Technology and Architectural Approaches to Address Continued Explosive Growth in Network Traffic Jane M. Simmons

Transcription

1 International Conference on Computing, Networking and Communications (ICNC 2017) Technology and Architectural Approaches to Address Continued Explosive Growth in Network Traffic Jane M. Simmons Monarch Network Architects January 28, 2017 Santa Clara, CA Copyright 2017 Monarch Network Architects LLC

2 Overview Explosive Growth in Network Traffic Huge Growth in System Capacity Impending Fiber Capacity Limits Technology Approaches to Address Fiber Capacity Limits Architectural Approaches to Address Fiber Capacity Limits Focus is on Transmission in the Long-Haul Network Expanded and updated version of: Saleh and Simmons, "Technology and Architecture to Enable the Explosive Growth of the Internet," Communications Magazine, Vol. 49, No. 1, Jan. 2011, pp Copyright 2017 Monarch Network Architects LLC J.M. Simmons 2

3 Explosive Growth in Network Demand Traffic Volume (PB/mo) Traffic Volume (PB/mo) If traffic growth is exponential If traffic growth rate is hyperbolic G(y) = 10 a (y y o ) b Actual Global Internet Traffic Growth a = 3, b = -1.14, y o = 1994 Traffic Growth is Best Modeled by a Hyperbolic Compound Annual Growth Rate (CAGR) Explosive Growth, not Exponential Growth Current CAGR between 20 and 25% S. Korotky, Bell Labs Technical Journal, Vol. 18, No. 3, Dec Lucent Technologies Inc. Copyright 2017 Monarch Network Architects LLC J.M. Simmons 3

4 Accompanied by Huge Growth in System Capacity 1995: Systems from this timeframe supported 8 wavelengths of 2.5 Gb/s on one fiber 20 Gb/s Total Bandwidth Today: 80 wavelengths of 100 Gb/s on one fiber 8 Tb/s Total Bandwidth Factor of 400 increase in system capacity in ~20 years Enabled by advances in: Modulation formats Transmitter / Receiver technology (advanced signal processing & high-speed electronics) Filtering technology Forward Error Correction (FEC) Accompanied by more than 3 orders of magnitude decrease in cost per Gb/sec km ~$1,000/(Gb/sec km) less than $0.50/(Gb/sec km) (Backbone Network) Can these trends regarding system capacity and cost continue? The capacity of a fiber has always been viewed as being almost infinite But Copyright 2017 Monarch Network Architects LLC J.M. Simmons 4

5 Fiber Capacity Isn t Infinite After All! System Spectral Efficiency metric The ratio of the information bit-rate to the total bandwidth consumed Current state-of-the-art systems: 100 Gb/s with 50 GHz wavelength spacing and dual-polarization Overall spectral efficiency is 2 bits/s/hz Theoretical analysis of the maximum supportable capacity on conventional fiber Assumptions: Raman amplification Single polarization Optical reach of 2,000 km (system capacity increases as the optical reach decreases) Optical reach is the distance an optical signal can travel before it degrades to a level that necessitates it be regenerated Essiambre et al., Capacity limits of optical fiber networks, J. of Light. Tech., February 15, 2010, pp IEEE Conclusion: Conventional-fiber Spectral Efficiency limit is 4 to 6 bits/s/hz per polarization Systems likely to be dual polarization Max Spectral Efficiency of 8 to 12 bits/s/hz Today s state-of-the-art systems are within a factor of ~5 of the limit! Copyright 2017 Monarch Network Architects LLC J.M. Simmons 5

6 How Should We Address This Impending Limit? Assuming a 20% compound annual growth rate, a 5 growth in traffic will occur in about 9 years Something new is needed! The end-game is not simply providing more capacity The solution(s) must be efficient with respect to cost, power consumption & space Technology approaches Expand into other regions of the spectrum Space Division Multiplexing (SDM) Multiple fiber-pairs per link Multiple cores per fiber Multiple modes per fiber Architectural approaches Improved packing of IP traffic Asymmetric traffic Caching Multicasting Dynamic networking Gridless/Elastic optical networks Optical reach vs. capacity For each approach, we discuss the possible benefits and the implementation challenges Some of the architectural approaches are dependent upon new technology Copyright 2017 Monarch Network Architects LLC J.M. Simmons 6

7 Technology Approaches for Addressing Fiber Capacity Limits Copyright 2017 Monarch Network Architects LLC J.M. Simmons 7

