Presentazione di PowerPoint - E

Download Report

Transcript Presentazione di PowerPoint - E

This teaching material is a part of e-Photon/ONe Master study
in Optical Communications and Networks
Course and module:
Photonics in Switching
Author(s):
Lena Wosinska
Royal Institute of Technology
[email protected]
No part of this presentation can be reused without the permission of author(s).
Users are requested to ask for permission by specifying the purpose of the usage.
http://www.e-photonone.org
The aim of this module
• To show principles for optical circuit
switching, burst switching and packet
switching
• To highlight the main technological problems
• To give an overview of the optical switching
node architectures
Revision: date
2( )
Outline
• Introduction
• Optical circuit switching (OCS)


Solving LTD and RWA problems
WDM network elements
• Optical packet switching (OPS)



Functions of an optical router
Contention resolution
OPS architectures
• Optical burst switching (OBS)
• Summary
Revision: date
3( )
Switching
• Needed to efficient utilize the network resources
• Full mesh connectivity vs. switched connectivity
• Resource sharing
Types
• Circuit Switching
• Packet switching
• Cell switching



a kind of packet switching
fixed packet size (e.g. ATM cells)
uses virtual circuits (VCS), routing decisions - during virtual circuit
setup.
Revision: date
4( )
Switched networks
1
Switching nodes

not concerned with contents of data
purpose: provide switching facility
in general not fully connected


End nodes

provides data to transfer
connected via switching nodes

A
Links

physical connections between nodes
Revision: date
5( )
Switching
Switching
Circuit
Switching
Packet
Switching
Connection Oriented
(Virtual Circuit)
Datagram
Revision: date
6( )
Photonics in switching
• Optical circuit switching (OCS)

Relatively mature technology today

Providing lightpaths

WDM network elements
•
OLT, OADM, OXC
• Optical packet switching (OPS)

Not available today due to some technological problems
•
Controllable optical memory for optical buffering
•
Control functions in the optical domain
•
Synchronization, etc
• Optical burst switching (OBS)

Hybrid packet switching: a feasible solution?
Revision: date
7( )
Optical circuit switching
Wavelength-Routed Networks
Revision: date
8( )
Wavelength-Routed Networks
• W-R switches
• Provide Lightpaths
• Problems:
• circuit switching

low bandwidth
efficiency
• Large granularity
Network
1
Wavelength-routing switch
A
Access (client) node (e.g., IP router):
contains (tunable) transmitters and receivers
Revision: date
9( )
Circuit switching
• Advantages

Once connection is established
•
Network is transparent, nodes seem to be directly connected  no variable
delay
• Drawbacks

Resources are permanently allocated to connection
•
Low utilisation for

Traffic with changing intensity or

Short lived connections

Connections with low capacity requirements
•

Grooming: Aggregation of tributaries
Delay before connection setup.
Revision: date
10 ( )
Optical circuit switching
• Solving Lightpath Topology Design (LTD) and Routing and
Wavelength Assignment (RWA) problems
• A lightpath corresponds to a circuit

Set-up a lightpath

The whole lightpath is available during the connection

Disconnect
• Network elements

Fiber

Optical line amplifier (OLA)

Optical line terminal (OLT)

Optical add-drop multiplexer (OADM)

Optical cross-connect (OXC)
Revision: date
11 ( )
Offline LTD and RWA: Objective
Seattle
New York
Given a traffic matrix (a forecast) and a fiber (physical) topology:
design the network that fits the traffic forecast
Revision: date
12 ( )
Definitions
• Lightpath topology:

Logical (virtual) topology
• Physical topology:

Fiber topology
• Grooming:

Kind of time multiplexing: packing a low speed channels into
higher speed channels. This function is provided by digital crossconnects (DCS)
Revision: date
13 ( )
The issues
• The WDM network transports data traffic from the client layer
• The WDM network is to establish lightpaths to support the client
traffic demands



