[PPT] - Future Internet Chapter 2: Optical Networking Holger Karl Computer PowerPoint Presentation

SLIDE 1

Computer Networks Group Universität Paderborn

Future Internet Chapter 2: Optical Networking

Holger Karl

SLIDE 2

Goals of this chapter

Optical networks realize today’s Internet’s immense

transport capacity

This chapter
Looks into some of the physical properties that make optical

networks possible

Identifies some of the resulting challenges (in particular, no fast

switches)

Discuss the algorithmic challenges ensuing from physical

properties/limitations of optical networks

SS 19, v 1.1.1 FI - Ch 2: Optical networking 2

SLIDE 3

Content

Optical devices
Optical networks
Optimization problems

SS 19, v 1.1.1 FI - Ch 2: Optical networking 3

SLIDE 4

SS 19, v 1.1.1 FI - Ch 2: Optical networking 4

Transmitters

Light-emitting diodes (LED, cheap)
Simple lasers
Work well, but downside: multiplexing 100 wavelengths needs 100

different simple lasers

Standard operation: 10s to 100s of wavelengths carried by a fibre
Tunable lasers (!)
Simplify practical logistics compared to simple lasers
Make reconfiguration of optical networks practical
Put n tunable lasers at a node for n lightpaths; pick wavelength
But: complex, expensive, only starting to be commercially

available

SLIDE 5

SS 19, v 1.1.1 FI - Ch 2: Optical networking 5

Directional Coupler

Combines and splits signals arriving on input towards output
Splitting ratio: Percentage of power going from one input to
pposite output
E.g., 3dB coupler: 50:50 power distribution; useful for switches
95:5 couplers for tapping, monitoring, …
Can be frequency-flat or frequency-selective
Phase is shifted by ¼/2 when crossing over to other arm

Input 1 Output 1 Output 2 Input 2 l (coupling length) Fibers or waveguides

Figure 3.1

A directional coupler. The coupler is typically built by fusing two fibers

together. It can also be built using waveguides in integrated optics.

From [1]

SLIDE 6

SS 19, v 1.1.1 FI - Ch 2: Optical networking 6

Isolators and Circulators

Isolator: only one-directional (unlike most optical devices)
Circulators: clever combination of isolators

From [1]

1 2 3 1 2 3 4 (a) (b)

Figure 3.3

Functional representation of circulators: (a) three-port and (b) four-port. The arrows represent the direction of signal flow.

SLIDE 7

SS 19, v 1.1.1 FI - Ch 2: Optical networking 7

Multiplexers and Filters

Multiplexers and filters:

In frequency domain

Demultiplexer

Multiplexer

Figure 3.7

A static wavelength crossconnect. The device routes signals from an input port to an output port based on the wavelength.

From [1]

Wavelength multiplexer

Wavelength

filter

(b)

(a)

SLIDE 8

SS 19, v 1.1.1 FI - Ch 2: Optical networking 8

Gratings: Mach-Zehnder Interferometer (MZI)

Couplers

combined with delay

Acts as filters
r

(de)multiplexer s

Input 1 Input 1 Output 1 Output 1 Output 2 Output 2 Input 2 Input 2 Path difference, L MZI ( ) L (a) (b) (c) Input 1 Input 2 MZI ( ) L MZI (4 ) L MZI (8 ) L MZI (2 ) L Output 1 Output 2

Figure 3.21

(a) An MZI constructed by interconnecting two 3 dB directional couplers. (b) A block diagram representation of the MZI in (a). L denotes the path difference between the two arms. (c) A block diagram of a four-stage Mach-Zehnder interferometer, which uses different path length differences in each stage.

From [1]

SLIDE 9

SS 19, v 1.1.1 FI - Ch 2: Optical networking 9

Gratings: Arrayed Waveguide Grating (AWG)

Think of AWG as an MZI generalized to multiple ports

$%&'( )*'&+,- .'(&'( )*'&+,- /--01,2 304,5'62,7 $%&'( 304,5'62,7 .'(&'( 304,5'62,7

Figure 3.24

An arrayed waveguide grating.

Arrayed

waveguide grating

Figure 3.25

The crossconnect pattern of a static wavelength crossconnect constructed from an arrayed waveguide grating. The device routes signals from an input to an output based on their wavelength.

