Data Networks:
Next-Generation Optical Access
toward 10 Gb/s Everywhere
Dr Kyeong Soo (Joseph) Kim (k.s.kim@swansea.ac.uk)
Multidisciplinary Nanotechnology Centre (MNC)
Outline
• Business and Architectural Issues
• Paradigm Shift in Optical Networking
• Ultimate Optical Network Architecture
• Toward Next-Generation Optical Access
• ECR-Based Quantitative Analysis Framework
• Summary
BUSINESS AND
ARCHITECTURAL ISSUES
Aim
• To identify promising routes forward in achieving the
goal of ―10 Gb/s everywhere‖, while making best use of
the existing knowledge in the literature and from earlier
projects.
– The solutions will show most promise of cost
effectiveness and power efficiency, and be future
proof (i.e., allowing bandwidth evolution and
infrastructure reuse).
Partners of Related Project
• 5 Industrial Partners
– Oclaro (Bookham)
– BT
– Ericsson
– CIP
– Gooch & Housego
• 4 Academic Partners
– Cambridge
– Essex
– Swansea
– UCL
Broadband Quality Score III
Univ. of Oxford and Universidad de
Oviedo,
sponsored by Cisco
September 2010
The State of the Internet by Akamai
(2nd Quarter, 2010 Report)
FTTH* vs. Cloud Computing**
SaaS*** User
SaaS Provider/
Cloud User
Cloud Provider
Web Apps
Utility computing
vs.
* NGOA Workshop, Mar. 2008.
** “Above the clouds”, UC Berkeley.
*** SaaS: Software as a Service
FTTH Business Perspective*
Layer Economic
Character
Life Cycle Cost per
Subscriber
Service Layer Low CapEx,
average to high
OpEx
1 to x years ?
Active Layer Average CapEx,
low OpEx
5 to 10 years €300~500
Passive Layer High CapEx, very
(very) low OpEx
25 to 50 years €500~700
•NGOA Workshop, Mar. 2008.
Cloud Computing: New Aspects in Hardware*
• The illusion of infinite computing resources available on
demand
– Through the construction of large-scale, commodity-computer
datacenters at low cost locations, and virtualization technique
• The elimination of up-front commitment by Cloud users
– Companies can start small and increase gradually
• The ability to pay for use of computing resources on a
short-term basis as needed
* “Above the clouds”, UC Berkeley.
Cloud Computing: Economic Benefits*
• Elasticity
– Ability to add or remove
resources at a fine grain
and with a small lead time
• Transference of risks of
– Overprovisioning
(underutilization)
– Underprovisioning
(saturation)
* “Above the clouds”, UC Berkeley.
Max. (=peak)
Min.
Avg.
Time
Demand
Cloud Computing: A New Killer Application for
Next Generation Optical Internet Access?
• Data transfer bottlenecks (to and from Clouds)
– Example: Move 10 TB from UC Berkeley to Amazon in
Seattle*
• WAN link of 20 Mb/s: 4 Msec ≈ 46 days
• Overnight shipping (FedEx): < 1 day (≈ 1.5 Gb/s)
• 10 Gb/s link: ≈ 2 hours
» Even better if we could use more than 10 Gb/s
for a short period!
* “Above the clouds”, UC Berkeley.
ULTIMATE OPTICAL NETWORK
ARCHITECTURE
Current Network Limitations
• Bandwidth-hungry services (e.g., VoD, IPTV):
– Increase the amount of network infrastructure
– Increase the network energy consumption
– Increase the data-driven network crashes
• Due to:
– Unbalance in capacity between core and access
– Mismatch between service/usage models and network
infrastructure
– Large number of power-hungry and error-prone electrical
components/systems
Paradigm Shift in Optical Networking
• Changes in network architectures
– Performance  Energy efficiency driven
– Static  Dynamically reconfigurable network
– Dedicated  Shared resources
– Separate & complicated  Integrated &
simplified management layers/interfaces
– Unbalanced  Balanced bandwidth link utilization
Traditional Way of Using Wavelengths
TX
TX
TX
TX
RX
RX
RX
RX
SW SW
Optical Network with
Passive/Semi-passive Nodes
New Way of Using Wavelengths
Tunable
TX
SW
Tunable
TX
SW
Tunable
TX
SW
Fixed
RX
SW
Fixed
RX
SW
Fixed
RX
SW
Continuous vs. Burst-Mode Communications
TX RXSW SW
...010110100101110100101001001010101111101001010101…
SONET/SDH
Packet Packet Packet
RX SW
10011…0110
Packet Packet Packet
011…010 011…010
Enabling Technologies
• Common denominator in technologies enabling
flexible, dynamically-reconfigurable optical networks
– New multiple access technologies
• e.g., Hybrid TDM/WDM, OFDMA with POLMUX
– Tunable transmitters (lasers) and receivers (filters)
– Burst-mode communications
• The paradigm shift pushes these technologies
toward the edge of the networks!
Ultimate Optical Network Architecture - 1
• A common network
architecture/infrastructure
for access/metro/backbone
• To enjoy the benefits of
Economy of Scale* by
maximizing statistical
multiplexing gain over
– Traffic burstiness
– Different usage patterns
• Challenge: How to
integrate them all?
