Submarine Transmission - AusNOG · LAN PHY – 10.313 Gbps “tradi9onal” framing WAN PHY –...

SubmarineTransmission

WherePhysicsbeatsStandards

2

TridentCableSystemSubseaOp9calfibrecable

Fourfibre-pairs,3570km

~45amplifiersinpath

10TbpsPerth-Singapore

16–2420-30TbpsPerth-Jakarta

Relies on non-standard optical technology

3

Outline

•  Standards•  Op9calTransmissionStandards(withnoapplica9ons)

•  RecentLong-HaulOp9calTransmissionReality(Non-Standards!)

4

Astandard-IEEE802.3(Ethernet)

•  Originally10Base5•  10Base2–IEEE802.3a(1985)

5

Op9calFibreEthernet1993-10BaseF–10Mbps1995–100BaseFX“FastEthernet”1998–1000Base-X“GigabitEthernet”2002–10GBASE-(SR/LR/ER)“10GigE”

MMF SMF

“-ZR”(80km)?Non-standardvendor“-BX”single-fibre?Non-standardvendorLANPHY–10.313Gbps“tradi9onal”framingWANPHY–9.953Gbps,SDHSTM-64framing

6

Op9calFibreEthernet1993-10BaseF–10Mbps1995–100BaseFX“FastEthernet”1998–1000Base-X“GigabitEthernet”

MMF SMF

2002–10GBASE-(SR/LR/ER)“10GigE”

1000xspeedin9years

8 years later…. 2010 – 40G/100GBase.XXX Ethernet

7

(SoManyStandardstochoosefrom…)

40GbE/100GbE

8 Considerations for scale NGN Access Networks

Multiple Parallel 10Gb channels

MMOF requires multiple fibres

No long-range 40G - datacentres and

server connect only

100GbE not compatible with

DWDM

9

Whymul9plelanes?

9

Moore’s Law Transistor count doubles

every 24 months

103.3 104.8

Bandwidth Requirement outstrips Moore’s Law by over 30 times!

Electronics scaling slowly

Nielson’s Law Link Capacity doubles

every 18 months

10

Whymul9plelanes?

Bitrate Bits/metre Bit time 10Gbps 54.8 91 ps/18.2 mm 40Gbps 220 22.7 ps/4.5mm

Speed of Light (in glass)?

Desktop CPUs: ~2-3GHz limit per core then 2/4/8 cores On/Off Lasers: ~10GHz-25GHz per lane then multi lanes

10Gbps is FAST! Only 20 cm per nanosecond 0.2 mm per picosecond

11

NewMul9modeCabling&Connectors

12

400GEthernet

8 years later…. ~2017/8 – IEEE802.3bs 200GE & 400GE MMF–32fibres(16ineachdirec9on)–anothernewMPOconnector/cable

SMFwithindatacentre–8fibres,newPAM4signalformat

SMF“longhaul”–only10kmmax.Formatsincompa9ble–nointerconnectwithaienuators

NO40kmop9on-Incompa9blewithDWDMlong-haul

13

StandardsBumpingintoPhysics•  25-50Gbaudpushesboundaryofphysicsandopto-

electronicinterfaces–anymoreincreaseisfromparalleliza9onandmul9plelanes

•  PaceofEthernetstandardshasslowed–from1000x/9yrto10x/8yrto2-4x/8yr

•  ButNielsen’sLawisnotslowingdown!

•  NocapacitybenefittoStructuredCabling–s9llonly10G/fibreMM,100G/fibreSM.MPOconnectorsdifficulttobreakout.

•  Addinglatency–every100G+linkrequiresFEC(ForwardErrorCorrec9on)&re-alignmentofparallelsignalsbeforecombining-~20-30microsecondsperlink

14

EthernetIrony!

Point-to-PointFullDuplex

Onetransmiier,Onereceiverper“wire”

NoCollisions!

EthernethasevolvedtomakeCSMA/CDsuperfluous!

EveryEthernetsegmentisnow…

15

Intercon9nentalCapacity

Non-Standards!

16

Long-haulOp9calTransmission•  Eachfibrecoreisexpensive–

requiresmanyamplifiers,repeaters,andpower

•  Verylowfibre-count–2,4,6pairs.