From [1]

SLIDE 10

SS 19, v 1.1.1 FI - Ch 2: Optical networking 10

Acousto-Optical Tunable Filter

Make an AWG controllable by external input (of sound

waves)

From [1]

SLIDE 11

SS 19, v 1.1.1 FI - Ch 2: Optical networking 11

Optical switches

Goals:
Provision new lightpaths
Switch to backup in case of failure (“protection switching”)
Ideally: switch on individual packets

From [1] Time BETWEEN two packets

SLIDE 12

SS 19, v 1.1.1 FI - Ch 2: Optical networking 12

Example technology: MEMS-based switch

From [1]

SLIDE 13

SS 19, v 1.1.1 FI - Ch 2: Optical networking 13

Wavelength converters

Device to convert one wavelength into another wavelength
Needed to adapt between different optical technologies
… or to increase switching flexibility
Variations:
Fixed input, fixed output
Variable input, fixed output: all incoming wavelengths (in range)

converted to one specific one on output

Fixed input, variable output: Convert a specific incoming

wavelength into a selectable output one

Variable input, variable output: …

SLIDE 14

Content

Optical devices
Optical networks
Optimization problems

SS 19, v 1.1.1 FI - Ch 2: Optical networking 14

SLIDE 15

SS 19, v 1.1.1 FI - Ch 2: Optical networking 15

Optical networks – Client layers

The large times necessary to modify a switch make an
ptical network essentially circuit-switched
How to transport packets (e.g., IP) over such a network?
“Client layers” introduce intermediate functions
SONET/SDH: Historical from telco carriers; complicated time

division multiplexing hierarchy based on a fixed basic rate (51.84 Mbit/s); framing structure (125 µs, irrespective of line rate)

Optical Transport Network (OTN, G.709): designed to bridge gap

between optics and IP; partly similar in spirit to SONET; adds FEC, management, protocol transparency (carry any kind of payload packet, e.g., IP or Ethernet frames)

Ethernet: conceive of optical circuit as a link between (electrical)

Ethernet switches

MPLS

SLIDE 16

SS 19, v 1.1.1 FI - Ch 2: Optical networking 16

Optical wavelength division networks

Goal: provide optical channels / lightpaths (LPs) between

client nodes across network nodes

Lightpath: a direct flow of light from source to destination without

need to convert to electrical signals

Network nodes may switch and convert wavelengths
Effectively routing wavelengths (not packets)
Questions:
Which elements are needed?
How to configure switches, wavelength converters to fulfil given

demand?

SLIDE 17

SS 19, v 1.1.1 FI - Ch 2: Optical networking 17

WDM elements

The usual: terminals, amplifiers, wavelength conversion, …
Also: Optical Add/Drop Multiplexer (OADM)
Ideally: configurable; choose wavelength to add/drop

Add/Drop Node A Node B Node B Node C (a) Add/Drop Node A Node B Node C (b) Transponder OLT OADM Optical passthrough

Figure 7.4

A three-node linear network example to illustrate the role of optical add/drop multi-

plexers. Three wavelengths are needed between nodes A and C, and one wavelength each between

nodes A and B and between nodes B and C. (a) A solution using point-to-point WDM systems. (b) A solution using an optical add/drop multiplexer at node B.

From [1]

SLIDE 18

SS 19, v 1.1.1 FI - Ch 2: Optical networking 18

WDM Optical Crossconnect (OXC)

Generalizes and combines switches and OADM, terminals
May internally be electrical, pure optics, hybrid

IP ATM SONET SDH OXC OLT

Figure 7.10

Using an OXC in the network. The OXC sits between the client equipment

f the optical layer and the optical layer OLTs.

From [1]

SLIDE 19

SS 19, v 1.1.1 FI - Ch 2: Optical networking 19

WDM main elements

OLT OADM OXC 1 1 1 1 2 2 2 Lightpath A B C D E F X IP router IP router IP router IP router SONET terminal SONET terminal

Figure 7.1

A wavelength-routing mesh network showing optical line terminals (OLTs),

ptical add/drop multiplexers (OADMs), and optical crossconnects (OXCs). The network

provides lightpaths to its users, such as SONET boxes and IP routers. A lightpath is carried on a wavelength between its source and destination but may get converted from

ne wavelength to another along the way.