Backbone/CoreBackbone/CoreMAN
Access
Access

Residential
Users
Business
Users
Access/MAN/Backbone


Residential
Users
Business
Users
* Factors of 5 to 7 decrease in cost (“Above the clouds”, UC Berkeley)
Ultimate Optical Network Architecture - 2
• Network resource as
utility
• Cut the (static) link
between fibre
infrastructure and pool of
network resources (e.g.,
transceivers)
• Challenge: Everything
(both up- and downstream)
in burst-mode
communications
Fibre Infrastructure
(Access/MAN) …
Transceivers
X
Ultimate Optical Network Architecture - 3
…
…
…
…
… …
…
P-T-P & WDM-PON
TDM-PON
Hybrid PON
(with advanced architecture)
Ultimate Optical Network Architecture:
Example
SUCCESS-HPON – Hybrid TDM/WDM-PONs
(2003-2005)
Central
Office
RN
RN
RN
RN
’
1, 2
1
2
21
22 23
’
1
’
3, 4, …
1, 2
3, 4, …
3
’
3
3
31
32
33
TDM-PON ONU
RN TDM-PON RN
WDM-PON ONU
RN WDM-PON RN
Central
Office
RN
RN
RN
RN
’
1, 2
1
2
21
22 23
’
1
’
3, 4, …
1, 2
3, 4, …
3
’
3
3
31
32
33
TDM-PON ONU
RN TDM-PON RN
WDM-PON ONU
RN WDM-PON RN
Protection & restoration is
possible by using different s
on east- and west- bound.
Benefits of Flexible Architecture
R
Tunable
TX 1
Power
Splitter
WDM
DEMUX
ONU 1
ONU 16
...
Start small and grow gradually
Benefits of Flexible Architecture
R
R
Tunable
TX 1
Tunable
TX 2
Power
Splitter
WDM
DEMUX
ONU 1
ONU 32
...
Start small and grow gradually
Benefits of Flexible Architecture
R
R
Tunable
TX 1
Tunable
TX 2
Power
Splitter
WDM
DEMUX
ONU 1
ONU 48
...
R
Tunable
TX 3
Start small and grow gradually
Benefits of Flexible Architecture
R
R
Tunable
TX 1
Tunable
TX 2
Power
Splitter
WDM
DEMUX
ONU 1
ONU 64
...
R
Tunable
TX 3
R
Tunable
TX 4
Start small and grow gradually
Benefits of Flexible Architecture
R
R
Tunable
TX 1
Tunable
TX 2
Power
Splitter
WDM
DEMUX
ONU 1
ONU 64
...
R
Tunable
TX 3
R
Tunable
TX 4
Flexibility and power efficiency
Usage = 50%
(Compared to Peak)
Turn off TX3 & TX4 to save energy
Benefits of Flexible Architecture
R
R
Tunable
TX 1
Tunable
TX 2
Power
Splitter
WDM
DEMUX
ONU 1
ONU 64
...
R
Tunable
TX 3
R
Tunable
TX 4
Redundancy and hot-swap capability
TX4 failed
The system is still running (with
slightly degraded performance)
TOWARD NEXT-GENERATION
OPTICAL ACCESS
Evolution of Optical Access
OLT
ONU
ONU
ONU
OLT
ONU
ONU

TDM-PON
OLT
ONU
ONU
ONU
ONU
OLT
ONU
ONU
ONU
? LR-PON
WDM-PON
Hybrid PON
Geneva, 19-20 June 2008
Evolution scenario
Now ~2010 ~2015
Power splitter deployed for Giga PON
(no replacement / no addition)
Splitter for NGA2
(power splitter or
something new)
G-PON
GE-PON
WDM option to
enable to overlay
multiple G/XGPONs
Co-existence
“Co-existence”
arrows mean to
allow gradual
migration in the
same ODN.
NG-PON2
E.g. Higher-rate TDM
DWDM
Elect. CDM
OFDM,Etc.
Equipment
be common
as much as
possible
NG-PON1 incl.
long-reach option
Capacity
XG-PON
(Up: 2.5G to 10G,
Down: 10G)
Co-existence
Component R&D to enable NG-PON2
A Suggested Time Line from ITU-T/IEEE*
* J. Kani and R. Davey , “Requirements for Next Generation PON,”
Joint ITU-T/IEEE Workshop on NGOA, Jun. 2008.
Areas of Improvement
• Reach
– Through amplification
• Bandwidth per subscriber
– Higher transmission rate in TDM-PON
– Introduction of WDM
• User base
– Serving both residential and business users
through common infrastructure
• Stronger protection capability for business users
Candidates for NGOA
• LR-PON
– 10 Gb/s over 100km with up to 1000:1 split ratios*
• WDM-PON
– Use of array of transceivers
– Lack of BW sharing
– Inventory management of ONUs with different s
– Need of colorless or sourceless ONUs
• Hybrid TDM/WDM-PON
– Use of fast tuneable lasers (and receivers)
– Flexible architecture, but complex MAC/scheduling
– How-swapping capability of tuneable components
* MIT CIPS Optical Broadband Working Group
Challenges
• Power Efficiency
– Number of high-powered transceivers and optical amplifiers in use
• Maintenance
– For active components and thermal optical devices in the field
• Backward compatibility
– For current-generation TDM-PONs
• Scalability
– Start small and grow gradually
• Integration with other services
– Wireless/Video overlay
37
BT’s Current
UK Network
BT-21CN
Simplified UK
Network
Current Status of Network
ECR-BASED QUANTITATIVE
ANALYSIS FRAMEWORK
Requirements for 10-Gb/s Optical Access
• ―10 Gbit/s everywhere‖ is taken to mean that any customer premises can
cost-effectively access useful end-to-end symmetrical throughputs of
10Gb/s data on demand (i.e., whenever they want it but it need not
necessarily always be there).‖ [Excerpt from TSB project requirements]
– Major focus on residential and SME customers.
– 10 Gb/s line rate in the access is a necessary but not sufficient condition.
