1 Optical switching Figures from http://www.leehansen.com/clipart/seasons/spring/pages/rainbow.htm and Lucent
2 References Keshav & Varghese doesn t cover this (except brief reference on K p. 15) :-( Other courses: e.g. ELEC9350/9355 (optical fibres/comms), PHTN4662 (Photonic Networks) Overviews: M. Maier and M. Reisslein: Trends in Optical Switching Techniques: A Short Survey, IEEE Network, 22(6):42-7, Nov.-Dec. 2008 Cisco s Fundamentals of DWDM Technology http://www.cisco.com/univercd/cc/td/doc/product/mels/dwdm/dwdm_ovr.htm D. Blumenthal et al: 'Photonic packet switches: Architectures and experimental implementations', Proc. IEEE 82(11):1650-67, 1994 M. Borella: 'Optical components for WDM lightwave networks', Proc. IEEE 85(8):1274-307, 1997 LightReading.com Industry news Later lecture on MPLS will discuss GMPLS for optical routing control & connection management
3 Easy-going Wikipedia pages http://en.wikipedia.org/wiki/multi-mode_optical_fiber http://en.wikipedia.org/wiki/single-mode_optical_fiber http://en.wikipedia.org/wiki/transverse_mode http://en.wikipedia.org/wiki/optical_amplifier http://en.wikipedia.org/wiki/wavelength-division_multiplexing http://en.wikipedia.org/wiki/tunable_laser http://en.wikipedia.org/wiki/routing_wavelength_assignment_%28rwa%29 http://en.wikipedia.org/wiki/arrayed_waveguide_grating http://en.wikipedia.org/wiki/four_wave_mixing http://en.wikipedia.org/wiki/optical_switching http://en.wikipedia.org/wiki/optical_burst_switching http://www.opticsjournal.com/tunablelasers.htm
4 References about specific technologies P. Ball: The next generation of optical fibers, Technology Review, 104(4):55, 7 pgs, May 2001 J. Kiniry: 'Wavelength Division Multiplexing: Ultra high speed fiber optics', IEEE Internet Comp. 2(2):13-5, 1998 X. Ma and G.-S. Kuo: 'Optical switching technology comparison: optical MEMS vs. other technologies', IEEE Comm. Mag. 41(11):S16-S23, Nov. 2003 P. Chu et al: 'MEMS: the path to large optical crossconnects', IEEE Comm. Mag. 40(3):80-7, 2002 S. Yao et al: 'Advances in photonic packet switching: an overview', IEEE Comm. Mag. 38(2):84-94, 2000 T. Battestilli and H. Perros: 'An introduction to Optical Burst Switching', IEEE Optical Communications 41(8):S10-5, Aug. 2003 H. Zang et al: A review of routing and wavelength assignment approaches for wavelength-routed optical WDM networks, Optical Networks, Jan. 2000.
5 Outline Optical components fibres amplifiers transmitters, receivers Optical switching overview Wavelength-routed networks Optical demultiplexing / filtering Switch construction Wavelength assignment problem & conversion All-wavelength switching (MEMS) Photonic packet switching Dealing with contention: Deflection routing
6 Optical fibre Fibre consists of: Glass (silica) core: 2-125μm in diameter. Glass cladding: Thickens fibre, making it stronger. n cladd < n core => internal reflection Rays ( modes ) of light are confined to fibre by total internal reflection Using a thin core restricts propagation to one mode only. Such single mode fibre offers higher data rates/distance than multi-mode fibre (but is more expensive) Plastic jacket: Protection Photo from Kurose & Ross. T Fig. 2.5 and 2.7
7 Evaluation of optical fibre Fibre doesn t conduct electricity Often need separate cable to provide power to devices. Good for insulating, e.g. protect computer from lightning striking outdoors antenna by connecting with fibre comms cables won t spread high voltages from failing machines in factories Immune to electromagnetic noise Low attenuation for wide bandwidth Note: Optical networks are not fast because the speed of light is fast (indeed, in glass fibre it is comparable to the speed of electrical signals) that would only affect propagation delays. They are fast because of their wide bandwidth high transmission rates.