From [1]

SLIDE 20

WDM features

Wavelength reuse: Same wavelength used by different

LPs in different parts of network (spatial reuse)

Wavelength conversion
Results in lightpath topology
Transparency: LPs can carry different protocols, are

agnostic to protocol on top (act as physical link)

Circuit switching
Today, fairly static; making it dynamic is current research
WDM does not provide optical packet switching (not ready yet)
Survivability: Backup LPs can be configured

SS 19, v 1.1.1 FI - Ch 2: Optical networking 20

SLIDE 21

SS 19, v 1.1.1 FI - Ch 2: Optical networking 21

WDM Network Design

How to combine these elements into a network?
Example: Three nodes with IP traffic A-B, B-C, A-C, WDM

links of 10 Gb/s, total traffic 50 Gb/s per pair

Compare
Number of router

ports

Equipment cost
IP topology, and

resulting delay

Questions: How to

design topology, route, assign wavelengths?

A B C Router Router Router Router Router Router (a) (b) (c)

Figure 10.1

(a) A three-node network. (b) Nodes A–B and B–C are interconnected by WDM links. All wavelengths are dropped and added at node B. (c) Half the wavelengths pass through optically at node B, reducing the number of router ports at node B.

From [1]

SLIDE 22

Content

Optical devices
Optical networks
Optimization problems

SS 19, v 1.1.1 FI - Ch 2: Optical networking 22

SLIDE 23

SS 19, v 1.1.1 FI - Ch 2: Optical networking 23

Lightpath Topology Design

Assume
n IP routers, with at most ¢ ports each (¢ much smaller than n)
IP routers switch infinitely fast
All lightpaths are bidirectional
We know traffic rate for each source-destination pair, ¸sd
Traffic can be split over several LPs, if necessary (flow

assumption, infinitely fine splits possible)

Reasonably matched by IP traffic, but still a simplification
Determine
Lightpath topology: bij 2 {0, 1}, indicating existence of LP between

nodes i, j

Note: We are not talking about optical switches yet; just which LPs we need

SLIDE 24

SS 19, v 1.1.1 FI - Ch 2: Optical networking 24

LPD – Notation & goal

Goal: Minimize congestion
Alternative goals??

From [1]

„link“ from IP prespective; lightpath from optical network perspective

SLIDE 25

SS 19, v 1.1.1 FI - Ch 2: Optical networking 25

Constraints (flow problem, mixed integer linear program)

MILP is NP hard
Try modified (ordered) LP relaxation for solution
Realistic? IP forwarding capacity? Can resulting bi,j be

realized?

From [1]

SLIDE 26

SS 19, v 1.1.1 FI - Ch 2: Optical networking 26

Routing and wavelength assignment

Suppose we are given a set of lightpaths to setup
E.g., from the LPD solution
Question: Can we actually realize it?
For each lightpath from node i to j, we must determine
A route: via which intermediate switches does it travel, using

which physical links

For each link it travels: the wavelength to use
Various constraints
Wavelength conversion capability assumed for the network
E.g.: lightpath keeps the same wavelength across all links; lightpath

can change wavelength at each link

E.g., at no link must any wavelength be assigned to different

lightpaths

SLIDE 27

SS 19, v 1.1.1 FI - Ch 2: Optical networking 27

RWA – Example: NSF network, 5 wavelengths

1 2 3 4 5 6 7 8 9 10 11 12 13 14

From [3]

SLIDE 28

SS 19, v 1.1.1 FI - Ch 2: Optical networking 28

RWA – Example: NSF network, 5 wavelengths

Lightpath from node 5 to node 13 (5 ⇒ 13)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 From [3]

SLIDE 29

SS 19, v 1.1.1 FI - Ch 2: Optical networking 29

RWA – Example: NSF network, 5 wavelengths

Conflict with demand 1 ⇒ 12: Use different frequencies

1 2 3 4 5 6 7 8 9 10 11 12 13 14 From [3]

SLIDE 30

SS 19, v 1.1.1 FI - Ch 2: Optical networking 30

RWA – Example: NSF network, 5 wavelengths

Conflict with demand 1 ⇒ 12: Use different path

1 2 3 4 5 6 7 8 9 10 11 12 13 14 From [3]

SLIDE 31

SS 19, v 1.1.1 FI - Ch 2: Optical networking 31

RWA – Example: NSF network, 5 wavelengths

Conflict with demand 1 ⇒ 12: Reject demand

1 2 3 4 5 6 7 8 9 10 11 12 13 14 From [3]

SLIDE 32

SS 19, v 1.1.1 FI - Ch 2: Optical networking 32

Jointly solving routing and wavelength assignment

Turns into an (M)ILP again, with many variations
Example: known, static lightpath requests; no wavelength

conversion at all;

Goal: realize traffic matrix/LPD with as few wavelengths as

possible

Or minimize number of used ports, …
Problems:
Finding edge-disjoint paths is hard
There might be multiple fibres

between two nodes

“Arcs” in a multi-graph
Wavelength reuse between arcs

(a) Physical network. (b) Optical network.