– Some degree of contention assumed at various points in the network
• What is missing here?
– Description/definition which is
• Specific (e.g., What is ―useful‖?)
• Practical & implementable (e.g., any shared architecture can achieve this?)
• Measurable (during the operation in the field)
What Does “10 Gb/s” Means?
• We need a quantifiable and
measurable definition of ―10 Gb/s‖
at the user side for
– Comparative study of candidate
architectures
– Actual implementations
• Our proposal is based on the
extension of the equivalent circuit
rate (ECR)*.
– For general services & applications
in addition to web-browsing and
interactive data
– Taking into account access/metro
part only
* N.K. Shankaranarayanan, Z. Jiang, and P. Mishra,
“User-perceived performance of web-browsing and
interactive data in HFC cable access networks,” Proc. Of
ICC, pp. 1264-1268, Jun. 2001.
Server
User
User
Candidate architecture
Server User
User
Y
Z = α*min(X, Y) (α < 1)
The same
perceived
performance
X
Implications on Metro/Access* Architectures - 1
• If we mean by ―10 Gb/s‖ the (extended) ECR of
the network architecture (i.e., Z), we can derive
the following conclusions:
– Point-to-point (including static WDM-PONs)
architectures with a UNI (i.e., Y) of 10 Gb/s can meet
the requirement.
• As far as the NNI (i.e., X) is not a bottleneck.
• But there is no statistical multiplexing gain (i.e., sharing of
resources) in this architecture.
* Not end-to-end.
Implications on Metro/Access Architectures - 2
– Shared architectures with a UNI of 10 Gb/s may not meet this
requirement (i.e., ECR < 10 Gb/s), irrespective of NNI.
• Need to increase either line rate (for TDM-PON & hybrid
TDM/WDM-PON) or number of WDM channels (for hybrid
TDM/WDM-PON) at the UNI.
• Note that the ECR is a function of the architecture, the number
of users, and the nature of services/applications.
ECR-Based Quantitative Analysis Framework –
Rationale
• To take into account the interactive nature of actual traffic (e.g.,
TCP flow control) and the performances perceived by end-users
(e.g., delay in web browsing) in quantification of the statistical
multiplexing gain.
• To capture the interaction of many traffic flows through TCP and a
candidate network architecture.
– Simulation models based on OMNeT++ and INET Framework have
been implemented, which provide models for applications as well as
a complete TCP/IP protocol stack.
Calculating ECR
•DW,R: Web page delay from reference architecture
•DW,C: Web page delay from candidate architecture
Start
i=0
R=R’=Ri
Two-sample hypothesis testing with
•H0: E[DW,R] = E[DW,C]
•H1: E[DW,R] < E[DW,C]
Reject
H0?
Yes
i=i+1
R’=R
R=Ri
Two-sample hypothesis-testing with
•H0: E[DW,R] = E[DW,C]
•H1: E[DW,R]  E[DW,C]
No
Reject
H0?
ECR=
(R + R’)/2
Yes
ECR=R
No
End
SYSTEM MODELLING
Overview of Hybrid TDM/WDM-PON
Simulation Setup: System Parameters
• N: Number of ONUs (subscribers)
• n: Number of hosts (users) per ONU
• RD: Rate of distribution fibre
• RF: Rate of feeder fibre
• RB: Rate of backbone network (>> N × RD)
• RTT: End-to-end round trip time
System Model - ECR Reference
• N = 16
• n = 1, 2, …
• RU = RD = RF = 10 Gbit/s
• RB = 1 Tbit/s (future standard or MUX of 100 Gbit/s links)
• RTT = 10 ms (including 600 µs RTT in 60-km PON)
App.
Server
ONU
1
ONU
N
…
RD=RF
Host 1
Host n
…
Host 1
Host n
…
RTT
RB
OLT
RU
System Model – Hybrid PON
• N = 16
• n = 1, 2, …
• RU = RD = RF = 10 Gbit/s
• TX = RX = 1, 2, …
• RB = 1 Tbit/s
• RTT = 10 ms
RF App.
Server
ONU
1
ONU
N
…
RD
RD
Host 1
Host n
…
Host 1
Host n
…
RTT
RB
OLT
RU
TX, RX
TRAFFIC MODELLING
Hierarchical Model Construction
Application
Host
(e.g., PC)
ONU
(w/ Ethernet Switch)
Service
User
Subscriber
(Household)
Overview of Host (User) Node - 1
HTTP 1
TCP
UDP
Network
and
Lower
Layers
HTTP nh
…
FTP 1
FTP nf
…
Video 1
Video nv
…
UNI
Overview of Host (User) Node - 2
• nh = nv = 1
– Assume that a user can watch only one video channel and
interact with only one web session simultaneously at any given
time.
• As far as user perceived (interactive) performance is concerned.
• nf should be kept large to load the high-speed access link.
– FTP is usually background process.
• This could be HTTP sessions just downloading files.
– Suggest 10 as a starting point.
Observations & Comments
• For study of network architectures/protocols, the
frame/packet-level traffic modelling is not very useful.
– e.g., Packet inter-arrival statistics highly depend on network
architectures/protocols.
• We will focus on application level traffic modelling, i.e.,
above transport layer (TCP/UDP).
– Statistics on sources (e.g., file size for FTP and frame size for
video) and user behaviour are critical.
– It is, however, extremely difficult to find such data!
HTTP Traffic Model - 1
• A behavioural model for user(s) web browsing based on [2] with following
simplification:
– No caching and pipelining
– Adapted for traffic generation at the client side
Server
Client
Request for
HTTP object
Request
for embedded
object 1
Response
Parsing Time Reading Time
…
Request
for embedded
object 2
Response to the last
embedded object
Request
for HTTP
object
Web page delay (= session delay*)
* Include connection (i.e., socket) set-up time as well (which is not shown in the figure).
