100Gbps for NexGen Content Distribution Networks
Martin ZirngiblDomain leader, Physical Technologies, [email protected] Labs Research
2 NANOG45, January 2009
Market Drivers for 100GDigital video trunking:
Part of growing “triple play” consumer packagesVideo bandwidth is easier to monetize(compare with p2p traffic)Requirement for high definition drives− High data rates per channel− Transport of uncompressed video for
access network specific compressionExplosive growth in programming content(estimated ~90% of BW in NG networks)
Video traffic:Requires minimal latency for effective deliveryRequires resilient transport (MPLS and transport layer)Benefits from multicast and from “asymmetric” design
Other factors: storage networking, carrier wholesaling, science applications, grid computing
What’s behind the growth?
VoIP
Enhanced Internet Services
Triple Playbundle
IPTV(incl. HDTV, VoD)
Blendedvoice,video& data
services
2000 2005 today Coming soon
MoreBandwidth
!MoreBandwidth
!!
ContentsourceBroadcast
CATV
IP TV
MPLS/Optical for video trunking
3 NANOG45, January 2009
NexGen Ethernet will be 40GbE and 100GbE
1980 1985 1990 1995 2000 2005 2010
0.01
0.1
1
10
100
10 Mb/s
100 Mb/s
1 Gb/s
10 Gb/s
Year
100 Gb/s
Today: 10 GbE
Good support for 100GbE in standards
IEEE to adopt a OTN compatible standard for 40GbE/100GbEIEEE and ITU-T timelines target mid-2010 for standards completion
Service providers and equipment OEMs also support 100 GbE
100G transport cost effective compared to 10G
A side-effect will be that 40G transport will also become more cost-effective as 100G techniques begin to be implemented in 40G
transmission
4 NANOG45, January 2009
10G I/O10G I/O10G I/O10G I/O10G I/O10G I/O10G I/O10G I/O
Benefits of 100G Networking10G interfaces 100G interfaces
10G 100G
On high-end core routers: Fabric capacity is poorly utilized with lower speed I/O (*)
Better utilization of fabric bandwidth due to improved access into it
(*) Example: Juniper T-1600
10G10G10G
10G10G
10G10G
10G10G10G
10G10G10G
10G10G
10G10G
10G10G10G
On high-end core routers: 10 x 10G interfaces subscribe 100G of BW
100G100G
The price of 100G interface on router will be much lower than 10 x 10G for the same capacity
A higher speed wavelength is better able to accommodate peak BW
Much larger LAG BW!!
Smaller wavelengths limit BW bursts
LAG groups are limited to 17 members
Large wavelength count consumes DWDM grid, represent higher cost for “commons”
Efficient use of DWDM grid.
No parallel wavelengths means lower OAM cost and fewer managed entities
100GN x 10G
100G I/O
100G I/O
100G I/O
100G I/O
5 NANOG45, January 2009
The boundary conditions of capacity growthGrowth of transport requires
Higher spectral efficiency or
Wider amplification bands or
Light another fiber
Latter two often too costly
Growth may have to be accommodated in existing systems
50-GHz channel spacing
Multiple ROADMs
Optical Reach
Capacity Growth based on 100G with advanced modulation format will lead to
very high spectral efficiency and be compatible with existing systems
6 NANOG45, January 2009
~112 Gbaud ~56 Gbaud ~28 Gbaud
1 bit/symbol 2 bit/symbol 4 bit/symbol
Required OSNR
Tolerance to CD, PMD, Filtering (in ROADMs)
Complexity of implementation
Tx
Rx
MZI-based Tx
NRZ/RZ Detector
I
Options for DQPSK
detection
DQPSK Tx
Polarization Multiplexing /
DQPSK TxLaser
Data
CDR
Coherent PM/DQPSK
detector
Multi-level Modulation Formats in Optics
/2π /2πLaser
I
Q
/4π /4π- CDR
DBSK Detector
/4π+ /4π+-
-
I
Q
CDR
/4π− /4π−
/2π
PBS
/2π
I
Q
I
Q
X
Y
Laser PBS
- CDR
- CDR
ADC+
DSP
OLO
hybrid2π hybrid2π
-
-
PBS
hybrid2π hybrid2π
-
-
hybrid2π hybrid2π
Rx
PBS ADC
+
DSP
OLO
7 NANOG45, January 2009
100 Gb/s serial modulation format choices
50 … 100 GHz
± 50 ps/nm
~ 10 ps
40G(Binary)
25 … 50 GHz
± 800 … 2000 ps/nm
~ 40 ps
10G(Binary)
Very large(coherent,
oversampled)± 26 ps/nm± 25 ps/nm± 8 ps/nm
CD tolerance(2-dB penalty)
50 GHz100 GHz150 … 200 GHz150 … 200 GHzSupported gridw/ ROAMDs
Very large(coherent,
oversampled)
100GPDM-QPSK
~ 8 ps
100G(RZ-)DQPSK
~ 3 ps
100GDuobinary
~ 3 psDGD tolerance(1-dB penalty)
100GNRZ
PMD-QPSK … Polarization-mode dispersion quarternary phase shift keyedDGD … Differential group delay
CD … Chromatic dispersionROADM … Reconfigurable optical add/drop multiplexer
~112 Gbaud ~56 Gbaud ~28 Gbaud
1 bit/symbol 2 bit/symbol 4 bit/symbol
Required OSNR / Tolerance to Noise
Tolerance to CD, PMD, Filtering (in ROADMs)
8 NANOG45, January 2009
At 100G impact of some fiber propagation effects (dispersion, PMD, single-channel nonlinearities) is so high that decreasing the baud-rate is mandatory
This requires necessarily the use of more complex modulation formats and receiver architectures
The combined use of:
Polarization-Division Multiplexing (PDM)
Quadrature-Phase Shift Keying (QPSK) allows to decrease baud-rate by a factor of four (100 to 25 Gbaud)
Coherent detection + processing to compensate for linear impairments
At 100G, PDM-QPSK does not suffer from nonlinear effects induced by 10G and 40G neighbors
PDM-QPSK having important issues at 40G becomes recommended at 100G!