Given the fiber (physical) topology
Determine the lightpath (virtual or logical) topology (Lightpath
topology design: LTD)
Routing and wavelength assignment (RWA)
• Traffic demands are expressed in a traffic matrix T=[lsd]

lsd is the demand in packets/second between source node s and

destination node d
The traffic matrix is obtained by forecasting
Revision: date
14 ( )
Heuristic solution
• Hard to determine the lightpath topology jointly with
the routing and wavelength assignment
• Split into separate LTD and RWA problems

Solve the LTD problem and then realize the obtained LTD
within the optical layer (i.e. for the obtained LTD solve RWA
problem).
Revision: date
15 ( )
The lightpath topology design (LTD)
• Given



The traffic demands
The maximum number of ports per client node, D
Lightpaths interconnect client nodes bidirectionally
• Determine the topology and routing of packets

Objective: Minimize the maximum load that any lightpath must
carry
Revision: date
16 ( )
LTD: mathematical formulation
• Assume n source and destination nodes


Up to (n-1)n connection demands (lij = 0, i=j)
Traffic between s and d might use several lightpaths
•

Conservation of flow s-d: rate lsd at source, –lsd at the destination,
and 0 at all intermediate nodes
Each lightpath starts and terminates at ports in client nodes
•
bij = 1 if nodes i and j are interconnected by a lightpath, 0 otherwise
• Goal: Find the bij under the objective
min lmax , lmax = max i , j  s ,d l ,
sd
ij
Revision: date
17 ( )
LTD computation
• This is a mathematical program



Linear if unconstrained (LP)
Mixed integer linear program (MILP) when

l’s are real, positive, and
•
bij are integer {0, 1}
Computational complexity an issue
• Heuristic approach



Assume bij real valued [0, 1]
Sort them in decreasing order
Round them to 0 or 1 while meeting degree requirements
Revision: date
18 ( )
RWA: problem definition
How to map the lightpath topology
to the fiber topology

Undirected or directed edges
•


Can one wavelength channel be used in both directions?
Lightpaths cannot share wavelength channels on a given link
Wavelength conversion
•
•
•
Allows the lightpath to use different wavelengths
No conversion: One wavelength for the entire path
Limited conversion


From a limited input set to a limited output set
 In some but not all nodes
Other constraints
•
•
Signal regeneration
Survivability
Revision: date
19 ( )
Objective

Realize the lightpath topology, meet all constraints

Minimize the number of wavelengths used per
link

Offline: For all lightpaths determined by the LTD

Online: For demands coming during operation
Revision: date
20 ( )
RWA approaches
• ILP formulation
• Heuristic


Routing sub-problem
Wavelength assignment sub-problem
Revision: date
21 ( )
RWA: ILP formulation
Minimize:
Such that:
Wmax
Wmax   W ijsd ij
s ,d
sd
sd
W

W
 ij  jk
i
k
if
 l sd

=   l sd
if
 0 otherwise

l sd :
W ijsd :
s= j
d = j
# of lightpaths between s and d.
# of lightpaths on link ij that is
following from s to d.
Revision: date
22 ( )
Routing sub-problem
•
•
•
•
Fixed routing
Fixed-alternate routing
Adaptive routing
Fault-tolerant routing
Revision: date
23 ( )
Wavelength assignment sub-problem
• Static WA sub-problem

Graph coloring
• Dynamic WA sub-problem

Random (R) Wavelength Assignment

First-Fit (FF)

Least-Used (LU)/SPREAD

Max-Used (MU)/PACK

Least loaded (LL)
etc.
Revision: date
24 ( )
Static (offline) WA: Graph coloring
• No wavelength converters  Wavelength continuity constraint
• Given: all connections and routes
• Objective: assign wavelengths (colors) to each lightpath so as to
minimize the number of wavelengths used under the wavelength
continuity constraint
• Construct a graph G so that each lightpath in the system is
represented by a node. Undirected edge between nodes in the G if
the corresponding lightpaths share a physical link
• Color the nods of G such that no two adjacent nodes have the
same color.
Revision: date
25 ( )
Dynamic (online) WA algorithms
•
•
Random (R)