HTTP Traffic Model -2
Parameters / Measurements Best Fit (Parameters)
HTML Object Size [Byte] /
Mean=11872, SD=38036, Max=2 M
Truncated lognormal (=7.90272,
=1.7643, max=2 M)
Mean=12538.25, SD=45232.98
Embedded Object Size [Byte] /
Mean =12460, SD=116050, Max=6 M
Truncated lognormal (=7.51384,
=2.17454, max=6 M)
Mean=18364.43, SD=105251.3
Number of Embedded Objects /
Mean=5.07, Max=300
Gamma (=0.141385, =40.3257)
Mean=5.70, SD=15.16
Parsing Time [sec] /
Mean=3.12, SD=14.21, Max=300
Truncated lognormal (=-1.24892,
=2.08427, max=300)
Mean=2.252969, SD=9.68527
Reading Time [sec] /
Mean=39.70, SD=324.92, Max=10000
Lognormal (=-0.495204, =2.7731)
Mean=28.50, SD=1332.285
Request Size [Byte] /
Mean=318.59, SD=179.46
Uniform (a=0, b=700)
Mean=350, SD=202.07
Streaming Video Traffic Model - 1
• HDTV quality, realistic, high bit-rate video traffic models are
needed for NGOA
– Use H.264/AVC video traces
– ―Terminator 2‖ VBR clip from ASU Video Trace Library
• Duration: ~10 min
• Encoder: H.264 FRExt
• Frame Size: HD 1280x720p
• GoP Size: 12
• No. B Frames: 2
• Quantizer: 10
• Mean frame bit rate: 28.6 Mbit/s
» ~334 streams needed to fill 10 Gbit/s line with the following assumption.
Streaming Video Traffic Model - 2
• Interface with OMNeT++/INET framework
– Through ―UDPVideoStream{Svr,Cli}WithTrace‖ modules:
• UDP server can handle multiple client requests simultaneously
• Random starting phase for each request
• Wrap around to generate infinite streams
• UDP client records the following performance metrics:
» Packet end-to-end delay (vector)
» Packet loss rate
» Frame loss rate
» Decodable frame rate (perceived quality metric)
FTP Traffic Model - 1
• A simple model for user(s) file downloading based on [3]:
– The model is for a data transfer connection only.
– Multiplexed (nf = 10) to emulate future FTP/data services at 10 Gbit/s rate
– Adapted for traffic generation at the client side
Server
Client
Request for
a file to download
Reading Time
Response to the last
embedded object
Request for
a file to download
File download delay
(= session delay)
FTP Traffic Model -2
Parameters Probability Distribution Function
(PDF)
File Size [Byte] /
Mean=2 M, SD=0.722 M, Max=5 M
Truncated lognormal (=14.45,
=0.35, max=5 M)
Mean=1995616(~2 M),
SD=700089.8(~ 0.70M)
Reading Time [sec] /
Mean=180
Exponential (=0.006)
Mean=166.667, SD=166.667
Request Size [Byte] /
Mean=318.59, SD=179.46
Uniform (a=0, b=700)
Mean=350, SD=202.07
Simulation Environment
OMNeT++ with
INET framework
Streamline Linux Cluster
• 22 computing nodes (each with 8 cores and 8GB memory)
• Total 176 cores and 176 GB memory
INITIAL RESULTS &
DISCUSSIONS
ECR Reference – Web Page Delay
Hybrid TDM/WDM-PON – Web Page Delay
Hybrid TDM/WDM-PON – ECR
Hybrid TDM/WDM-PON –
Min. Number of TXs to Achieve ECR of Rtarget
Discussions - 1
• Dedicated architectures with 10-Gb/s line rate — including
static WDM-PON — can provide 10-Gb/s ECR (by
definition).
– As far as there is no contention in the network side.
– But, we cannot enjoy any statistical multiplexing gain
(i.e., sharing of resources) other than some fibre
infrastructure in case of WDM-PON.
Discussions - 2
• Hybrid TDM/WDM-PON with 10-Gb/s line rate can also provide
10-Gb/s ECR with multiple transceivers whose number
depends on traffic load.
– It is remarkable that hybrid PON with just one transceiver
can achieve 10-Gb/s ECR until n reaches 5.
• When n=5, streaming video traffic alone pushes about
150-Mb/s stream into ONU and 2.4-Gb/s multiplexed
stream into OLT (out of 16 ONUs).
– An ideal shared architecture would be that of large split
ratio with multiple wavelength channels.
• i.e., SuperPON + hybrid TDM/WDM-PON
Summary
• Changing business environment and demands are driving
forces behind the paradigm shift in optical networking toward
– Flexible, dynamically-reconfigurable network to better utilize network
resources
– Passive/semi-passive network to maximise energy efficiency
– A common network infrastructure for access/metro/backbone
• We have been working on the following tasks to realize 10-
Gb/s NGOA solutions:
– Investigate candidate architectures in terms of cost, power efficiency,
maintenance, scalability, and extensibility.
– Propose ECR-based comparative analysis framework and
demonstrate benefits of shared architecture (e.g., hybrid PON) based
on it.
References
1. N.K. Shankaranarayanan, Z. Jiang, and P. Mishra, ―User-
perceived performance of web-browsing and interactive
data in HFC cable access networks,‖ Proc. Of ICC, pp.
1264-1268, Jun. 2001.
2. J. J. Lee and M. Gupta, ―A new traffic model for current
user web browsing behavior,‖ Research@Intel, 2007
[Available online].