8 Expand into Other Regions of the Spectrum Fiber Loss (db/km) S-Band C-Band L-Band Wavelength (nm) C-band = central or conventional band S-band = shorter-wavelength band L-band = longer-wavelength band Currently, most systems use the C-band (and perhaps part of the L-band or S-band) Expand further into S and/or L bands to increase the system capacity Perhaps will increase system capacity by a factor of 2 to 3 Desired components: An optical amplifier that can operate over the whole expanded system Requiring multiple amplifiers will not improve on the system economics Transponders that are tunable over the whole spectral region Copyright 2017 Monarch Network Architects LLC J.M. Simmons 8

9 Space Division Multiplexing (SDM) Option 1: Multiple Fiber Pairs Per Link Light up N fiber-pairs per link instead of one fiber-pair Increases the capacity of a link by a factor of N Does not require new fiber types (in contrast to the SDM options that are discussed next) Two architectural options: Deploy parallel systems, where N ROADMs are deployed at a node Maintain a single system, where all fiber pairs feed into one ROADM at a node This may lessen wavelength contention Not all links have to be upgraded to multi-fiber at once But the required ROADM degree may become very large Disadvantages Does not improve on the system economics (i.e., cost per bit/sec) or the system power consumption (i.e., energy per bit) E.g., The number of optical amplifiers increases by N (However, there is ongoing research into developing amplifiers that can amplify signals in more than one fiber) Copyright 2017 Monarch Network Architects LLC J.M. Simmons 9

10 Space Division Multiplexing (SDM) Option 2: Multiple Cores Per Fiber Multicore Fiber (MCF) Current conventional fibers have a single core (SCF) MCF: Increase the number of cores to N Benefits The system capacity theoretically increases by N Significant improvement in system economics and power consumption Amplification of all cores with a single amplifier has been demonstrated The cores are tightly packed and can be processed together A single connector can interconnect multicore fibers rather than requiring one connector per core Concerns and Challenges Need to install new fiber Cross-talk between the cores; Tradeoff between low cross-talk and high core density Various techniques to reduce cross-talk (see next slide) Non-linear impairments worsen with more cores (reduces the optical reach) Due to the accompanying decrease in fiber effective area Two types of MCF systems Weakly-coupled: Signals in the cores remain uncoupled can route the signals in the cores independently. Perhaps 12 cores (or more?) in a long-haul network. Strongly-coupled: Signals in the cores become coupled as the number of cores increases e.g., the l i s in each one of the cores must be routed together Pay attention to the achievable distance Some experimental systems with many cores have an optical reach of 10 s of kms not suitable for long-haul transmission Copyright 2017 Monarch Network Architects LLC J.M. Simmons 10

11 Multi-Core-Fiber Cross-Section Examples 2009: Cross-section of a fiber with 7 cores ~100 km K. Imamura et al., Multi-core holey fibers for the long-distance (>100 km) ultra large capacity transmission, OFC/NFOEC OSA 2013: Cross-section of a fiber with 12 cores with Trenches Saitoh et al., JLT, Jan. 1, IEEE 2013: Cross-section of a fiber with 12 cores, Bidirectional ~1500 km T. Kobayashi et al., ECOC ECOC 2015: Cross-section of a fiber with Heterogeneous cores; 30 cores Remains Weakly Coupled at 100 km Saitoh et al., JLT, Jan. 1, IEEE Copyright 2017 Monarch Network Architects LLC J.M. Simmons 11

12 Space Division Multiplexing (SDM) Option 3: Multiple Modes Per Fiber Mode Division Multiplexing Current conventional long-distance fibers have a single mode (SMF) Consider using fiber with multiple modes: Few-Mode Fiber (FMF) Small number of modes (e.g., 3 or 6), not hundreds of modes Benefits The system capacity theoretically increases in proportion to the number of modes Improvement in system economics Amplification of all modes with a single amplifier has been demonstrated Amplified 5 modes with one EDFA (Genevaux et al., JLT, Jan. 2016) Concerns and Challenges Need to install new fiber The modes become coupled together Requires electronic multiple-input multiple-output (MIMO) processing at the receiver May consume a significant amount of power Minimizing inter-mode cross-talk allows less complex MIMO DSP to be used All modes of a given wavelength need to be routed together; a ROADM cannot drop a subset of the modes As with MCF, need to pay attention to the achievable optical reach Copyright 2017 Monarch Network Architects LLC J.M. Simmons 12