Determine the set of wavelengths that are available at a given route

Pick one with uniform probability
First-Fit (FF)


•
Least-Used (LU)/SPREAD

•
Select the wavelength that is the least used in the network
Max-Used (MU)/PACKED

•
All wavelengths are numbered
Assign the first available wavelength
Select the most used wavelength in the network
Least-Loaded (LL) designed for multi-fiber networks

Select the wavelength with largest residual capacity on the most loaded link
along the connection.
Revision: date
26 ( )
WA: Alternate routing lowers blocking
•
Random-1: (no alternative routes)

•
Random-2:



•
Fix two shortest paths between every s-d
Choose at random one of the available wavelengths on the first shortest path
If no such wavelength is available choose a random one on the second shortest
path
Max-used-1:

•
Choose at random one of the available wavelengths on a fixed shortest path
between s-d
On a fixed shortest path between s-d choose the available wavelength that is
used most in the network at that time
Max-used-2:


Use the wavelength for first path that is used most in network
Try second path when first path is full
Revision: date
27 ( )
Dimensioning wavelength routing networks
Common practice
Revision: date
28 ( )
Deterministic vs. Statistical dimensioning
Deterministic network dimensioning:
•
Forecast traffic demands
•
Solve the LTD
•
Solve the RWA (network dimensioning problem)
•
Iterate every 6 to 12 months
•

Upgrade the network to meet all demands

Constrain RWA not to disturb established lightpaths
Deterministic setting
Alternatively:
Statistical dimensioning using statistical models
•
First-passage model
•
Blocking model
Revision: date
29 ( )
RWA Statistical dimensioning: First-passage model
Transient network state
• Network starts without any lightpaths
• Demands arrive randomly and lightpaths are set up one by one

Lightpaths might be terminated
•


Rate of termination strictly less than rate of arrivals of demands
Number of lightpaths increase
Eventually, a demand cannot be met
Dimensioning of WDM network
• Blocking should not occur before a given time
• Time chosen to be long enough for network upgrade
• Goal is probabilistic
• Problem with tractability
Revision: date
30 ( )
RWA. Statistical dimensioning: Blocking model
Assume stochastic equilibrium
•
•
Arrival rate strictly smaller than departure rate
Determine maximum offered traffic


•
Reuse factor: offered load per wavelength at given blocking probability.
Depends on




•
Given a upper bound on blocking probability
Usually assumes arrival as Poisson process, lightpaths with exponentially
distributed durations and uniform traffic distribution
network topology
traffic distribution
actual RWA algorithm
number of wavelengths available
Usually evaluated by simulation
Revision: date
31 ( )
Analysis problem vs. design problem
•
For the statistical models the analysis problem is easier to solve than a
design problem.
•
For a blocking model:

•
Easier to calculate blocking probabilities on each link given link capacities and the
traffic model than to determine link capacities given a blocking probability
constraint.
For First-passage model:

Easier to calculate the statistics of the time to first blocking given link capacities
than to determine link capacities to achieve a pre-specified first-passage time.
Revision: date
32 ( )
Maximum load dimensioning models
• With full wavelength conversion the problem is identical to the
dimensioning of circuit-switched networks with time slot
interchange
• Maximum load dimensioning models to determine how
limitations in conversion affects the dimensioning


The same set (or sets) of lightpaths supported
•
Offline requests (static network design problem)
•
Online requests (dynamic network design problem)
Determine the increase in number of wavelengths to
compensate for lack of conversion
Revision: date
33 ( )
WDM network elements
Optical line terminals
(OLTs)
WDM network example
Optical add-drop
multiplexers (OADMs)
Optical cross-connects
(OXCs)
Optical line amplifiers
Revision: date
34 ( )
Optical line terminal (OLT)
• Transponders

Adaptation from/to access
links
•
•
•

Wavelength and fiber
Overhead and FEC
BER monitoring
Main cost of OLT
• Multiplexer

To merge the incoming
channels
• Amplifier (optional)
• Optical supervisory channel
(OSC)