3. cdma2000 Evaluation Methodology, 3GPP2 C.R1002-B,
3GPP2 Std., Rev. B, Dec. 2009 [Available online].
70
Questions?
Thank you for your time!
For more information on today’s
presentation, please visit
http://iat-hnrl.swan.ac.uk/~kks/

Data Networks: Next-Generation Optical Access toward 10 Gb/s Everywhere

  • 1.
    Data Networks: Next-Generation OpticalAccess toward 10 Gb/s Everywhere Dr Kyeong Soo (Joseph) Kim (k.s.kim@swansea.ac.uk) Multidisciplinary Nanotechnology Centre (MNC)
  • 3.
    Outline • Business andArchitectural Issues • Paradigm Shift in Optical Networking • Ultimate Optical Network Architecture • Toward Next-Generation Optical Access • ECR-Based Quantitative Analysis Framework • Summary
  • 4.
  • 5.
    Aim • To identifypromising routes forward in achieving the goal of ―10 Gb/s everywhere‖, while making best use of the existing knowledge in the literature and from earlier projects. – The solutions will show most promise of cost effectiveness and power efficiency, and be future proof (i.e., allowing bandwidth evolution and infrastructure reuse).
  • 6.
    Partners of RelatedProject • 5 Industrial Partners – Oclaro (Bookham) – BT – Ericsson – CIP – Gooch & Housego • 4 Academic Partners – Cambridge – Essex – Swansea – UCL
  • 7.
    Broadband Quality ScoreIII Univ. of Oxford and Universidad de Oviedo, sponsored by Cisco September 2010
  • 8.
    The State ofthe Internet by Akamai (2nd Quarter, 2010 Report)
  • 9.
    FTTH* vs. CloudComputing** SaaS*** User SaaS Provider/ Cloud User Cloud Provider Web Apps Utility computing vs. * NGOA Workshop, Mar. 2008. ** “Above the clouds”, UC Berkeley. *** SaaS: Software as a Service
  • 10.
    FTTH Business Perspective* LayerEconomic Character Life Cycle Cost per Subscriber Service Layer Low CapEx, average to high OpEx 1 to x years ? Active Layer Average CapEx, low OpEx 5 to 10 years €300~500 Passive Layer High CapEx, very (very) low OpEx 25 to 50 years €500~700 •NGOA Workshop, Mar. 2008.
  • 11.
    Cloud Computing: NewAspects in Hardware* • The illusion of infinite computing resources available on demand – Through the construction of large-scale, commodity-computer datacenters at low cost locations, and virtualization technique • The elimination of up-front commitment by Cloud users – Companies can start small and increase gradually • The ability to pay for use of computing resources on a short-term basis as needed * “Above the clouds”, UC Berkeley.
  • 12.
    Cloud Computing: EconomicBenefits* • Elasticity – Ability to add or remove resources at a fine grain and with a small lead time • Transference of risks of – Overprovisioning (underutilization) – Underprovisioning (saturation) * “Above the clouds”, UC Berkeley. Max. (=peak) Min. Avg. Time Demand
  • 13.
    Cloud Computing: ANew Killer Application for Next Generation Optical Internet Access? • Data transfer bottlenecks (to and from Clouds) – Example: Move 10 TB from UC Berkeley to Amazon in Seattle* • WAN link of 20 Mb/s: 4 Msec ≈ 46 days • Overnight shipping (FedEx): < 1 day (≈ 1.5 Gb/s) • 10 Gb/s link: ≈ 2 hours » Even better if we could use more than 10 Gb/s for a short period! * “Above the clouds”, UC Berkeley.
  • 14.
  • 15.
    Current Network Limitations •Bandwidth-hungry services (e.g., VoD, IPTV): – Increase the amount of network infrastructure – Increase the network energy consumption – Increase the data-driven network crashes • Due to: – Unbalance in capacity between core and access – Mismatch between service/usage models and network infrastructure – Large number of power-hungry and error-prone electrical components/systems
  • 16.
    Paradigm Shift inOptical Networking • Changes in network architectures – Performance  Energy efficiency driven – Static  Dynamically reconfigurable network – Dedicated  Shared resources – Separate & complicated  Integrated & simplified management layers/interfaces – Unbalanced  Balanced bandwidth link utilization
  • 17.
    Traditional Way ofUsing Wavelengths TX TX TX TX RX RX RX RX SW SW
  • 18.
    Optical Network with Passive/Semi-passiveNodes New Way of Using Wavelengths Tunable TX SW Tunable TX SW Tunable TX SW Fixed RX SW Fixed RX SW Fixed RX SW
  • 19.
    Continuous vs. Burst-ModeCommunications TX RXSW SW ...010110100101110100101001001010101111101001010101… SONET/SDH Packet Packet Packet RX SW 10011…0110 Packet Packet Packet 011…010 011…010
  • 20.
    Enabling Technologies • Commondenominator in technologies enabling flexible, dynamically-reconfigurable optical networks – New multiple access technologies • e.g., Hybrid TDM/WDM, OFDMA with POLMUX – Tunable transmitters (lasers) and receivers (filters) – Burst-mode communications • The paradigm shift pushes these technologies toward the edge of the networks!
  • 21.
    Ultimate Optical NetworkArchitecture - 1 • A common network architecture/infrastructure for access/metro/backbone • To enjoy the benefits of Economy of Scale* by maximizing statistical multiplexing gain over – Traffic burstiness – Different usage patterns • Challenge: How to integrate them all? Backbone/CoreBackbone/CoreMAN Access Access  Residential Users Business Users Access/MAN/Backbone   Residential Users Business Users * Factors of 5 to 7 decrease in cost (“Above the clouds”, UC Berkeley)
  • 22.
