Vincent Chan 1

LIDSLIDS

Future Optical Network Architecture

Vincent Chan, Asuman Ozdaglar, Devavrat Shah

MIT

NSF FIND Meeting Nov 2006

Vincent Chan 2

LIDSLIDS

LocalNetwork

LocalTraffic

BlockingFilter

Optical Router/switch

Freq.Convert

User

User

User

User

User

User

User User

OpticalAmp

UserLocalNetwork

LocalTraffic

BlockingFilter

Optical Router/switch

Freq.Convert

User

User

User

User

User

User

User User

OpticalAmp

User

Optical Networks

•WDM, Optical amplifiers high rates, long reach multicasting

•Optical routing and switching power localization, narrow casting, long reach, high utilization?

•Increase in capacities (major difference between fiber bandwidth and link rates) decrease in cost?

Can we trade bandwidth utilization for lower cost ?Perhaps but with new architectures!

Vincent Chan 3

LIDSLIDS

Future Wide Area Optical Network

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDMOptical

X-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

Optical Network – Near future•Optical switching – GMPLS bypass, load balancing, …

•Packet processing cost dominates

OpticalX-Co nn ect

I P R ou ter

W D M

OpticalX -Co nn ect

I P R ou ter

W D M

Op ticalX-C on nect

IP Ro uter

W D M

Op ticalX- Co nnec t

I P R outer

W D M

OpticalX-C on nect

IP Ro uter

W D M

Op tic alX- Co nn ect

IP R outer

W D M

Op tica lX-C on nec t

IP Ro uter

W D M

Optica lX-Connect

IP Route r

WDM

Otherdata

OXC

IPdata ctrl

Vincent Chan 4

LIDSLIDS

1970 1974 1978 1982 1986 1990 1994 1998 2002 2006 2010 2014 2018 2020

Fiber trunksIncreasing line speeds

Dispersion managed

e-switched architecture

Optical switching

•1st disruptive technology - WDM fiber links•2nd disruptive technology - optical switching•3rd disruptive technology - direct optical access•4th disruptive technology - new transport mechanisms

Electronic access

Optical access

Limit of WDM/optical switching technology ?

Optical network evolution/revolution and disruptive technologies

Computing

1

10

1

02

103

1

04

105

106

Subscriber cost

Can we trade bandwidth utilization for lower cost ?

Vincent Chan 5

LIDSLIDS

Optical Networks

CO

AN

AN

AN

AN

AN

Distribution Rings Access Node

Distribution Tree

Metro/access

Wide area

Feeder

Distribution bus

•Physical and logical architecture

•Transport mechanisms –flow switching

•Routing: separate IP and optical control planes

•Very fast dynamics < 100mS

•Scalable

•Low cost

Vincent Chan 6

LIDSLIDS

Optical flow switching (OFS)

Electronic packet switching (EPS)

Generalized multiprotocol label switching (GMPLS)

Tell-and-Go / burst switching (TaG)

Candidate Transport Mechanisms

LANLAN

mux

WAN

w dedicated wavelength

channels

X

X

OXC

X

X

LANLAN

mux

schedulerWAN

w dedicated wavelength

channels

X

X

OXC

X

X

WANMAN

MAN

LANWANrouter

MANrouter

WAN

MAN

MAN

LAN

LANw dedicated wavelength

channels

MANrouter

X

X

OXC

WANrouter

Vincent Chan 7

LIDSLIDS

Optical Flow Switching and Bypass

User 1 User 2

Router 1 Router 2 Router 3

•End-to end (user-to-user) flows bypassing routers

•Very challenging IP/optical control planes (<100ms)

•Architecture provide multiple services including overlays.

•Supports virtualization

•Security? Optical infrastructure isolation

WDM layer

. . . . . .

Network control

Decreasing cost to scale

Vincent Chan 8

LIDSLIDS

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

TGiven

dynamictraffic

matrices

Derive desired logical topology (multiple, dynamic)

Design sensible fiber plant topology

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

Design physical topology – fixed part of LTD

OpticalX-Connect

IP Router

WDMOpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDMOpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

OpticalX-Connect

IP Router

WDM

Logical topology realized by routing and wavelength assignment, RWA (dynamic part of LTD)

•When failure occurs or traffic changes, tunable XCR & OXC take care of maintaining or providing new logical connection via RWA

•When needed physical topology fixed part of LTD can be redone to get better connections when traffic changes

• Physical topology is made changeable by OXC, slow or fast.