8 Fibre Attenuation in silica 0.85μm band (cheap LEDs) 1.3μm and 1.5μm bands (lasers) approximate wavelengths of visible light 167 THz 121 THz Bandwidth of tens of THz (tens of Tb/s with simple modulation) Monochrome figure from Cisco Monochrome figure from Cisco http://www.cisco.com/univercd/illus/4/87/48087.gif
9 Erbium Doped Fibre Amplifiers (EDFAs) Amplify (30dB+) signals (including noise, no regeneration) Cover a broad bandwidth, e.g. 35nm Operation: pump @ 980 or 1480nm, signal @ 1525-1560nm Switching relevance: Many fabrics exhibit high loss amplify with EDFA after switching EDFA forms a basic electronically-controlled on/off gate for switching EDFAs are the original transparent all-optical network element for trans-oceanic links motivate all-optical networks & all-optical switching.
10 Mismatch between fibre and transceivers 40Tb/s of fibre >> 10Gb/s optoelectronics (today) + fibre is often scarce => use multiple optoelectronic systems tx/rx @ different λs => Wavelength Division Multiplexing (optical equivalent of FDM in the RF domain) Wavelengths = lambdas (λ) (sometimes spelled lamda) Figure from from Cisco Cisco
11 WDM technology DWDM = Dense WDM, e.g. 0.8nm spacing => about 64λs (Coarse DWM: 12λs, Ultra Dense WDM >100λs) Example of state of the art (world record) [NTT press release, 2006sep29]: 140 channels @ 111 Gb/s (14Tb/s) Tunable transmitters (multiple or tunable laser) Tuning times range from ns (Distributed feedback) to ms (mechanical/acousto) Tunable receivers Burst mode receivers can synchronize quickly, unlike continuous mode receivers. Figure from Cisco
12 Broadcast and select network using tunable tx/rx Combiner Fibre Splitter Tunable lasers Broadcast medium Tunable filters Photodiodes poor signal power because of broad splitting Note that a splitter doesn t demultiplex or switch: input is sent to all outputs.
13 Outline
14 Optoelectronic networks Simplest use of optics to networking merely involves replacing wired links with fibre. Where link joins switch, line interface card translates signal between optical and electronic forms. Expensive optoelectronic parts are replicated (on switch interfaces, not just terminal interfaces) Electronics (particularly energy flow: power in, heat out) limits switch throughput
15 Motivation for all-optical networks Benefits: High transmission speed (switching speed may not be as impressive, e.g. Gb/s throughput on each port, but reconfiguration (changing port mappings) only once per ms) Eliminate optoelectronics from network Low cost network Transparency: Payload rate and format can change (with OE progress) without changing network elements. Network must recognize control rate and format for packet switching. Particularly for trans-oceanic cables original motivation for EDFA amplifiers Costs: Add optical switching equipment Analog: Signals are amplified, not regenerated Error rate monitoring is hard
16 Where are optical networks used? LANs: Rarely is optical switching needed (exception: CERN,?NSA?!) Access networks: Passive Optical Networks (PONs) (see next slide) Metropolitan Area Networks Traditionally SDH optoelectronic rings (s.t. can adapt around single failure) Uncertainty about provisioning rings (e.g. how much BW needed?) can be accommodated by subdividing rings & adding lambdas to serve new rings. Long haul (e.g. inter-city): Optical transmission preferable because of high bandwidth for low attenuation. Core networks (e.g. switch in Sydney switching traffic between Melbourne, Canberra, Brisbane): High transmission speeds But reconfiguration needn t be frequent.