    Ultimate Optical NetworkArchitecture - 2 • Network resource as utility • Cut the (static) link between fibre infrastructure and pool of network resources (e.g., transceivers) • Challenge: Everything (both up- and downstream) in burst-mode communications Fibre Infrastructure (Access/MAN) … Transceivers X
  • 23.
    Ultimate Optical NetworkArchitecture - 3 … … … … … … … P-T-P & WDM-PON TDM-PON Hybrid PON (with advanced architecture)
  • 24.
    Ultimate Optical NetworkArchitecture: Example SUCCESS-HPON – Hybrid TDM/WDM-PONs (2003-2005) Central Office RN RN RN RN ’ 1, 2 1 2 21 22 23 ’ 1 ’ 3, 4, … 1, 2 3, 4, … 3 ’ 3 3 31 32 33 TDM-PON ONU RN TDM-PON RN WDM-PON ONU RN WDM-PON RN Central Office RN RN RN RN ’ 1, 2 1 2 21 22 23 ’ 1 ’ 3, 4, … 1, 2 3, 4, … 3 ’ 3 3 31 32 33 TDM-PON ONU RN TDM-PON RN WDM-PON ONU RN WDM-PON RN Protection & restoration is possible by using different s on east- and west- bound.
  • 25.
    Benefits of FlexibleArchitecture R Tunable TX 1 Power Splitter WDM DEMUX ONU 1 ONU 16 ... Start small and grow gradually
  • 26.
    Benefits of FlexibleArchitecture R R Tunable TX 1 Tunable TX 2 Power Splitter WDM DEMUX ONU 1 ONU 32 ... Start small and grow gradually
  • 27.
    Benefits of FlexibleArchitecture R R Tunable TX 1 Tunable TX 2 Power Splitter WDM DEMUX ONU 1 ONU 48 ... R Tunable TX 3 Start small and grow gradually
  • 28.
    Benefits of FlexibleArchitecture R R Tunable TX 1 Tunable TX 2 Power Splitter WDM DEMUX ONU 1 ONU 64 ... R Tunable TX 3 R Tunable TX 4 Start small and grow gradually
  • 29.
    Benefits of FlexibleArchitecture R R Tunable TX 1 Tunable TX 2 Power Splitter WDM DEMUX ONU 1 ONU 64 ... R Tunable TX 3 R Tunable TX 4 Flexibility and power efficiency Usage = 50% (Compared to Peak) Turn off TX3 & TX4 to save energy
  • 30.
    Benefits of FlexibleArchitecture R R Tunable TX 1 Tunable TX 2 Power Splitter WDM DEMUX ONU 1 ONU 64 ... R Tunable TX 3 R Tunable TX 4 Redundancy and hot-swap capability TX4 failed The system is still running (with slightly degraded performance)
  • 31.
  • 32.
    Evolution of OpticalAccess OLT ONU ONU ONU OLT ONU ONU  TDM-PON OLT ONU ONU ONU ONU OLT ONU ONU ONU ? LR-PON WDM-PON Hybrid PON
  • 33.
    Geneva, 19-20 June2008 Evolution scenario Now ~2010 ~2015 Power splitter deployed for Giga PON (no replacement / no addition) Splitter for NGA2 (power splitter or something new) G-PON GE-PON WDM option to enable to overlay multiple G/XGPONs Co-existence “Co-existence” arrows mean to allow gradual migration in the same ODN. NG-PON2 E.g. Higher-rate TDM DWDM Elect. CDM OFDM,Etc. Equipment be common as much as possible NG-PON1 incl. long-reach option Capacity XG-PON (Up: 2.5G to 10G, Down: 10G) Co-existence Component R&D to enable NG-PON2 A Suggested Time Line from ITU-T/IEEE* * J. Kani and R. Davey , “Requirements for Next Generation PON,” Joint ITU-T/IEEE Workshop on NGOA, Jun. 2008.
  • 34.
    Areas of Improvement •Reach – Through amplification • Bandwidth per subscriber – Higher transmission rate in TDM-PON – Introduction of WDM • User base – Serving both residential and business users through common infrastructure • Stronger protection capability for business users
  • 35.
    Candidates for NGOA •LR-PON – 10 Gb/s over 100km with up to 1000:1 split ratios* • WDM-PON – Use of array of transceivers – Lack of BW sharing – Inventory management of ONUs with different s – Need of colorless or sourceless ONUs • Hybrid TDM/WDM-PON – Use of fast tuneable lasers (and receivers) – Flexible architecture, but complex MAC/scheduling – How-swapping capability of tuneable components * MIT CIPS Optical Broadband Working Group
  • 36.
    Challenges • Power Efficiency –Number of high-powered transceivers and optical amplifiers in use • Maintenance – For active components and thermal optical devices in the field • Backward compatibility – For current-generation TDM-PONs • Scalability – Start small and grow gradually • Integration with other services – Wireless/Video overlay
  • 37.
    37 BT’s Current UK Network BT-21CN SimplifiedUK Network Current Status of Network
  • 38.
  • 39.
    Requirements for 10-Gb/sOptical Access • ―10 Gbit/s everywhere‖ is taken to mean that any customer premises can cost-effectively access useful end-to-end symmetrical throughputs of 10Gb/s data on demand (i.e., whenever they want it but it need not necessarily always be there).‖ [Excerpt from TSB project requirements] – Major focus on residential and SME customers. – 10 Gb/s line rate in the access is a necessary but not sufficient condition. – Some degree of contention assumed at various points in the network • What is missing here? – Description/definition which is • Specific (e.g., What is ―useful‖?) • Practical & implementable (e.g., any shared architecture can achieve this?) • Measurable (during the operation in the field)
  • 40.