13 Are Coupled Spatial Carriers Good or Bad? The modes (and possibly the cores) of a fiber will be coupled together Example: if there are 6 modes, then all 6 modes of l1 will be coupled together A ROADM will not be able to pick out just one of the modes All modes will likely be routed as an inseparable unit, whether desired or not Glass Half Full View: Coupling results in Spatial Super-Channels Allows for efficient transmission of very high bandwidth connections Example: A 600 Gb/s connection can be carried as one Gb/s super-channel More cost-effective ROADMs and transponders One MEMS mirror can steer all of the wavelengths in a superchannel Component sharing across a superchannel in the transponders Glass Half Empty View: Coupling is similar to Wavebands in standard SMF Waveband is a group of wavelengths routed as a bundle Wavebands were first proposed in the late 1990s to reduce switching costs Wavebands require more complex algorithms to use the bandwidth efficiently Service providers did not like wavebands 20 years ago (due to loss of flexibility) Will they like coupled modes / cores any better? May not have a choice! With wavebands, traffic analyses were performed to determine optimal band size With coupled modes, the size is determined by the limits of the technology Copyright 2017 Monarch Network Architects LLC J.M. Simmons 13

14 Space Division Multiplexing (SDM) Option 4: Multiple Cores and Multiple Modes Per Fiber Multiple Modes can be Combined with Multiple Cores (FM-MCF) Multiplicative capacity increase Inter-core cross-talk increases with higher-order modes To maintain weak coupling among cores, larger inter-core distance is required Fewer cores Example: 12 cores, each with 3 modes (527-km reach) (Shibahara et al., OFC 2015) The potential capacity benefits of multi-core and/or few-mode fiber are very large But a lot of implementation challenges still need to be addressed Unlikely to be deployed in the near-term Copyright 2017 Monarch Network Architects LLC J.M. Simmons 14

15 Architectural Approaches for Addressing Fiber Capacity Limit Copyright 2017 Monarch Network Architects LLC J.M. Simmons 15

16 Architectural Approaches for Addressing Capacity Limits Architectural strategies do not increase the system capacity to handle more traffic In contrast to the technology options Rather, they reduce the effective traffic load so that more traffic can be carried A variety of architectural approaches can be utilized to reduce capacity requirements Improved Packing of IP Traffic Asymmetric Traffic Caching Multicasting Dynamic Networking Gridless/Elastic Optical Networks Trade off optical reach vs. capacity Most of these approaches are likely more feasible in the near-term than the technology options that were presented For each approach, we estimate the capacity benefits and the amount of traffic in the network that can take advantage of the approach Copyright 2017 Monarch Network Architects LLC J.M. Simmons 16

17 Architectural Approaches for Addressing Capacity Limits: Improved Flow Management in IP Networks Reduce the need for excessive headroom when packing IP traffic onto wavelengths Some carriers run IP wavelengths at ~35% fill, to allow for burstiness of IP traffic and rerouting during failures Large pipes (e.g., 40 Gb/s,100 Gb/s) carry a large number of flows, which allows for smoother statistical multiplexing of the traffic Increase fill of IP wavelengths to more than 65% Still need some headroom for restoration from failures Estimate a 2 Benefit in Capacity Estimate 80% of the traffic is carried in IP Packets Remainder may be, for example, wavelength services carried directly in the optical layer Copyright 2017 Monarch Network Architects LLC J.M. Simmons 17

18 Peak Traffic Architectural Approaches for Addressing Capacity Limits: Asymmetric Connections Typically, symmetric connections are established in a network X Gb/s from Node A to Node Z, and X Gb/s from Node Z to Node A However, many applications are asymmetric X Gb/s from Node A to Node Z, and Y Gb/s from Node Z to Node A, where Y X Video distribution, data backup, distributed data processing, etc. AT&T Study 8 Weeks of IP Data Approximately a 2:1 asymmetry ratio network-wide IP Link Direction with Higher Peak Traffic IP Link Direction with Lower Peak Traffic S. L. Woodward et al., Asymmetric optical connections for improved network efficiency, JOCN, vol. 5. no. 11, Nov Optical Society of America IP Link Establish asymmetric connections accordingly (e.g., 10 Gb/s A to Z, 5 Gb/s Z to A) Will require changes to provisioning systems, element management systems, etc. Equipment savings are possible as well, in addition to capacity savings Transponders typically include both a transmitter (Tx) and a receiver (Rx) (or N of each) If Tx s and Rx s are deployed on separate cards, then some number of Tx s and Rx s can be removed Separating the Tx and Rx is likely less efficient in terms of cost, power, space (on a normalized basis) However, if enough Tx and Rx can be removed, the overall benefit may be positive Estimate a 1.3x Benefit in Capacity Estimate this applies to 80% of the traffic Copyright 2017 Monarch Network Architects LLC J.M. Simmons 18

