Turning elastic metro optical networks into reality...ELASTIC TRANSPONDERS (TRx) Reconfigurable Set...
Transcript of Turning elastic metro optical networks into reality...ELASTIC TRANSPONDERS (TRx) Reconfigurable Set...
1 © Nokia 2018
Turning elastic metro optical networks into reality
Séminaire Telecom ParisTech 2019
• Patricia Layec, team
• 11-01-2019
Public
2 © Nokia 2018
1. Introduction
2. Today’s elastic optical networks
3. Towards true elastic networks
4. What needs to be replaced
5. What needs to be added
6. Conclusions
Public
Agenda
3 © Nokia 2018 Public
Introduction
Market segments
Picture from description of work of H2020 METRO-HAUL project
Access ~50km
Metro network ~100-500km
Core network ~500-10000km
~10Gb/s ~10-100Gb/s ~100-400Gb/s
4 © Nokia 2018
• Metro segment is growing 2x faster than core segment
• Key drivers for this growth
Metro video traffic will increase by 720% by 2017
Metro cloud and DC traffic will increase by 440% by 2017
- DC traffic includes DC-to-end-user & DC-to-DC (i.e. DCI) traffic
- Growth rate of DCI is 100% per year since 2012
Public
Introduction
Traffic is growing fast
[Bell Labs Consulting, “Metro network traffic growth: an architecture impact study,” Dec. 2013]
[A. Chen (Alibaba), W3B4, OFC 2016]
0
20
40
60
80
100
120
Tra
ffic
(T
b/s
)
metro backbone
2012 2013 2014 2015 2016 2017
5.6x
3.2x
560% INCREASE IN TOTAL METRO TRAFFIC
METRO TRAFFIC GROWS ALMOST
2X FASTER
5 © Nokia 2018
• Service providers are deploying video caching and cloud computing platforms within metro networks
- Better performance
- Better user experience
more traffic terminated in metro and not going to the core networks
Public
Introduction
Traffic distribution is changing
[Bell Labs Consulting, “Metro network traffic growth: an architecture impact study,” Dec. 2013]
Total IP traffic
IP traffic terminates in
metro network
57%
2012
Total IP traffic
IP traffic terminates in metro network
75%
2017
6 © Nokia 2018
• Traffic patterns are becoming meshed and variable
3-fold heterogeneity:
- connection lengths (and impairments),
- bandwidth requests on demand (capacity market)
- Smaller duration of connections (from months to minutes)
• Reach << 1000km
- Less DSP constraints
• Metro networks are also influenced by over-the-top (vs. telco)
Public
Introduction
Mismatch between dynamics of new services and optical transport
Most innovative solutions should appear in the metro networks
Metro Core
Meshed & variable traffic
patterns
7 © Nokia 2018
• Traffic patterns are becoming meshed and variable
3-fold heterogeneity:
- connection lengths (and impairments),
- bandwidth requests on demand (capacity market)
- Smaller duration of connections (from months to minutes)
• Reach << 1000km
- Less DSP constraints
• Metro networks are also influenced by over-the-top (vs. telco)
Public
Introduction
Mismatch between dynamics of new services and optical transport
Metro Core
Meshed & variable traffic
patterns
Most innovative solutions should appear in the metro networks
Try and learn Continuously adjust Always zero margins
Working with just-enough performance
Design and optimize Set and forget (for 15years) Guaranteed (large) margins Seeking top performance (bit-rate/distance)
OTTs Telcos
8 © Nokia 2018
• Elastic optical networks introduced in 2008 [M. Jinno et al, Th3F6, ECOC’08]
Public
Today’s elastic optical networks
Finisar
From BPSK to 16QAM 40G, 100G, 200G
Cisco
Multirate linecard BPSK, QPSK, 16QAM
50G, 100G, 200G
But in 1 touch can be upgraded to 200G.
Nokia
QPSK, 16QAM 100G, 200G
Infinera
QPSK N x 100G
superchannel
Reconfigurable transponder but in set & forget configuration
9 © Nokia 2018
Universal transponder: 1 does it all !