Fig. 1. Multigraphs and multifiber optical networks.

From [2]

SLIDE 33

SS 19, v 1.1.1 FI - Ch 2: Optical networking 33

RWA as MILP: Notation

Notation
Multigraph G=(V,E)
¤ is set of wavelengths per arc, W = |¤ |
Traffic matrix (or set of lightpaths) T with K requested connections,

each connection requests data rate of exactly one wavelength

Goal: For each accepted connection, find a lightpath (p, ¸)
No clashes of lightpaths on any arc
Decision variables:

From [2]

SLIDE 34

SS 19, v 1.1.1 FI - Ch 2: Optical networking 34

RWA as MILP: Constraints in a link formulation

max zKS1(x) =

k⇧K

xk subject to:

e⇧ω+(vi)

xλ

ke =

e⇧ω(vi)

xλ

ke

k ⇧ K, λ ⇧ Λ, vi ⇧ V \ {sk, dk} (1)

e⇧ω+(sk)

xλ

ke

e⇧ω(sk)

xλ

ke = xλ k

k ⇧ K, λ ⇧ Λ (2)

e⇧ω(dk)

xλ

ke

e⇧ω+(dk)

xλ

ke = xλ k

k ⇧ K, λ ⇧ Λ (3)

k⇧K

xλ

ke ⇤ 1

e ⇧ E, λ ⇧ Λ (4)

λ⇧Λ

xλ

k = xk

k ⇧ K (5) xλ

ke ⇤ xλ k

k ⇧ K, e ⇧ E, λ ⇧ Λ (6) xk, xλ

k , xλ ke ⇧ {0, 1}

k ⇧ K, e ⇧ E, λ ⇧ Λ. (7)

From [2]

Wavelength continuity, similar to flow conservation in multi-commodity flow problem Maximize number of accepted connections No two lightpaths sharing a link can have same ¸ Connection has only one ¸

SLIDE 35

SS 19, v 1.1.1 FI - Ch 2: Optical networking 35

RWA as MILP: Variations & Solution approaches

Many variations of this problem formulation exist
E.g., incorporate wavelength conversion (arbitrary or limited)
E.g., demands for more/less than just one single wavelength
E.g., hop/delay constraints
E.g., backup path for each primary path (edge- or node-disjoint)
Formulate via sources or entire paths, not via flows on edges
Solution approaches
MILP solution in general NP complete
Common technique: Linear relaxation
Turn the integer or 0/1 variables into real numbers, solve the resulting

linear program (simplex algorithm)

Round the real solution back to integers; take care not to violate

constraints

Often works reasonably well here

SLIDE 36

SS 19, v 1.1.1 FI - Ch 2: Optical networking 36

Dimensioning a WDM network

Suppose we can solve, for given network and traffic

matrix, the LTD/RWA problem

An online problem, solve it when demand changes
Related problem: How to dimension a WDM network to

serve traffic requests?

An offline, planning problem – long time scales
Where to run links (fibres)?
How many wavelengths to provision on each link?
For a static traffic matrix: Very similar to LTD/RWA
What happens if traffic matrix only known stochastically? How to

plan a network, then?

SLIDE 37

SS 19, v 1.1.1 FI - Ch 2: Optical networking 37

Dimensioning WDM for statistical traffic models

Class 1: First-passage model
Assume network is empty, lightpath requests arrive according to

some stochastic model

Lightpaths do not depart => network will become overloaded at

some time T

Goal: Postpone time of overload T as far as possible
Formally: With high probability p, all requests in [0,T] can be fulfilled
Useful in particular for upgrade decisions
Variant: some LPs do depart, but LPs come in faster than they

depart

SLIDE 38

SS 19, v 1.1.1 FI - Ch 2: Optical networking 38

Dimensioning WDM for statistical traffic models

Class 2: Blocking model
Lightpaths come and go; arrival rate = departure rate
Network is in equilibrium
Load metrics
Offered load: arrival rate * average duration (avg. number of LPs in

system, compare Little’s Law)