Added and terminated
Revision: date
35 ( )
Optical add/drop multiplexer (OADM)
• Drop and add one or more wavelength channels
 To and from equipment at local node
 Remaining channels pass transparently
• Channel selection
 Any channel or only some
 Static  Requires careful planning
 Reconfigurable: software configurable remotely
 One, a few or any number of channels
•
•
Modularity
Loss dependence on number of dropped channels
Revision: date
36 ( )
OADM Structures
•
•
Parallel structure

Arbitrary channels

Fixed loss

Demux of all channels
•
Inefficient for small drop
•
Higher loss
•
Many filters
Serial structure

Each OADM drops one channel

Modular cost
•
More modules for more drops
•
Disrupts service
•
Loss increases
Revision: date
37 ( )
Reconfigurable OADMs
• Fixed transponders
 Deployed but might not be
used
 Constrains planning
• Tunable transponders
 Send on any wavelength
 Receive any wavelength
Revision: date
38 ( )
Ideal OADM
• Would drop and add any channel

And any number of channels
• Remotely controlled


Reconfiguration without disturbance to unaffected channels
No plan-ahead needed
• Low and fixed loss

Independent of set of wavelengths dropped
Revision: date
39 ( )
Optical Cross-connects (OXC)
• Switching of wavelengths channels



From input to output ports
From input to output wavelengths
Does not include input/output OLTs
• Functions


Provide lightpaths
Protection switching (rerouting)
•

Performance monitoring
•
•


Restoration of failed lightpaths
Access to test signals
Bridging: Split signal
Wavelength conversion
Multiplexing and grooming
•
•
Aggregate tributary data streams
Only possible in the electronic domain
Revision: date
40 ( )
Types of OXCs
• Electronic or optical core



Transparency
Cost
Size
• Transparent or opaque core


Signal converted
Electronic signal
•

Signal monitoring
Optical signal
•
Transparency
Revision: date
41 ( )
All-optical OXCs
• Transparency at the cost of

Grooming
•

Wavelength conversion
•

Higher demand for lightpaths
 No aggregation of low bitrate demands
Higher blocking of lightpath demands
Signal regeneration
•
More constrained routing of lightpaths
• Switching wavelengths separately


Switch core composed of smaller switches
Additional switch needed for local add/drop
•
Size WF x T for non-blocking access to any T channels
Revision: date
42 ( )
Ex. 1: All-optical OXC. Clos architecture
Three stage strict internal non-blocking Clos architecture.
Size: 128x128
L. Wosinska et al.: ”Large Capacity Strictly Non-Blocking OXCs Based on MEOMS Switch
Matrices. Reliability Performance analysis,” IEEE/OSA JLT,
Vol.19,
No.8, Aug.
Revision:
date
43 (2001
)
Ex. 2: All-optical OXCs. WR architecture.
Strict internal non-blocking wavelength routing architecture.
Size 128x128
L. Wosinska et al.: ”Large Capacity Strictly Non-Blocking OXCs Based on MEOMS Switch
Matrices. Reliability Performance analysis,” IEEE/OSA JLT,
Vol.19,
No.8, Aug.
Revision:
date
44 (2001
)
Shortcomings with circuit switching
• Much idle time  low utilisation of links
• Constant data rate  hard to connect devices with different speeds
• Circuit-switched network are more sensitive to faults

Common synchronization

If a part of the connection fails, the whole transfer fails
Solution: Packet Switching

Data transmitted in packets
•

Large messages broken up into a number of packets
Store and forward
•
A node receives, stores and transmits packets to next node
Revision: date
45 ( )
Optical packet switching
The third generation optical networks ?
Revision: date
46 ( )
OPS Networks
•
•
•
•
Large capacity
High bandwidth efficiency
Rich routing functionalities
Great flexibility and reliability
Revision: date
47 ( )
An Example
Revision: date
48 ( )
Packet switching
• Advantages

Link capacity dynamically shared
•
•

For traffic with changing intensity
Connections with different bandwidth needs
Well suited for data communication
• Drawbacks