100G Transmission: Picking the best modulation format
PDMPDM--QPSK with QPSK with Coherent DetectionCoherent Detection offers the best option for 100Goffers the best option for 100G
9 NANOG45, January 2009
Support for 100G Multi-wavelength Transmission
Our optical test-bed has shown support for 16.4 Tbpstransmission over 164 channels at greater than 2,500 km
[G.Charlet et al., OFC PDP3, 2008]12
11
10
9
8
7
FEC limit
•7
•8
•9
•10
•11
•12
Q²-
fact
or (
dB)
1525 16051565Wavelength (nm)
Wavelength (nm)
Pow
er (
5dB/
div)
1525 1565 1605
16.4 Tbit/s: 164ch x 111Gb/s over 2,550km
[G. Charlet et al., OFC PDP3, 2008]
FEC Limit
Test-bed for 164 channels
x3
A.O.
GFF GFF
+D +D
-D
GFF
GFF
Tunable filter
5
65km55km
A.O.
PBC
164x100Gb/s 2x27.75Gbit/s
164
wav
elen
gths
Coherentmixer sc
ope
com
pute
r
QPSK
QPSK
2x27.75Gbit/s
CL
CL
LRA
Local oscillator
L
C Booster C
L
C
Booster L
LRA GFF
x3
A.O.
GFF GFF
+D +D
-D
+D +D
-D
GFF
GFF
Tunable filter
5
65km55km
A.O.
PBC
164x100Gb/s 2x27.75Gbit/s
164
wav
elen
gths
Coherentmixer sc
ope
com
pute
r
QPSK
QPSK
2x27.75Gbit/s
CL
CL
LRA
Local oscillator
L
C Booster C
L
C
Booster L
LRA GFF
10 NANOG45, January 2009
The Network Evolution Strategy Supporting 100G
IP Access
SONET/SDH
Electrical/Optical Service Aggregation
Transparent (Photonic) Domain
teranode BW management
Direct (colored) optics on
SONET/SDH
OAM&P for SONET/SDH and DWDM
Evolution to 100G is assured through a transparent photonic core which does not hinder introduction of higher bit-rates. Adherence to ITU-T OTN principles ensures non-intrusive wavelength management
Introduction of new capabilities is achieved at the edges through the introduction of transponder (OT) capabilities and enhancements to electrical domain BW management such as the Teranode OXCs
Consolidation of functionality is achieved through integration via a common management and control plane and usage of colored optics on client devices
G.709 OTN for transparency management
Transparent (Photonic) Domain
11 NANOG45, January 2009
Impacts of Filters on 100G and Existing Systems
Optically transparent channels see decreasing channel BW after traversing each filter (ROADM, WSS, etc.)