17 Passive Optical Networks (PONs) Why?: Higher rates to subscribers than wire Options: Individual fibres from Exchange to N subscribers N fibres to Exchange (Central Office CO) 2N transceivers Fibre to the curb, switch to fibres to subscribers (a) Point-to-point network N fibers 2N transceivers 1 fibre to Exchange 2N+2 transceivers Either optoelectronic switch (may fail, CO (b) Curb-switched network 1 fiber 2N+2 transceivers expensive optoelectronic tx/rx) or photonic switch (advanced technology) Fibre curb + passive optical splitter to subs 1 fibre to Exchange N+1 transceivers CO (c) Passive optical network 1 fiber N transceivers Power loss from splitting limits fan-out; need to deal with shared medium (security, MAC) Terms: Fibre To The x (FTTx): Home, Building, Premises, Curb, Node,... CO L km L km L km Curb switch Passive optical splitter N subscribers N subscribers N subscribers Image from slides by Vijay Sivaraman For more, see www.ponforum.org
Repeat of previous slide without animation, e.g. to facilitate printing Passive Optical Networks (PONs) 18 Why?: Higher rates to subscribers than wire Options: Individual fibres from Exchange to N subscribers N fibres to Exchange (Central Office CO) 2N transceivers Fibre to the curb, switch to fibres to subscribers (a) Point-to-point network N fibers 2N transceivers 1 fibre to Exchange 2N+2 transceivers Either optoelectronic switch (may fail, CO (b) Curb-switched network 1 fiber 2N+2 transceivers expensive optoelectronic tx/rx) or photonic switch (advanced technology) Fibre curb + passive optical splitter to subs 1 fibre to Exchange N+1 transceivers CO (c) Passive optical network 1 fiber N transceivers Power loss from splitting limits fan-out; need to deal with shared medium (security, MAC) Terms: Fibre To The x (FTTx): Home, Building, Premises, Curb, Node,... CO L km L km L km Curb switch Passive optical splitter N subscribers N subscribers N subscribers Image from slides by Vijay Sivaraman For more, see www.ponforum.org
19 Optical switching applications Figure from LightReading OADM = Optical Add-Drop Multiplexer OXC = Optical Cross-Connect
20 Optical switching technologies Figure from LightReading MEMS = MicroElectroMechanical Systems
21 How the optical domain differs High-rate transmission Re-evaluate decisions made in electronic networks Processing e.g. optical address matching Buffering Electronic: Packet switching uses processing (routing decisions) to save transmission capacity Optical: Circuit switching is currently more cost effective Electronic: Route along shortest path (to save transmission) and buffer when contention on that path (buffering is cheap). Optical: Deflection routing: Use longer path to avoid buffering
22 Outline
23 Wavelength routed networks Motivated by WDM not all wavelengths may have same destination. Directing signal propagation alleviates power problems of broadcast-and-select network. Components are also called wavelength selective
24 Optical (de)multiplexing Multiplexing is trivial: Combine signals by splicing fibres together Optical Add-Drop Multiplexer (OADMs): Drop wavelengths (demux) Add wavelengths (mux) Many access ports; few wavelengths change Figure from Cisco
25 Static demultiplexing... Static demultiplexers: Physical configuration assigns λs to ports. Generally require precise physical alignment, e.g. by using photolithography - need to assess thermal stability... using refraction: Figure from Cisco
26... using diffraction Diffraction grating Diffraction grating Figure from Cisco
27... using interference Arrayed Waveguide Grating (AWG) aka optical waveguide router! up to 250 ports Input cavity: splits light into waveguides Waveguides have differing lengths => delays Delayed waveforms interfere in output cavity Position output ports @ points of max. interference e.g. Cisco ONS 15454 based on Reconfigurable Optical Add Drop Mux (ROADM) using Silicon Array Waveguide Grating (AWG) Figure from Cisco
28 Photo from Y. Hibino: An array of photonic filtering advantages: arrayed-waveguide-grating multi/demultiplexers for photonic networks; IEEE Circuits and Devices Magazine, 16(6):21-7, Nov. 2000
29... using (interference) filters Interference filters transmit light of one wavelength, and reflect others. Good thermal stability (variation of vertical/horizontal dimensions handled by filters having vertical range) and isolation between channels, at moderate cost. High loss, e.g. (yellow) wavelength that is reflected many times. Figure from Cisco
30 Controllable refractive indices Control the refractive index of the medium by applying: Electric field (electro-optic) e.g. Lithium Niobate (LiNbO 3 ) or liquid crystal ns switching times substantial loss Sound (acousto-optic), 10us switching times Heat(thermo-optic), ms switching times Silica has lower optical loss than polymer, but requires more heat and conducts heat more (=> switch requires more space) Form basic switching element using interferometers...
31 Interferometeric filters/gates delay + = Mach-Zehnder interferometer: Controllable delay element determines relative phase of combined wavelengths. 180 o phase difference => filtered out + = Drawbacks: Few ports (2 2) Crosstalk between outputs
32 Building a switch from filters/gates Splitter Combiner Filter Power budget: Each input is split between n outputs (like broadcast&select) (n=2 in this example) Combiner has only one active input => readily amplify using EDFA
33 Wavelength assignment problem Can concurrently use the same wavelength in spatially separate areas of the network Q: What λ should be used for each lightpath? A: Complex optimisation problem. Wavelength assignment can be simplified with expensive wavelength converters.