Joint optimization

The Optical Network Architect’s Problem

100ms can be as fast as 5ms + 1 roundtrip time

Vincent Chan 9

LIDSLIDS

This plot assumes that there are 10,000 users per MAN, including both active and dormant users. It is assumed that 10% of the number of users in each MAN are active (i.e. transmitting) at any instant in time. It is also assumed that MAN and WAN routers run at 20% utilization.

10-4

10-3

10-2

10-1

100

101

10210

-6

10-5

10-4

10-3

10-2

10-1

100

Bandwidth per active user (Gbps)

Netw

ork

cost

per

use

r per

bps

($/u

ser/

bps)

OFSEPS (optical transport)EPSGMPLSTaG (OBS?)

Replacement of electronicNICs with optical transceiversat end users

Replacement of electronicMAN transport with opticalMAN transport

Bump and flattening curve representthe addition of expensive opticaltransceivers at end users

Cost comparison of transport mechanisms

Vincent Chan 10

LIDSLIDS

Conventional multicast

Optical multicast for input 1

Optical multicast for input 2

Optical multicast for both inputs

Achievable rate for input 1

Ach

ieva

ble

rate

for

inp

ut 2

2-inputs/2-outputs multicast rates

Conventional multicast

Optical multicast for input 1

Optical multicast for input 2

Optical multicast for both inputs

Achievable rate for input 1

Ach

ieva

ble

rate

for

inp

ut 2

2-inputs/2-outputs multicast rates

Feeder Network

Distribution network

Broadcast flow-switching All other colors for e-IP

WAN

S

Feeder Network

Distribution network

Broadcast flow-switching All other colors for e-IP

WAN

S

Opticalswitch

λ-converter Optical splitter

Optical Multi-cast

……

…Optical splitter λ-converter

Optical tree aggregation

……

…

Opticalswitch

Opticalswitch

λ-converter Optical splitter

Optical Multi-cast

……

…Optical splitter λ-converter

Optical tree aggregation

……

…

……

…

Opticalswitch

•Large optical switches used for aggregation and multi/narrow-cast

•Reconfigurable at mS rates

•Allows dynamic group formation for active flow switching users

•Optical multicast create new reachable regions with networking coding

•Simplifies hardware

Large reconfigurable optical switches as architecture building blocks

Vincent Chan 11

LIDSLIDS

• Two main challenges in the design of routing and flow control mechanisms:

– Design of distributed asynchronous algorithms that work with local information

– Nonconvexities due to integrality constraints, and nonlinear dependencies on the lightpaths owing to fiber nonlinearities.

• Previous Work: RWA problem formulated as a mixed integer-linear program (computationally very hard)

• Two approaches:

– Multi-commodity flow formulation

– Statistical techniques for routing, scheduling and admission control

Routing & Wavelength Assignment and Flow Control Algorithms

Vincent Chan 12

LIDSLIDS

• Optimal multi-commodity flow formulation

• fl : Total flow of link l

• The link cost function convex and monotonically increasing

– Keep link flows away from link capacity

– The link cost function piecewise linear with integer breakpoints

• We proved in some topologies that the relaxed problem has an integer optimal solution and provided an efficient algorithm to find it.

Multi-commodity Flow Formulation

Vincent Chan 13

LIDSLIDS

Algorithms based on state statistics

• Algorithms need to operate at the granularity of flows

• Primary network layer tasks in flow-level network

– Admission control

• Buffering, admitting or dropping flows arriving at network

• Interacts with Routing and Scheduling to make decisions

– Routing and wavelength scheduling

• Assign rates to end-hosts at network layer based on available statistical information

• Given rate requirement by interacting with routing, it allocates physical resources such as lightpaths and wavelengths to end-hosts

Vincent Chan 14

LIDSLIDS

• The algorithms utilize statistical information about network

– Dynamics of network affects the confidence in statistical information

– Complexity of feedback can reduce effect of dynamics Trade-off between complexity and effect of dynamics

• The confidence in statistical information affects performance

– Less accurate statistical information will lead to wastage of resources

• Thus, for algorithms operating in such network

– Trade-off between performance, complexity and network dynamics plays an important role in design

• Traffic statistics collection algorithms are essential in the network performance

Trade-off between performance, complexity and network dynamics

Vincent Chan 15

LIDSLIDS

•‘New technology’

•New transport mechanisms

•New architectures

•New applications

•Grows faster than Moore’s Law

•New opportunities