• Single hardware: single development cost, fast price erosion
• Similar technology for all rates: jointly optimized link design, limited XPM degradation
good side effects on engineering margin and planning tools
• Limit spares: single hardware device whatever the selected data rate [A. Morea et al, Th10F5, ECOC’10]
• Tunable devices: support easily traffic increase and network upgrades [O. Rival et al, P5.12, ECOC’12]
no need to uninstall and replace low rate devices with higher ones
Confidential
Today’s elastic optical networks
Set & forget configuration
Sp
ares
TS
P R
1
TS
P R
2
TS
P R
2
TS
P R
m
TS
P
TS
P
TSP
100Gbps 200Gbps
400Gbps
10 © Nokia 2018
Most popular usage is trade-off between capacity and optical reach
Fine granularity of flexrate technologies
• Set partitioning [J. Renaudier et al, We1C5, ECOC’12]
• Time hybrid QAM [X. Zhou et al, IEEE Com Mag., 2013]
• Probabilistic shaping [F. Buchali, et al., arXiv:1509.08836, 2015]
• FEC with puncturing
Today’s elastic optical networks
700 km 6000 km
+25% +50%
Fixed modulation, fixed power Adapted modulation, optimized power
[D.J. Ives et al., PNC, 29 (3), 2015]
t
Polar. X
8QAM
QPSK 8QAM QPSK
QPSK 8 QAM
Polar. Y
QPSK
8QAM
X Y
11 © Nokia 2018
PDM-QPSK optical spectrum
• Current networks mostly use a 50GHz fixed-grid
• Elastic networks enable flexgrid thanks to:
- Flexible channel spacing
- Tunable symbol rate of the transponder
=> Irregular grid with variable bandwidth of channels
- Techno-economics show a 20% spectrum savings with uniform traffic [A. Morea et al., OFC’11, JWA62]
Public
Today’s elastic optical networks
Spectrum
savings
60Gb/s in 100G l
75Gb/s in 100G l
50Gb/s in 100G l
50GHz
50 ps/div
FPGA electrical output
[A. Dupas et al., OFC’15, M3A2] Commercial WSS and ROADM have now very narrow granularity
12 © Nokia 2018 Public
Towards true elastic networks
ELASTIC TRANSPONDERS (TRx)
Reconfigurable
Set & forget configuration [Y. Zhou, et al. “1.4 Tb Real-Time Alien Superchannel Transport Demonstration Over 410 km Installed Fiber Link Using Software Reconfigurable DP-16 QAM/QPSK,” in JLT 2015]
Measured switching time (field trial with commercial hardware)
QPSK 16QAM : 40 sec 16QAM QPSK : 20 sec
TSP
100Gbps 200Gbps
400Gbps
13 © Nokia 2018 Public
Towards true elastic networks
ELASTIC TRANSPONDERS (TRx)
Reconfigurable
Set & forget configuration
SELF-EVERYTHING ELASTIC OPTICAL NETWORKS
Fast reconfigurable
Automated
Easy to use & configure
TSP
100Gbps 200Gbps
400Gbps
14 © Nokia 2018
• Easy to use and configure
- To trigger intents rather than specific hardware configuration => Tell what you need, not how to do
- e.g. low latency, 200G bandwidth, secure end-to-end connection, …
• Boulder project at Open Networking Foundation (ONF) is the first that really started working on the specifications and design of intent-based
- Open-source projects are developing intent-based networking
Public
Towards true elastic networks
NFV / Applications
Intent-based mediation
SDN controller
Data plane
vQOT vRSA vBoD …
Open day light
15 © Nokia 2018 Public
Towards true elastic networks
Perfect future network
M. Weldon et al., “The Future X Network: A Bell Labs Perspective,” ISBN 978-1498759267, CRC Press, 2015.
16 © Nokia 2018
• Highly scalable, tunable network fabric including:
- Disaggregated SDN network – to decouple the HW and SW that different lifecycle
- Adaptive impairment-aware optimization – to leverage monitoring and machine learning to optimize and automate the network
- Elastic optical metro transport – to allow specific innovation into the metro segment (i.e. low cost, low energy) while today innovation is driven by the core segment (i.e. performance)
- Ultra-scalable routers
• Network OS that fully automates and exposes the tunable network fabric
Confidential
Towards true elastic networks
Perfect future network
17 © Nokia 2018
KEY BENEFITS
• Assumptions:
- 10x traffic increase in 5 years
- Perfect future network
• TCO savings:
- OPEX & CAPEX for Optical (15%)
- OPEX & CAPEX for IP (43%)
And…
- Automation & optimization (42%)
Confidential
Towards true elastic networks
M. Weldon et al., “The Future X Network: A Bell Labs Perspective,” ISBN 978-1498759267, CRC Press, 2015.
IP/Optical cost/Gb
10x traffic
-71%
SDN automation / optimization (42%)
High scalable IP (43%)
High rate WDM (15%)
18 © Nokia 2018
• Data plane is already in good way
- CDC-F ROADMs
- Flexible elastic transponder (data rate, spectrum, tunable laser, etc.)