19 Architectural Approaches for Addressing Capacity Limits: Multicasting Typically, unicast connections are established in a network One source, one destination Some applications are inherently multicast in nature; e.g., video distribution More efficient to route the multicast traffic over one tree rather than over multiple unicast connections Assume that Node A is transmitting the same data to Nodes W, X, Y, and Z E A E A B B F C D F C D W X Y W X Y G H Z 4 Separate Unicast Connections G H One Multicast Tree Z Backbone network study comparing capacity requirements with a multicast tree vs. multiple unicast connections: If number of destinations is uniformly distributed between 5 and 15: factor of ~3 savings If number of destinations is uniformly distributed between 2 and 6: factor of ~1.5 savings Estimate a 2x Benefit in Capacity Estimate this applies to 20% of the traffic Copyright 2017 Monarch Network Architects LLC J.M. Simmons 19

20 Architectural Approaches for Addressing Capacity Limits: Caching Distributed caching via Content Distribution Networks (CDN) Content is stored on multiple servers to be closer to the consumers of the content Fewer connections are needed; connections are routed over shorter distances Lower latency is an additional benefit Percentage of traffic that is distributed via CDNs is increasing Cisco Visual Networking Index 2013: 34% percent of global Internet traffic crossed CDNs in 2012 Cisco Visual Networking Index 2016: 45% percent of global Internet traffic crossed CDNs in % percent of global Internet traffic will cross CDNs in 2020 Caching algorithm improvements increase the probability that the desired data are stored on a nearby server A study by AT&T estimates CDNs reduce capacity requirements by a factor of 3 as compared with content distribution via a centralized server A. Gerber and R. Doverspike, Traffic types and growth in backbone networks, OFC/NFOEC 11, Paper OTuR1 Estimate a 3x Benefit in Capacity Estimate this applies to 50% of the traffic Copyright 2017 Monarch Network Architects LLC J.M. Simmons 20

21 Architectural Approaches for Addressing Capacity Limits: Dynamic Optical Networks Networks have historically been quasi-static; connections remain in place for months / years Networks have gradually become more configurable Operations personnel initiate the provisioning process Connections are established remotely through software control Takes advantage of flexible technology such as ROADMs and tunable transponders The next step in this evolution is Dynamic Networking Deliver bandwidth when & where needed, instead of having a fixed pipe Connections are rapidly established and torn down without the involvement of operations personnel The higher layers of the network automatically request bandwidth from the optical layer, which is then reconfigured accordingly; completely under software control Connections are provisioned and torn down in seconds, or possibly sub-seconds Software Defined Networking (SDN) may enable more dynamic networking There are still challenges Finding the correct balance between centralized and distributed operation Dealing with resource contention and latency. Tradeoff with optimization. Strategies for pre-deployment of equipment: Cost vs. blocking Backbone network study compared capacity requirements using a dynamic vs. a static optical layer A subset of the traffic was assumed to have ON/OFF cycles, where it was ON 10% of the time If this traffic subset is 25% of the total traffic: factor of ~5 savings (for this traffic) If this traffic subset is 10% of the total traffic: factor of ~4 savings (for this traffic) Estimate a 4x Benefit in Capacity Estimate this applies to 20% of the traffic Copyright 2017 Monarch Network Architects LLC J.M. Simmons 21

22 Architectural Approaches for Addressing Capacity Limits: Gridless/Elastic Optical Networks Assign spectral resources to meet the required rate of the carried services The spectrum on each fiber is sliced up as needed: Optical-layer Grooming Example: a 15 Gb/s connection is assigned 15 Gb/s worth of spectrum as opposed to grooming the connection with other connections to fill up a 100 Gb/s wavelength Allow the assigned spectral resources to grow or shrink: Elastic Network If the connection now requires 20 Gb/s as opposed to 15 Gb/s, the spectrum allotted to that connection expands (assuming the spectrum is available) Do not adhere to the 50-GHz ITU-T grid: Gridless Architecture Jinno et al., Spectrum-Efficient and Scalable Elastic Optical Path Network: Architecture, Benefits, and Enabling Technologies, IEEE Communications Magazine, November IEEE Copyright 2017 Monarch Network Architects LLC J.M. Simmons 22