Packets stored and delayed at the switching nodes
•
•
Larger and variable end-to-end delay
Packets are lost when packet buffers are filled
Revision: date
49 ( )
Optical packet switching
• Advantages

Complements WDM
•
•
•
Allows grooming in optical domain
Allows statistical multiplexing
 Can improve bandwidth utilization within the optical layer
Increase flexibility
• Problems

Technological problems
•
•
•

Optical control functions
Synchronization
Optical buffering
High complexity
•
High cost
•
Low reliability
Revision: date
50 ( )
Functions of a packet switch (router)
•
•
•
Receives packet on input link
Processes the header
Decides where to forward the packet

•
•
•
Address look-up  output link and output (wavelength) channel
Sets up the switch matrix.
Stores packet until the output link is free
Transmits packet
inputs
outputs
IC
IC
IC
OC
switch
fabric
OC
OC
controllers Revision: date
51 ( )
Optical router (OPS node): Needed functions
•
Decoding of packet header

•
Could be electronic: header encoded at lower bit rate
Setup of switch fabric


Packet delayed until setup done (a fixed delay)
Setup requires scheduling of packets from all inputs
•

•
Fast reconfiguration of fabric (200 ns for 250 byte packet at 10 Gb/s)
Synchronization: Elastic buffering of packets to align packets at all inputs

Only needed when switch fabric is synchronous
•
•
•
Simplified for fixed packet size and synchronized operation
Synchronous fabric has better throughput
Multiplexing of lower-speed streams (and reverse operation, i.e. demultiplexing)
Contention resolution (e.g. buffering of packets if output busy)
Revision: date
52 ( )
Contention resolution in OPS networks
• Congestion is inherent in packet switching
• Contention may be dealt with in

Time

Wavelength

Space
• Electronic packet switching typically rely on the time domain by
means of queuing

Queuing in optics is not feasible

Queuing may be “emulated” by delaying packet in fiber loops
• What about optical packet switching ?
Revision: date
53 ( )
If output busy: Handling packet contention
• Drop a packet

Packet loss probability can be high even at moderate loads
• Deflect the packet


Send it on a free output
Restrict the deflection
•
•
•


Output that leads to destination
Output with a route to the destination that is at most m hops longer
Also called hot-potato routing
Increases delay and network load
Creates variable delays and potentially reordering
• Change the wavelength (TWC)


Chose a wavelength available at the output
All-optical tunable wavelength converters not available yet
• Buffer the packet

Store the packet until the output is available

Applying TWC may allow for decrease of the buffer size
Revision: date
54 ( )
Contention resolution techniques
•
Bufferless architectures


•
Deflection routing
TWC
Optical buffers

Placement at a node
•
•
•


Output buffer
Input buffer
Recirculation buffer
Dedicated or shared buffers
Technology
•
•
FDLs
Novel optical memories
 EIT
 Opt. resonators
Revision: date
55 ( )
Special Features
• Deflection Routing

A contention resolution scheme in case of no optical buffer
• Packet Aggregation


To relieve the optical packet switches of the heavy burden of
processing each individual (small) packet
Possible schemes: mixed-flow, per-flow
Revision: date
56 ( )
Buffer placement at the node
IC
Input buffer
IC
•
•
IC
Simple, FIFO
Head of the line (HOL) blocking
Output buffer
IC
•
•
IC
No HOL blocking
More difficult to implement
switch
fabric
•
•
OC
OC
OC
switch
fabric
OC
IC
Recirculation buffers
OC
OC
buffer
Shared by all inputs
Requires larger switch size
Revision: date
57 ( )
Optical buffering
•
Fiber delay lines (FDLs)



Not random access
Require synchronization
Supported packet format
•
•
•

Long fiber delay lines
•
•

•
Constant packet size
Some configurations support variable packet size
 A certain granularity
Not compatible with packet formats of different packet size
Not very practical solution
Ex.: For packets containing 53 bytes (ATM cell) at 2.5 Gb/s the length of fiber in the
FDLs needs to be the multiples of 640 m
Feed-forward or feed-back configurations
Novel solutions for optical memory