    What Does “10Gb/s” Means? • We need a quantifiable and measurable definition of ―10 Gb/s‖ at the user side for – Comparative study of candidate architectures – Actual implementations • Our proposal is based on the extension of the equivalent circuit rate (ECR)*. – For general services & applications in addition to web-browsing and interactive data – Taking into account access/metro part only * N.K. Shankaranarayanan, Z. Jiang, and P. Mishra, “User-perceived performance of web-browsing and interactive data in HFC cable access networks,” Proc. Of ICC, pp. 1264-1268, Jun. 2001. Server User User Candidate architecture Server User User Y Z = α*min(X, Y) (α < 1) The same perceived performance X
  • 41.
    Implications on Metro/Access*Architectures - 1 • If we mean by ―10 Gb/s‖ the (extended) ECR of the network architecture (i.e., Z), we can derive the following conclusions: – Point-to-point (including static WDM-PONs) architectures with a UNI (i.e., Y) of 10 Gb/s can meet the requirement. • As far as the NNI (i.e., X) is not a bottleneck. • But there is no statistical multiplexing gain (i.e., sharing of resources) in this architecture. * Not end-to-end.
  • 42.
    Implications on Metro/AccessArchitectures - 2 – Shared architectures with a UNI of 10 Gb/s may not meet this requirement (i.e., ECR < 10 Gb/s), irrespective of NNI. • Need to increase either line rate (for TDM-PON & hybrid TDM/WDM-PON) or number of WDM channels (for hybrid TDM/WDM-PON) at the UNI. • Note that the ECR is a function of the architecture, the number of users, and the nature of services/applications.
  • 43.
    ECR-Based Quantitative AnalysisFramework – Rationale • To take into account the interactive nature of actual traffic (e.g., TCP flow control) and the performances perceived by end-users (e.g., delay in web browsing) in quantification of the statistical multiplexing gain. • To capture the interaction of many traffic flows through TCP and a candidate network architecture. – Simulation models based on OMNeT++ and INET Framework have been implemented, which provide models for applications as well as a complete TCP/IP protocol stack.
  • 44.
    Calculating ECR •DW,R: Webpage delay from reference architecture •DW,C: Web page delay from candidate architecture Start i=0 R=R’=Ri Two-sample hypothesis testing with •H0: E[DW,R] = E[DW,C] •H1: E[DW,R] < E[DW,C] Reject H0? Yes i=i+1 R’=R R=Ri Two-sample hypothesis-testing with •H0: E[DW,R] = E[DW,C] •H1: E[DW,R]  E[DW,C] No Reject H0? ECR= (R + R’)/2 Yes ECR=R No End
  • 45.
  • 46.
    Overview of HybridTDM/WDM-PON
  • 47.
    Simulation Setup: SystemParameters • N: Number of ONUs (subscribers) • n: Number of hosts (users) per ONU • RD: Rate of distribution fibre • RF: Rate of feeder fibre • RB: Rate of backbone network (>> N × RD) • RTT: End-to-end round trip time
  • 48.
    System Model -ECR Reference • N = 16 • n = 1, 2, … • RU = RD = RF = 10 Gbit/s • RB = 1 Tbit/s (future standard or MUX of 100 Gbit/s links) • RTT = 10 ms (including 600 µs RTT in 60-km PON) App. Server ONU 1 ONU N … RD=RF Host 1 Host n … Host 1 Host n … RTT RB OLT RU
  • 49.
    System Model –Hybrid PON • N = 16 • n = 1, 2, … • RU = RD = RF = 10 Gbit/s • TX = RX = 1, 2, … • RB = 1 Tbit/s • RTT = 10 ms RF App. Server ONU 1 ONU N … RD RD Host 1 Host n … Host 1 Host n … RTT RB OLT RU TX, RX
  • 50.
  • 51.
    Hierarchical Model Construction Application Host (e.g.,PC) ONU (w/ Ethernet Switch) Service User Subscriber (Household)
  • 52.
    Overview of Host(User) Node - 1 HTTP 1 TCP UDP Network and Lower Layers HTTP nh … FTP 1 FTP nf … Video 1 Video nv … UNI
  • 53.
    Overview of Host(User) Node - 2 • nh = nv = 1 – Assume that a user can watch only one video channel and interact with only one web session simultaneously at any given time. • As far as user perceived (interactive) performance is concerned. • nf should be kept large to load the high-speed access link. – FTP is usually background process. • This could be HTTP sessions just downloading files. – Suggest 10 as a starting point.
  • 54.
    Observations & Comments •For study of network architectures/protocols, the frame/packet-level traffic modelling is not very useful. – e.g., Packet inter-arrival statistics highly depend on network architectures/protocols. • We will focus on application level traffic modelling, i.e., above transport layer (TCP/UDP). – Statistics on sources (e.g., file size for FTP and frame size for video) and user behaviour are critical. – It is, however, extremely difficult to find such data!
  • 55.
    HTTP Traffic Model- 1 • A behavioural model for user(s) web browsing based on [2] with following simplification: – No caching and pipelining – Adapted for traffic generation at the client side Server Client Request for HTTP object Request for embedded object 1 Response Parsing Time Reading Time … Request for embedded object 2 Response to the last embedded object Request for HTTP object Web page delay (= session delay*) * Include connection (i.e., socket) set-up time as well (which is not shown in the figure).
  • 56.