34 Wavelength converters Primitive: Perform opto-electronic-opto conversion, and use different λ for output laser Better: Electrical signal directly modulates laser Best: Optical conversion using coherence effects From non-linear response of medium in the presence of multiple waves. e.g. Four-Wave-Mixing, Difference Frequency Generation
35 Outline
36 MicroElectroMechanical Systems It s all done with mirrors (MEMS) Mechanical => switching time 4 I => ms switching times (too slow for burst or packet switching) Loss of approx 1.25dB / mirror Photos from Lucent Photos from Lucent
37 MEMS Implemented on silicon using photolithography precise positioning of mirrors can integrate mechanical system with control electronics density should advance with progress in photolithography technology (driven by VLSI) Current densities of 256-1024 mirrors We have built MEM mirrors on an 8-in. wafer with a million MEM mirrors on it, each individually moveable Jeffrey Jaffe, president of research and advanced technologies at Bell Labs in J. Ribeiro: 'Bell Labs grapples with VoIP, open-source', Jan. 05 Photo from Lucent
38 MEMS switches 2D 3D Figures from Chu02
39!!!
40 Other switching technologies Bubble (see Agilent Video) Liquid crystal attenuators/switches Liquid crystal gratings Holograms Wavelength-selective technologies electro/thermo/acousto-optics
41 Outline
42 Electro-optic packet switching Payload travels over a lightpath from source to destination; no optoelectronic conversion within network switching elements. Control information may incur optoelectronic conversion, may even flow on auxiliary electronic control network. Payload Control Figure from Blumenthal94 Distinct from optoelectronic
This slide only omits the animation of its predecesor to aid printing Electro-optic packet switching Payload travels over a lightpath from source to destination; no optoelectronic conversion within network switching elements. Control information may incur optoelectronic conversion, may even flow on auxiliary electronic control network. 43 Payload Control Figure from Blumenthal94 Distinct from optoelectronic
44 Outline
45 Ways to deal with contention Time domain: optical buffering Space domain: deflection routing Frequency domain: Wavelength conversion Creates more options, but may still have contention for specific frequency on output port.
46 Optical buffering Long loops of fibre form delay lines How long? spool of 0.9km of fibre Electricity travels a foot in a nanosecond. G. Hopper In vacuum v 0 =3E8 m/s In silica: slower than in a vacuum: n SiO2 =1.5, v SiO2 =2E8m/s e.g. 1500B Ethernet frame (12kb) @ 1Gb/s (1ns/b) = 12us delay = 2.4km Information continuously physically moves (c.f. electronic buffers) Memory provides sequential, not random, access But that s fine for queueing/buffering Photo from www.go4fiber.com/bargain/image/products/50ss.jpg
47 Optical buffering: Problems Attenuation in delay lines Complexity (to implement variable delay ) Some say: unnecessary modeled on electronic networks
48 Optical buffer implementation Fibre delay line Capacity length loss Programmable fibre delay line Enables control of holding time. Loss from splitting signal Active recirculating delay line Enables control of holding time. Amp noise limits recirculations Figure from Blumenthal94
49 Optical buffering for TSI Switching time 1 2 3 4
50 Outline
51 Deflection routing Route some inputs ( ) on shortest path Deflect others to other outgoing ports (longer paths) Links between switches effectively become passive delay line buffers for deflected packets Often prioritise packets: Better service for originally high priority packets, Escalate priority after deflect s.t. not deflected endlessly. Assume that the ports are bidirectional but can only support one flow in either direction
52 Deflection routing: Evaluation Reduces network throughput because packets take longer paths Manhattan street network (regular mesh) & deflection: 55-70% of throughput with infinite buffering. May later revert to shortest path => mis-sequence packets Deflecting towards source further increases path length, but creates backpressure & may provide congestion control
53 Summary of optical switching Switching at the physical layer Fibre enables massive transmission capacity, but optical processing and buffering are difficult. WDM allows many flows to share one fibre Various types of demultiplexers based on refraction, diffraction, interference MicroElectroMechanical Systems: Slow reconfiguration, but no optoelectronic conversion bottleneck Packet switching is possible, but is it out of context?