Material subjected to EIT (Electromagnetically induced transparency)
Optical Cavities
Revision: date
58 ( )
Optical memory
• Optical memory

a medium, a device or a system with an ability to delay (slow down) or
stop a light pulse by e.g. reduce the group velocity of light with a limited
distortion.
• Examples:


Fiber Delay Line (FDL)
Material subjected to EIT

Optical Cavities
Revision: date
59 ( )
Fiber delay line
1 km
•Cheap and easy to manufacture
• Several kilometers long
• Slow down factor - zero
• No flexibility in terms of storage time
•Requires synchronization
Many architectures proposed to introduce variable
delay
Revision: date
60 ( )
Types of buffers based on FDL
Revision: date
61 ( )
Switched Delay Line (SDL) element
SDL: A network element that is built by optical
crossbar switches and fiber delay lines.
An optical memory cell: (a) writing the packet
(b) circulating the packet (c) reading the packet
(a)
1
(b)
1
(c)
Revision: date
1
62 ( )
22 time slot interchange
Using a memory cell as a 22 time slot interchange
(a)
1
(b)
1
(c)
1
To reverse the order, set the connection in (b) to
the “bar’’ state.
To preserve the order, set the connection in (b)
to the “cross’’ state.
Revision: date
63 ( )
Novel types for optical memory: EIT
• Electromagnetically induced transparency (EIT) – an artificially
created spectral window of transparency used to slow and spatially
compress light pulses.
• Inside the memory cell, light is converted into a spin excitation of
atoms and its velocity drops down to zero once the coupling beam is
turned off. After the coupling beam is turned back on the atomic
coherence is converted back into light signal.
Revision: date
64 ( )
EIT, cont.
coupling
beam
Arriving IP packet
Memory cell
Revision: date
Optical fiber
65 ( )
EIT, cont.
Phase 1 : writing
Light slows down inside the cell and
is spatially compressed
Revision: date
66 ( )
EIT, cont.
Phase 1 : writing
Cell length
The memory cell needs to be long enough to fit the
entire packet
IP packet of 1500bytes at 2,5Gb/s is 1,4 km long in
free space and about 400m long in an optical fiber
Revision: date
67 ( )
EIT, cont.
The coupling
beam is turned
off
Phase 2 : storage
The light is stored in the material
Revision: date
68 ( )
EIT, cont.
The coupling
beam is turned
back on
Phase 3 : reading
The light is recovered
and leaves the cell
Revision: date
69 ( )
EIT, cont.
Variable coupling
power
The packet is slowed down in the cell
No storage of light
We regulate the slowdown factor by varying the coupling power
Revision: date
70 ( )
EIT, cont.
material
slow down factor
storage time
Quantum dots
40 in room temperature
107 in very low temperature
8.7ns
Atomic vapor
105
up to 0.5 ms
depends on the gas
Slow down factor and storage time depend on the
material, temperature, coupling power, bandwidth
and wavelength
Revision: date
71 ( )
Novel types for opt. memory - Optical cavities
Optical cavities use optical resonance in photonic structures
Slow down factor of 104 (depending on the number of
side cavities)
Storage time: 50 ns
Chip scale implementation of the system foreseeable
Revision: date
72 ( )
Requirement for optical memory
Telecommunications
• Wavelength
• Attenuation and
distortion
• Bandwidth
• Packet length
• Control memory cells
separately
Technology
• Compression rate (cell size)
• Tuning of the intensity of the
control field
• Temperature and mechanical stress
• Cost
QoS
• The storage time
• Pulse distortion
• Priority classes
Revision: date
73 ( )
Comparison
storage time
cell size
temperature
bandwidthwavelength
EIT
Up to 0.5 ms
Order of cm
Close to 0K or 80C
Depends on the
material
Order of ns
Size of a chip
Room temp.
No limitations
