Optical Packet Switching

Techniques
Walter Picco
MS Thesis Defense December 2001

Fabio Neri, Marco Ajmone Marsan

Telecommunication Networks Group
http://www.tlc-networks.polito.it/
1
 Overview
• Introduction and motivations

• Goals of the thesis

• State-of-the-art and enabling technologies

• SIMON: an optical network simulator

• Optical networks design

• Obtained results

2
 The need of optics
Future network requirements:
• High bandwidth capacity
• Flexibility, robustness
• Power supply and equipment footprint
reduction

Optics offers a good evolution perspective

3
 Optical framework today
• Point to point communications
• Circuit switching with packet switching
electronic control

why ?

• Optical packet switching:

– no optical memories
– slow optical switches
4
 Optical packet switching
Bandwidth is not a problem

Network cost is in the commutation

New protocols and architectures needed

• New tools to measure performance

• New design techniques
5 more
 Overview
• Introduction and motivations

• Goals of the thesis

• State-of-the-art and enabling technologies

• SIMON: an optical network simulator

• Optical networks design

• Obtained results

6
 Goals
• New optical network simulator

Topology Simulation Performance

7
 Goals
• New analysis and design method for optical
networks

Resources Analysis Topology

8
 Overview
• Introduction and motivations

• Goals of the thesis

• State-of-the-art and enabling technologies

• SIMON: an optical network simulator

• Optical networks design

• Obtained results

9
 Transmitting data
Wavelength Division Multiplexing: the huge
bandwidth of an optical fiber is divided in many
channels (colors)

Each channel occupies a

different frequency slot
10
 Storing data
• Optical RAM is not available yet
• Fiber Delay Lines (FDLs) are used instead

FDLs

FDL

Forward usage Feedback usage

11
 Processing data
• Electronics limits the speed in data forwarding
• Optical 3R regeneration (and wavelength
conversion) is now possible

1 2
3R

• Physical layer is not a matter of concern

• All-optical solutions are currently at the study

12
 Switching data
• Today: Semiconductor Optical Amplifiers

• Tomorrow (a
possibility): Micro
Electro Mechanical
Systems

13
 Overview
• Introduction and motivations

• Goals of the thesis

• State-of-the-art and enabling technologies

• SIMON: an optical network simulator

• Optical networks design

• Obtained results

14
 The starting simulator: CLASS
• Simulator of ATM networks
• Topology independent

Adaptable tool
• Fixed routing
implementation } fiber
channel
Not good for WDM
15
 CLASS modifications
• Dynamic routing strategy
} fiber
channel

• Each WDM channel must be listed in the

network description file

Maximum flexibility in the network description

16
SIMON node architecture
3R
3R

1 1
3R

2 2 3R
3R
m m

1 1 3R

n-1 n-1

n n

3R
3R

SWITCH 3R
CONTROL
UNIT 17
 Time division
• Slotted network:

timeslot

P2
C1 t
P1
C2 t
C3 t
t0 t1

18
 Overview
• Introduction and motivations

• Goals of the thesis

• State-of-the-art and enabling technologies

• SIMON: an optical network simulator

• Optical networks design

• Obtained results

19
 Designing WDM networks
• Given:
Network topology and the traffic matrix
• Find:
Number of WDM channels on each link
• Optimizing:
Network throughput
• Meeting a cost constraint:
Network cost Fixed number
 of ports for all
commutation the switches
20
 The optimization problem
• Mathematical statement:
Find minimum (maximum) of a non-linear
function in the discrete domain, meeting
some constraints

NP-complete problem

Only heuristic solutions are possible

21
 Proposed approach
1)Find:
Ptot  f  n1  nM 
– Ptot : packet loss probability of the whole
network
– ni : number of WDM channels on link i

2)Elaborate a heuristic solution to find the

minimum of Ptot

22
 Link model
• Classical queueing theory: M/M/L/k queue
k
1

buffer L
servers
• server  WDM channel
• buffer slot  FDL
23 more
 Node model

Input Output
fibers fibers

24
 Model limitations
• FDLs can’t be modeled as a simple buffer
– discrete storage time
– noise addition at each recirculation

FDL  SNRB  SNR A

B A 
channel  t B  t A  t

• All the FDLs of a node are shared among the

different queues
25
 Network model
• The packet loss probability (Pf) of a flow is:


Pf  1   1  P i 
iL f
• The packet loss probability (Ptot) of the whole
network results:

t P f f

Ptot  f  n1  nM 
f F
Ptot 
t
f F
f

• First step completed

26
 Searching the minimum

Storage capacity
(number of FDLs)
Level
Network connectivity
(number of channel
ports)
• Cost constraint:
(channel ports + FDLs ports) = constant
• optimum balance  optimum solution
27
 Heuristic approach
• Starting topology: maximum connected

Highest possible level

• Iteration steps:
– the current topology is perturbed
– if the perturbed topology has a lower Ptot
the topology is modified
28
 Heuristic approach
• Topology perturbation:
– all the links are analyzed

added

cancelled
– the link that modified gives the lower Ptot is
memorized
29
 Overview
• Introduction and motivations

• Goals of the thesis

• State-of-the-art and enabling technologies

• SIMON: an optical network simulator

• Optical networks design

• Obtained results

30
 General backbone: topology
Node
6 7
User

5 1 2 8

12 4 3 9

11 10

31
General backbone: throughput
Fraction of packets successfully transferred

0.95

0.9 1
2
3
4
M/M/L/k (4 MR)
M/M/L/k (  MR)
0.85
0 2 4 6 8 10 12 14 16 18

Total network load [Gbps] 32

 General backbone: delay
9

8 1
2
7 3
4
M/M/L/k (4 MR)
Packets net delay

6
M/M/L/k (  MR)
5

0
0 2 4 6 8 10 12 14 16 18

Total network load [Gbps] 33

 USA backbone: topology
1 8 17 23
5 9 14
22
2 4 6 10 15 18 24

11 16 19 25
3

7 12 20 26

13 21 27

28
34
Fraction of packets successfully transferred USA backbone: throughput
1

0.98

0.96

0.94

0.92

0.9
1
2
3
0.88
M/M/L/k (4 MR)
M/M/L/k (  MR)
0.86
0 5 10 15 20 25 30 35 40

Total network load [Gbps] 35 more

 Conclusions
Two key elements:

• A new tool capable to simulate the next

generation optical networks

• A new optimization target in the optical

networks design giving good results

36 more
ES

37
 Optical Burst Switching
• Packets are assembled in the network edge,
forming bursts
• Advantages:
– More efficient exploitation of the bandwidth
– Possibility to implement Service
Differentiation
• Disadvantages:
– More complicated network structure
– More complicated forwarding process
38
continue
 Link model
• Packet loss probability P on the link:
–  link capacity
–  link traffic load
– offered load [Erlangs],

 
 r
 L
  L 1  1 r 
i k 1
1

k    if r  1
  L
 i 0 i! 1 r 
P 0   0   1
L!  L    L 1 i
 L

   k  1  if r  1
 i 0 i! L! 
39
continue
Japan backbone: topology
1

2 3
4

5 6

8
7

9 10

11 40
 Japan backbone: throughput
Fraction of packets successfully transferred

0.99

0.98

0.97

0.96

0.95

0.94

0.93
1
0.92 2
3
0.91 M/M/L/k (4 MR)
M/M/L/k (  MR)
0.9
0 5 10 15 20 25 30

Total network load [Gbps] 41

continue
 Future work
• Simulator:
– Support for different architectures
– FDLs of variable length

• Heuristic approach:
– More detailed model for FDLs

42
continue
End of presentation
43