Advanced modulation format allows flexible deployment of 10/40/100G channels
8
9
10
11
20 40 60 80 100
0.2 nm/div
10 d
B/di
v
0.2 nm/div
10 d
B/di
v
Filter width (GHz)
Q²-
fact
or (
dB)
8
9
10
11
20 40 60 80 100
0.2 nm/div
10 d
B/di
v
0.2 nm/div
10 d
B/di
v
0.2 nm/div
10 d
B/di
v
Filter width (GHz)
Q²-
fact
or (
dB)
Our test-beds show− Almost constant performance obtained for
optical receiver bandwidth above 35GHz.− No crosstalk from adjacent channels when filter
width is above 50GHz thanks to coherent detection and sharp filtering provided in the electrical domain
− Conclude that we can expect tolerance to ROADM cascades
12 NANOG45, January 2009
100G Transmitter at theTampa Central Office
100G Receiver next to LambdaXtreme®at the Miami Central Office
Tampa to Miami 504-km
field route operating
LambdaXtreme®
100G Field Trial over installed, live Verizon system
13 NANOG45, January 2009
100G Field Trial continued
-65-60-55-50-45-40-35-30-25-20
1588.5 1589 1589.5 1590
100G DQPSK
10G OOK
-55
-50
-45
-40
-35
-30
-25
-20
1580 1590 1600 1610 1620
100G DQPSK
10 d
B
1590 1600Wavelength [nm]
Trial included transport of HDTV channel in 100G signal
14 NANOG45, January 2009
Fully Integrated Transmitter/Receiver for DQPSK
π/2
Transmitter Receiver
3.2 mm
Eye diagram 107 Gb/s
1.7 mm
37%
37% +90°
225°
submount
Mask Layout
InP chip
Schematics
OFC2008 postdeadline
15 NANOG45, January 2009
16 QAM Transmitter for High Spectral Efficiency
EAM based MZM modulator
16-quadrature amplitude modulation (QAM) modulator and demonstrate it at 43 Gb/s(10.7Gbaud)
Scalable to higher rates
4 modulators to drive
Compact 3mm long chip
A BC
D
E
(b)(a)
Data #1 Data #2
Data #3 Data #4
0°180°45°
90°-90°
0.150.30
0.10
0.150.30
Im
Re
OFC2008 postdeadline
16 NANOG45, January 2009
Future: 100G Optical Packet Switching Metro Ring Node without WDM
Static optical links between next-neighbor nodes
Requires only one transponder per node
But Thru-traffic reduces add/drop capacity
Scales poorly for large number of nodes N
WDM Metro Ring Node
Static optical links between any two nodes (full mesh)
Eliminates thru-traffic
But requires N – 1 transponders
ROADMs can reduce the number of transponders,but reintroduce thru-traffic
Optical Packet Ring Node
Dynamic optical packet links between all nodes
Requires only one (tunable) transponder per node
Thru-traffic is bypassed, doesn’t reduce add/drop
Adjusts to rapid changes in traffic patterns
Node k
RxTx Processthru-traffic
add drop
Node k
Tx N-1
Tx 1 Rx 1
Rx N-1
Node k
PacketRx
add
Tunable Tx
De-Aggregator
Aggregator
dropk
17 NANOG45, January 2009
Optical Packet Switching Demonstration at 10G
OC-12 generator produces scrambled SONET frames with PRBS 231-1 payload
Unused 10 Gb/s time slots are filled with dummy packets
All Tx’s switch packets based on a global, periodic schedule
De-aggreators reassemble packets with SONET data back into OC-12 stream
BER measurements on OC-12 outputs, < 0.5 dB penalty
Span 1: 21 dB, +400 ps/nm. Span 2: 22 dB, -100 ps/nm-23 -22 -21 -20 -19 -18 -17 -16 -15
12111098
7
6
5
4
3
-log
(BER
)
Power @ Rx [dBm]
b2b Tx1-Rx2 Tx2-Rx3 Tx1-Rx3
18 NANOG45, January 2009
Bell-Labs fast-tuning laser module
Two double-sided PCB boards, including
FPGA w/ look-up table & microprocessor
Fast D/A converters
TEC controller
Bus interfaces
External wavelength-locker:
ADC is sampled once per packet to update look-up table
Improves frequency accuracy from ±6 GHz to ±3 GHz , correcting for temperature & aging
Feedback is slow (order of seconds)
Switching between all combinations of 64 channels in < 50 ns
Improved laser design and advanced current driving for < 5 ns switching
Not implemented in small module
Accuracy ±12 G
75 mm
GainSource
TEC Control
Laser
Wavelocker
50 mm
75 mm
GainSource
TEC Control
Laser
Wavelocker
50 mm
Microprocessor
ADC
Gain andTemperatureControl and
Monitor
LookupTable
Low
-spe
edR
S232
bus
Hig
h-sp
eed
para
llel b
us DAC
DAC
DAC
Amp
Amp
Amp
Laser
ADC TIA
TIA
DiffAmp
DiffAmp
Wavelocker
19 NANOG45, January 2009
Take Away Messages
Video will dominate network traffic and require another order of magnitude more bandwidth
100G will be necessary to cost-effectively satisfy network demands
Advanced modulation formats will allow 100G to have similar reach and ROADM cascadability as 10/40G and to be compatible with 50GHz spacing thus allowing edge upgrade of current and future transparent networks
Photonic Integrated circuits will make 100G transmitters/receiver compatible with los cost, low footprint and high reliability
Optical packet switching may become attractive at 100G to allow for sub-channel connectivity in optical domain