Commercial products (planned to be) deployed on field
Hardware foundations are ready for perfect future network
• Optical management tools
- Technology-specific
- Backward-compatible to previous generation of hardware
- SDN should become a carrier-grade system with resilient and secure functionalities
• Manual intervention
- Cabling between router ports and photonic nodes
To reduce delays of service creation (from weeks to days)
• Network programmability and applications come from upper layers
- Already happening in our day-to-day life
- dynamic bandwidth consumption,
- network as a service
Public
What needs to be replaced
Bottom-up
Top-down
19 © Nokia 2018
• Hitless data rate change for bandwidth on demand on optical transport
• State-of-the-art:
- 2017: commercial Field Trial, 20-35s switching time [JLT 2017, Y. Zhou et al., Vol 35, n°3]
- 2016: 450 µs switching time between 10 different bit rates [OFC’2016, A. Dupas et al., Th3I.1]
• Experimental proof-of-concept on 100Gbps real-time transmitter:
- Zero loss of data during data rate reconfiguration
- Flex symbol rate transponder (10G, 20G, …, 100G)
- a novel hitless protocol that buffers and synchronizes commands with data flows after all framing processing => No additional latency on data flows and service hitless switching.
=> Max switching time ~12µs.
Public
What needs to be added
Hitless data rate change (speed)
A. Dupas et al., “Ultra-fast Hitless 100Gbit/s Real-Time Bandwidth Variable Transmitter with SDN optical control,” Proc. OFC, 2018
To
fib
er
link
SDN Controller
Local controller
OTUflex-like
from 10 to
100Gbps
Hitless
coding
Tunable High
speed
interfaces
Hitless flexrate transmitter
Synchronization at the bit level
10ns/div
FPGA Clock For 160 bits
Command from SDN controller
Switching
10ns/div
20 © Nokia 2018
• Why?
- When crossing optical nodes, filter cascade reduces the available bandwidth => less tolerant to detuning
• Key goal: A self-optimizing transponder leveraging monitoring at the receiver side (no need for calibration)
- To reduce filter penalties by several dBs (worst case)
- To reduce the large filter design margin provisioned
• We propose an algorithm to safely cancel Tx/Rx lasers and filters misalignments
- Tx and Rx lasers are never perfectly aligned (+/-1.5GHz), and are not aligned with the optical filter cascade either (+/- 3GHz)
- Exact misalignments cannot be predicted
- Closed-loop automated algorithm that correct the effect of Tx/Rx frequency fluctuations using analytical metric
- No need for extra equipment (e.g. OSA)
Public
What needs to be added
Monitoring & closed-loop systems: filter mitigation
[O. Bertran Pardo et al., OFC’15]
Tx frequency correction
Tx Rx Detuned filter cascade (WSS)
Control plane
[C. Delezoide et al., OFC’19]
21 © Nokia 2018
• Analytical metric [C. Delezoide et al., OFC’19]
- Step #1: Filter effects are mitigated by the equalizer, hence we need to compute the power spectral density (PSD) after the Analog-to-Digital Converters (ADCs)
- Step #2: Metric based on the center of mass :
where Fs is the ADC sampling frequency and f is the baseband frequency
- Step #3: ensure the Tx laser is aligned with the Rx one. This permits an optimal monitoring of the signal-filter detuning
- Step #4: minimize the metric |M1|, hence filter penalties
• Experimental measurements
- Tx laser frequency known at +/- 1.5GHz and filter slot center at +/-2.5GHz
Public
What needs to be added
Monitoring & closed-loop systems: filter mitigation
22 © Nokia 2018 Public
What needs to be added
Monitoring & machine learning: architectures
Low latency High Cognition
Vendor A
Rx adaptation
Vendor A
Cloud
Program devices
Monitor
SDN controller
Learning & decision making
Vendor A
Program learning
Monitor
SDN controller
Cloud
Rx learning & adaptation
Learning
23 © Nokia 2018
Monitoring & machine learning: examples
What needs to be added
Public
Low latency High Cognition
QoT prediction “A learning living network”
[Oda et al., JLT’17] Vendor A
Program learning
Monitor
SDN controller
Cloud
Rx learning & adaptation
Learning
Vendor A
Cloud
Program devices
Monitor
SDN controller
Learning & decision making
ML at DSP level [Zibar et al., ECOC’15]
24 © Nokia 2018
QoT prediction with learning
Public
• Network margins (as defined in Augé et al., OFC’13):
- Unallocated margins – where transmission reach exceeds the transmission distance
- Design margin – unwanted, worst case QoT prediction (imperfect model + input uncertainties)
- System margins – time-dependent, fixed margins (e.g. polarization effects, ageing, network load)