23 Gridless/Elastic Optical Networks Pose Numerous Management and Practical Challenges Adds network management complexity Need to track how the spectrum is sliced up on each fiber Routing and Spectrum Assignment (RSA) vs. Routing and Wavelength Assignment (RWA) May lead to stranded spectral resources (especially in a dynamic network) The available spectrum may consist of narrow, non-contiguous spectral regions May limit opportunities for optical bypass (especially in a dynamic network) The spectrum may be sliced up differently on each fiber entering a ROADM Need to Defragment the network to create larger blocks of free spectrum Defragmentation involves modifying live connections; needs to be done carefully Need guardbands between each optical slice Wasted bandwidth Need bandwidth-variable transponders (BVTs) and bandwidth-variable ROADMs Likely not problematic: e.g., transponders based on OFDM subcarriers, ROADMs using LCoS Many bandwidth-variable transponders are required If transponders transmit/receive just one connection, then many more transponders are required as compared to a standard grid-based system Alternative: Multi-Flow Transponders one transponder can support multiple independent connections The granularity of the optical slices cannot be arbitrarily fine E.g., Filtering limitations in the ROADM Electronic grooming is still needed for the relatively low-rate connections (e.g., 10 Gb/s) Estimate a 1.5x Benefit in Capacity Estimate this applies to ~100% of the traffic Copyright 2017 Monarch Network Architects LLC J.M. Simmons 23

24 Architectural Approaches for Addressing Capacity Limits: Trade Off Optical Reach vs. Capacity State-of-the-art transponders typically include powerful Digital Signal Processing (DSP) Example: DSP required for coherent detection, a key enabler of 100 Gb/s transmission Take advantage of the DSP to enable software-adaptable transponders Adaptable transponders can trade off bandwidth versus optical reach Short connection Requires reduced optical reach Utilize less bandwidth to carry it Use a modulation format with higher spectral efficiency so that less bandwidth is required Line Rate Modulation Format Bandwidth Optical Reach 100 Gb/s DP-BPSK 75 GHz 3,800 km 100 Gb/s DP-QPSK 50 GHz 2,500 km Use the same modulation format but with reduced bandwidth DP-BPSK Dual Polarization Binary Phase Shift Keying DP-QPSK Dual Polarization Quadrature Phase Shift Keying Line Rate Modulation Format Bandwidth Use in conjunction with the ITU-T Flex-Grid option Optical Reach 400 Gb/s 16 QAM 100 GHz 400 km 400 Gb/s 16 QAM 75 GHz 300 km Channel spacing is a multiple of 12.5 GHz instead of the current 50-GHz-spacing QAM Quadrature Amplitue Modulation Teipen, Griesser, and Eiselt, ICTON 2012 Estimate a 1.3x Benefit in Capacity Estimate this applies to ~25% of the traffic Copyright 2017 Monarch Network Architects LLC J.M. Simmons 24

25 Summary of Architectural Approaches Approximate Benefit Factor Approximate Percentage of Traffic Subject to Benefit Effective Capacity Multiplier More Efficient IP Packing 2 80% 1.7 Asymmetric Connections % 1.2 Multicast 2 20% 1.1 Caching 3 50% 1.5 Dynamic Optical Networking 4 20% 1.2 Gridless/Elastic Optical Networks % 1.5 Optical Reach vs. Capacity % 1.1 Total Effective Capacity Multiplier ~ 6 ~6 more traffic can be carried as compared to today, using the same amount of bandwidth Benefits are not as significant as those provided by technology advancements But likely more feasible in the near-term (using existing fiber infrastructure) These are Rough Estimates!! (some of the benefits may overlap, so full multiplicative effect may not be achieved) Copyright 2017 Monarch Network Architects LLC J.M. Simmons 25

26 Conclusions Fiber Capacity Limits Need to be Addressed Technology Advancements: Numerous implementation challenges Most of the advancements presented are not near-term solutions Architectural Options Most of the advancements presented are near-term solutions Require changes in network management Gridless / elastic networks more challenging to implement Only Transmission and Fiber Capacity Limits were Discussed Scalable switching is needed as well Need to attack the problem from both technology and architectural perspectives Combined, may provide an overall 250 effective capacity increase (We ll be fine for 30 years!) Copyright 2017 Monarch Network Architects LLC J.M. Simmons 26