•  Big$$$onterminalequipmenttomaximisethroughputperpairusingDWDMtechniques

•  40–80amplifiersmeanshighaienua9on,lowSNR,closetonoisefloor

17

SubmarineChannelCapacity

Capacity per wavelength Year Cable

2.5 Gbps 1995-2000 various

10 Gbps 2000-2010 Japan-US, AJC

40 Gbps 2011 APCN2 upgrade

100 Gbps 2012- SCCS upgrade, FASTER, many

200 Gbps (trial 2014) (Japan-US 630km

segment)

18

Mul9pleChannels

•  ITU-TG.694.1SpectralGridsforDWDM

•  Defined ~80 channels, 50 GHz spacing 80 x 10 Gbps = 800 Gbps per fibre-pair

19

LongHaul100G•  LinecodingdevelopedbyNortel

–DP-QPSK–adoptedbyOIF(Op9calInterworkingForum)

•  Combine2separateparallelsignals

•  ~112Gbpsfrom2x25-28Gbaudperseccarriers

•  Works through existing 50 GHz DWDM equipment •  More robust than 40Gbps NRZ (On-Off-Keying) •  shipping equipment from 2010 •  Completely different from 100G Ethernet signalling

80 x 100 Gbps = 8 Tbps per fibre-pair

No interoperable standard

20

100Gbps–Non-Standard!

•  ITU-TG.959.1OTNPhysicalLayerInterfaces(2012)

Class Speed

OTU-1 2.5Gbps

OTU-2 10Gbps

OTU-3 40Gbps

•  No100Gbpsphysicalinterfacedefined! No standard (use 100GbE)

21

CoherentDetec9on•  ~2010dedicatedDSPprocessorsbecamefastenoughto

directlyprocessop9calsignal

•  Soswarecorrec9onforChroma9c&Polarisa9onDispersion

•  Locked-Laserfrequencyvastlybeiernoiserejec9on&sensi9vity

•  Instantlyenabledjumpfrom10Gbpsto100Gbps

22

NarrowerChannels

Re-paintthelane-lines–squeezechannelsinto37.5GHzspacing

•  80(+25%)100channelsinthesamespectrumband

100 x 100 Gbps = 10 Tbps per fibre-pair

Non-standard

23

Wider‘Superchannels’

Combineadjacentchannelstocarryhigherbit-ratesinlessspectrum

•  400Gbpsin(2x50)100GHzsuperchannel600 GHz 488 GHz 462.5 GHz

Infinera White Paper - https://www.infinera.com/wp-content/uploads/2015/07/Infinera-WP-Evolution_of_Next-Gen_Optical_Networks.pdf#page=9&zoom=auto,-12,499

40 x 400 Gbps = 16 Tbps per fibre-pair

Non- standard!

24

MoreChannels•  ITU-TG.694.1SpectralGridsforDWDM

•  37.5Ghz channels & superchannels completely non-standard….

•  …until Feb 2012 Edition 2 defined ‘Flexible Grid’ on 12.5 GHz granularity.

25

200G,300G,400G….•  Morecomplexencoding-8QAM,16QAM(followingradiotech)

6240km 2880km 1680km @25Gbaud 100G 150G 200G

@50Gbaud 200G 300G 400G

26

BeierFibre

•  “Standard”fibres–0.22dB/km,core65μm2

•  Low-LossFibre–0.17dB/km

•  ~25%longerspans,fewerrepeaters,lowernoise•  LargeEffecGveAreafibre–core130μm2

•  SupportsHighersignalpower,biggeramplifiers

27

Compromises

•  Currentlyatlimitofop9cal&electricalphysicsandspeeds

•  Tradeoffofratevsreachvsbandwidthvsstability

FurtherintotheFuture….

(Spurringanewwaveofnewcables…)

29

Mul9-coreSingle-mode

30

NTTlabs

https://www.ntt-review.jp/archive/ntttechnical.php?contents=ntr201412ra3.pdf&mode=show_pdf

31

www.tridentsc.com.au

Paul.Brooks@tridentsc.com.au

Ques9ons?