    HTTP Traffic Model-2 Parameters / Measurements Best Fit (Parameters) HTML Object Size [Byte] / Mean=11872, SD=38036, Max=2 M Truncated lognormal (=7.90272, =1.7643, max=2 M) Mean=12538.25, SD=45232.98 Embedded Object Size [Byte] / Mean =12460, SD=116050, Max=6 M Truncated lognormal (=7.51384, =2.17454, max=6 M) Mean=18364.43, SD=105251.3 Number of Embedded Objects / Mean=5.07, Max=300 Gamma (=0.141385, =40.3257) Mean=5.70, SD=15.16 Parsing Time [sec] / Mean=3.12, SD=14.21, Max=300 Truncated lognormal (=-1.24892, =2.08427, max=300) Mean=2.252969, SD=9.68527 Reading Time [sec] / Mean=39.70, SD=324.92, Max=10000 Lognormal (=-0.495204, =2.7731) Mean=28.50, SD=1332.285 Request Size [Byte] / Mean=318.59, SD=179.46 Uniform (a=0, b=700) Mean=350, SD=202.07
  • 57.
    Streaming Video TrafficModel - 1 • HDTV quality, realistic, high bit-rate video traffic models are needed for NGOA – Use H.264/AVC video traces – ―Terminator 2‖ VBR clip from ASU Video Trace Library • Duration: ~10 min • Encoder: H.264 FRExt • Frame Size: HD 1280x720p • GoP Size: 12 • No. B Frames: 2 • Quantizer: 10 • Mean frame bit rate: 28.6 Mbit/s » ~334 streams needed to fill 10 Gbit/s line with the following assumption.
  • 58.
    Streaming Video TrafficModel - 2 • Interface with OMNeT++/INET framework – Through ―UDPVideoStream{Svr,Cli}WithTrace‖ modules: • UDP server can handle multiple client requests simultaneously • Random starting phase for each request • Wrap around to generate infinite streams • UDP client records the following performance metrics: » Packet end-to-end delay (vector) » Packet loss rate » Frame loss rate » Decodable frame rate (perceived quality metric)
  • 59.
    FTP Traffic Model- 1 • A simple model for user(s) file downloading based on [3]: – The model is for a data transfer connection only. – Multiplexed (nf = 10) to emulate future FTP/data services at 10 Gbit/s rate – Adapted for traffic generation at the client side Server Client Request for a file to download Reading Time Response to the last embedded object Request for a file to download File download delay (= session delay)
  • 60.
    FTP Traffic Model-2 Parameters Probability Distribution Function (PDF) File Size [Byte] / Mean=2 M, SD=0.722 M, Max=5 M Truncated lognormal (=14.45, =0.35, max=5 M) Mean=1995616(~2 M), SD=700089.8(~ 0.70M) Reading Time [sec] / Mean=180 Exponential (=0.006) Mean=166.667, SD=166.667 Request Size [Byte] / Mean=318.59, SD=179.46 Uniform (a=0, b=700) Mean=350, SD=202.07
  • 61.
    Simulation Environment OMNeT++ with INETframework Streamline Linux Cluster • 22 computing nodes (each with 8 cores and 8GB memory) • Total 176 cores and 176 GB memory
  • 62.
  • 63.
    ECR Reference –Web Page Delay
  • 64.
    Hybrid TDM/WDM-PON –Web Page Delay
  • 65.
  • 66.
    Hybrid TDM/WDM-PON – Min.Number of TXs to Achieve ECR of Rtarget
  • 67.
    Discussions - 1 •Dedicated architectures with 10-Gb/s line rate — including static WDM-PON — can provide 10-Gb/s ECR (by definition). – As far as there is no contention in the network side. – But, we cannot enjoy any statistical multiplexing gain (i.e., sharing of resources) other than some fibre infrastructure in case of WDM-PON.
  • 68.
    Discussions - 2 •Hybrid TDM/WDM-PON with 10-Gb/s line rate can also provide 10-Gb/s ECR with multiple transceivers whose number depends on traffic load. – It is remarkable that hybrid PON with just one transceiver can achieve 10-Gb/s ECR until n reaches 5. • When n=5, streaming video traffic alone pushes about 150-Mb/s stream into ONU and 2.4-Gb/s multiplexed stream into OLT (out of 16 ONUs). – An ideal shared architecture would be that of large split ratio with multiple wavelength channels. • i.e., SuperPON + hybrid TDM/WDM-PON
  • 69.
    Summary • Changing businessenvironment and demands are driving forces behind the paradigm shift in optical networking toward – Flexible, dynamically-reconfigurable network to better utilize network resources – Passive/semi-passive network to maximise energy efficiency – A common network infrastructure for access/metro/backbone • We have been working on the following tasks to realize 10- Gb/s NGOA solutions: – Investigate candidate architectures in terms of cost, power efficiency, maintenance, scalability, and extensibility. – Propose ECR-based comparative analysis framework and demonstrate benefits of shared architecture (e.g., hybrid PON) based on it.
  • 70.
    References 1. N.K. Shankaranarayanan,Z. Jiang, and P. Mishra, ―User- perceived performance of web-browsing and interactive data in HFC cable access networks,‖ Proc. Of ICC, pp. 1264-1268, Jun. 2001. 2. J. J. Lee and M. Gupta, ―A new traffic model for current user web browsing behavior,‖ Research@Intel, 2007 [Available online]. 3. cdma2000 Evaluation Methodology, 3GPP2 C.R1002-B, 3GPP2 Std., Rev. B, Dec. 2009 [Available online]. 70
  • 71.
    Questions? Thank you foryour time! For more information on today’s presentation, please visit http://iat-hnrl.swan.ac.uk/~kks/