Or normalized to amount of wavelengths:

reuse factor = offered load / number of wavelengths in network

Goal: Try to serve as many requests as possible
Ensure that probability of blocking a lightpath request is below a given

bound, say, 1%

Determine maximum offered load that the network can sustain while

meeting this bound on blocking probability

More flexibility in LPs than what is commonly provided today

SLIDE 39

SS 19, v 1.1.1 FI - Ch 2: Optical networking 39

Dimensioning problem not tractable

Dimensioning problems as outlined above not easily

tractable analytically

Only under very strong, impractical limiting assumptions
Turn to simulation
Input: Number of nodes/network graph, arrival rate ¸, LP lifetime

1/µ, number of wavelengths

For W wavelengths, rewrite the graph into W copies, with additional

nodes for source/destination proper

Differs without/with wavelength conversion
Output: Blocking probability, reuse factor
Algorithm (e.g.): For each LP request, do shortest path routing on

first available wavelength

SLIDE 40

SS 19, v 1.1.1 FI - Ch 2: Optical networking 40

Some analytic expectations, observations

We can make some educated guesses about what to

expect!

Look at blocking probability of an LP, consider only a

single link, average load ½ = ¸ / µ

This turns into a model with Poisson arrivals, W servers, no

queue: M/M/W/0 queuing model!

Blocking probability is known:

Erlang-B formula

Now vary number of wavelengths W and load

proportionally

What happens?
Effect is called trunking efficiency

SLIDE 41

SS 19, v 1.1.1 FI - Ch 2: Optical networking 41

Some analytic observations – Independence assumption

Assume wavelengths on a link are free or used (with

probability ¼), independent of each other/of situation on

ther links
Consider blocking probability for an LP request of H hops
Without wavelength conversion
With full wavelength conversion

SLIDE 42

SS 19, v 1.1.1 FI - Ch 2: Optical networking 42

Some analytic observations – Independence assumption

Look at this for small blocking probabilities
Consider ratio of the two usage probabilities, expresses

gain by wavelength conversion

So-called conversion gain
Consequence: Gain of wavelength conversion increases

with the average hop count / the network size

SLIDE 43

SS 19, v 1.1.1 FI - Ch 2: Optical networking 43

Required number of wavelengths

From [1]

SLIDE 44

SS 19, v 1.1.1 FI - Ch 2: Optical networking 44

Outlook: Photonic Packet Switching

So far (and current technology): Lightpaths are setup like

circuits, switched like circuits

Packet switching via electronics: Trivial, but squanders the

benefits

Can we switch on packets, using only optical technology?
Challenges
Process packet header?
Buffering?
And others (multiplexing, synchronization, …)

SLIDE 45

SS 19, v 1.1.1 FI - Ch 2: Optical networking 45

Header processing

Header has constant size
Process the header electronically, buffer payload of the

packet until done, then switch to output port from buffer

Needs electronically controlled switches, e.g., a device where

grating/coupling can be controlled electronically at very high speeds

SLIDE 46

SS 19, v 1.1.1 FI - Ch 2: Optical networking 46

Buffering

Buffering needed:
Header processing – fixed delay
When output port is busy (buffer at input or output) – variable delay
How to buffer a photon, which only exists when it moves?
… Give it a long way to move – a delay line
Example: 200 m = 1 µ s, corresponds to 10 packets of size 1000 bits

at 10 Gbit/s

Figure 12.15

Example of a ! × ! routing node using a feed-forward delay line architecture. Figure 12.18

Example of a !× ! routing node using a feedback delay line architecture.

From [1]

SLIDE 47

SS 19, v 1.1.1 FI - Ch 2: Optical networking 47

Conclusion

Optical networking provides powerful tools and unique

challenges to construct, dimension, and operate a network

Current practice is still circuit-switched networking
Burst/packet switches still on-going research
Integration between the network fabric as such and control

logic complex

Both from a (signalling) protocol as from an algorithmic

perspective

Mutual adaptation between traffic matrixes and networking not

trivial (part of our own research)

SLIDE 48

SS 19, v 1.1.1 FI - Ch 2: Optical networking 48

References

1. R. Ramaswami, K. N. Sivarajan, G. H. Sasaki, Optical

Networks, A Practical Perspective. Morgan Kaufmann, 2010.

2. B. Jaumard, C. Meyer, and B. Thiongane, “Comparison of

ILP formulations for the RWA problem,” Optical Switching and Networking, vol. 4, no. 3–4, pp. 157–172, Nov. 2007.

3. H. Simonis, Using Mixed Integer Linear Programming