• Many recent works on QoT prediction with learning
- Oda et al., JLT, vol. 35, no. 8, 2017
- E. Seve et al., OFC’17
- S. Yan et al., ECOC’17 PDP
- M. Bouda et al., JOCN vol.10 (1), 2018
- …
[Seve et al., OFC’17]
25 © Nokia 2018
• E. Seve et al., OFC’17
- QoT model, evaluation of power uncertainty
- Simulations with statistical error on measured power P
Examples
Public
• S. Oda et al., JLT 2017
- Experimental testbed with 6 ROADMs, 88 channels at 100Gbps with 50GHz spacing
- Monitoring BER
No learning
#Demands
SN
R e
rro
r (d
B)
0
-1
-2
1
2
3
QoT prediction with learning
Prediction instead of (worst-case) forecast
26 © Nokia 2018
Monitoring & machine learning: examples
What needs to be added
Public
Proactive fiber break detection
[Boitier et al., ECOC’17]
Low latency High Cognition
QoT prediction “A learning living network”
[Oda et al., JLT’17] Vendor A
Program learning
Monitor
SDN controller
Cloud
Rx learning & adaptation
Learning
ML at DSP level [Zibar et al., ECOC’15]
27 © Nokia 2018
[F. Boitier et al., ECOC’17]
Proactive Fiber Break Detection
Public
• To proactively detect fiber break to use cost-effective optical restoration
- Be as reliable (as 1+1 protection) and low cost (as in shared restoration)
- Use the transmission fiber as a sensor
- Unleash coherent receiver for monitoring
• We develop a real-time proof-of-concept with monitoring and learning
- We compute State-of-Polarizations (SOP) based on existing FIR filters
- We classify events at the DSP level to lower false alarms and re-route traffic in case of a risky event before the cut occurs
28 © Nokia 2018
[F. Boitier et al., ECOC’17]
Proactive Fiber Break Detection
Public
• Training phase
- Control & management plane
- 16548 events divided into training and test, with 10-fold cross-validation
- 4 different SOP events generated, usage of polarization scrambler for robustness
- We evaluate accuracy with naïve Bayesian classifier for various number of parameters
• Real-time measurements
- Small number of variables required for event classification
=> Easy to embed in real-time boards
- Reduce false alarms and avoid unnecessary re-routing of traffic
Real-time receiver board
Step #1
Calculate SOP
Step #2
Calculate rotation
speed
Step #3
Compare to
threshold
Step #4
Save SOP
during few sec
t
SOP
Alarm
Yes
Event
recognition
!
Embedded CPU in receiver board
29 © Nokia 2018
Real-time proof-of-concept
Proactive Fiber Break Detection
Elastic Tx
Node C Node B
Emulated Rx
Elastic Rx
SDN Controller
Node D
Node A Working path
Restoration path
30 © Nokia 2018 Public
Real-time proof-of-concept
Proactive Fiber Break Detection
31 © Nokia 2018
• Metro is the meeting point between telcos and OTTs with less stringent requirements in long reach / high capacity performance than core networks
Good playground for disruptive research
• Need to change perspective from “set & forget” to “easy to use & (re)-configure”
- Good news is that data plane is already deploying products with high level of flexibility
- Good news: many talks at OFC & ECOC on automation, online optimization
- Need for faster reconfiguration
• Unified control may be more challenging
• Machine learning is a natural evolution to
- Optical monitoring
- Traffic patterns observation
Predicting (deterministic models) instead of forecasting (uncertainty)
Public
Conclusions
Work partly supported by H2020 European project METRO-HAUL
33 © Nokia 2018
• Superchannel optimization example
Public
What needs to be added
Plug and play, monitor and optimize
F. Cugini, et al., “Towards Plug-and-Play Software-defined EONs: Field Trial of Self-Adaptation Carrier Spacing,” in Proc ECOC, 2015
Installation with QoT prediction
subcarrier packing Sequential shift
Monitor BER
Monitor BER => FEC limit reached
WSS reconfiguration
Increase #iterations of FEC decoder
SDN control + supervisory channel
34 © Nokia 2018
• Optical white box [M. De Lenheer, et al., Th1A.7, OFC’16], Architecture on Demand [N. Amaya et al, ICTON’11]
more flexibility to use hardware, open possibilities by software
• Chassis, hence backplane disappears but this has a cost:
- loss of flexibility with manual cabling between hardware elements
- Insertion loss of about 3.5dB due to a fiber switch matrix
• Such losses may be well-accepted in metro networks where the need for high performance is not always mandatory.
• But, back-2-back performance are key in metro networks to increase network capacity [D. Boertjes, Workshop flexible optical networks,ECOC’16]
Public
What needs to be added
Open architecture