley 41
Hierarchical Optical Network n Optical MAN may be n n n Packet Switched (feasible since lower speeds) Burst Switched Sub- circuit switching by wavelength merging Interfaces boxes are All-Optical and merge multiple MAN streams into destination-specific core stream Relatively static Optical Core Control distributed to intelligent edge boxes 4/27/2001 EE 122, UC Berkeley 42
Outline n n Why Optical? Intro to Optical Hardware Three generations of Optical Various Switching Architectures n n Circuit, Packet and Burst Protection and Restoration 4/27/2001 EE 122, UC Berkeley 43
Link vs Path Protection n For failure times, need to keep available s on backup path Link: Need to engineer network to provide backup Path: need to do end-to-end choice of backup path 4/27/2001 EE 122, UC Berkeley 44
Types of Protection n Path protection n 4/27/2001 n Dedicated (1+1) – send traffic on both paths Dedicated (1: 1) – use backup only at failure Shared (N: 1) – many normal paths share common backup EE 122, UC Berkeley Link Protection n n Dedicated (each is also reserved on backup link) Shared (a on backup link is shared between many) 45
Restoration n n Do not calculate protection path ahead of time Upon failure, use signalling protocol to generate new backup path Time of failover is more But much more efficient usage of s Need also to worry about steps to take when the fault is restored 4/27/2001 EE 122, UC Berkeley 46
Protection and Restoration n Time of action n n Path calculation (before or after failure ? ) Channel Assignments (before or after failure ? ) OXC Reconfiguration AT&T proposal n n n 4/27/2001 Calculate Path before failure Try channel assignment after failure Simulations show 50% gain over channel allocation before failure EE 122, UC Berkeley 47
Protection Algorithms n Various flavors n n n n Shortest path type Flow type ILP (centralized) Genetic programming In general, centralized algos are too inefficient Need distributed algos, and quick signalling Have seen few algos that take into account the different node types (LWC/FWC) 4/27/2001 EE 122, UC Berkeley 48
Conclusion n n Optical is here to stay Enormous gains in going optical O/E/O will soon be the bottleneck Looking for ingenious solutions n n 4/27/2001 Optical Packet Switching Flavors of Circuit Switching EE 122, UC Berkeley 49
Collective References n n n “Optical Networks: A practical perspective” by Rajiv Ramaswami and Kumar Sivarajan, Morgan Kaufman. IEEE JSAC n September 1998 issue n October 2000 issue IEEE Communications Magazine n March 2000 issue n September 2000 issue n February 2001 issue n March 2001 issue 4/27/2001 n n n INFOCOM 2001 n ‘Optical Networking’ Session n ‘WDM and Survivable Routing’ Session INFOCOM 200 n ‘Optical Networks I’ Session n ‘Optical Networks II’ Session RFC 2702 for MP S www. cs. buffalo. edu/pub/WWW /faculty/qiao/ www. lightreading. com EE 122, UC Berkeley 50
- Slides: 50