Optical cavities
Revision: date
74 ( )
Ex. 1: Broadcast-and-select OPS
P. Gambini, et al: “Transparent Optical Packet Switching: Network Architecture and Demonstrators in the
KEOPS Project,” IEEE JSAC, vol. 16, no. 7, pp. 1245-1257, Sept. 1998.
Revision: date
75 ( )
Ex. 2: Fiber loop memory based OPS
Revision: date
76 ( )
Ex. 3: Broadcast-and-select optical ATM switch
J.M. Gabriagues, et al.: ”Design, modeling and implementation of the ATMOS project fiber delay
line photonic switching matrix” Optical and Quantum Electronics, vol. 26, no.5, pp. 497-516,
May 1994.
Revision: date
77 ( )
Ex.4: OPS with recirculation buffer.
Optical
demultiplexer
Parallel electrical
and optical buffer
positions
Switching
matrix
Optical
inputs
Optical
outputs
L. Wosinska, ” Buffering and control in all-optical packet switching nodes”, (Invited paper),
Revision: date
78 ( )
in Proc. of ICTON’05, Barcelona, Spain, July 2005
Optical packet switch. Example,
cont.
one module
Electrical buffer
Buffer
1
1
N
.
.
.
.
.
.
.
.
.
.
.
.
Optical buffer
N’
N’+1
N’ + N
1
N
The storage time in an optical buffer:
• is limited
• can be composed of fix write and read
time and variable storage time
or variable write and read time
• can only take pre-defined values
Revision: date
79 ( )
Optical packet switch - simulation
Simulation assumptions:
ATM traffic: storage time 0.256μs, granularity: 2ns*
IP traffic: 0.6144μs for IP granularity 4,8ns
Priority classes:
• Transparency Class 20%
• Low Loss Class 20%
• Normal 60%
Equal load at each input and output
Exponentially distributed inter-arrival time
*The granularity has been chosen for the simplicity of the control unit to
obtain 128 (27=1 byte) values of possible delay.
Revision: date
80 ( )
Optical packet switch - simulation results
IP traffic, transparency class
packet loss probability
1,E+00
load 0.2
load 0.5
load 0.7
1,E-01
1,E-02
1,E-03
0
2
4
6
8
10
12
14
16
number of optical buffer positions
The loss probability goes down to a certain point and than
stays constant as the buffer increases.
Revision: date
81 ( )
Optical packet switch - simulation results
ATM traffic, transparency class
loss probability
1,E+00
3 buffer positions
5 buffer positions
10 buffer positions
1,E-01
1,E-02
1,E-03
1,E-04
1,E-05
1,E-06
0
0,05
0,1
0,15
0,2
0,25
maximum storage time in μs
The lowest achievable packet loss probability for a given
number of buffer positions reaches a limit that cannot be
overcome by increasing the maximum storage time.
Revision: date
82 ( )
Optical packet switch - simulation results
packet loss probability
ATM traffic, trasnparency class
1,E+00
load 0.2
1,E-01
load 0.5
1,E-02
load 0.7
1,E-03
1,E-04
1,E-05
1,E-06
0
2
4
6
8
10
12
number of optical buffer positions
Storage time of 0.5ms is enough to obtain any value of loss
probability for any traffic load.
Revision: date
83 ( )
Optical burst switching
Hybrid packet switching: a feasible solution?
Revision: date
84 ( )
Optical Burst-Switched Networks
• A compromise between circuit-switching and
packet-switching
• A control packet is first sent to set up the
“connection” for a burst, which is released as
soon as the burst is sent
• Bandwidth is reserved for a shorter time than
circuit-switching  higher bandwidth efficiency
Revision: date
85 ( )
OBS: the main idea
•
Sort data at optical network ingress
according to destination



•
OBS is a hybrid of circuit and packet
switching

•
Collect a burst of data for a destination
Send a control packet to set up a path
Send burst when path should be
established
Tends towards CS when burst is large
Main performance parameter: burst
loss
Revision: date
86 ( )
Design issues
• Burst assembly and scheduling
• When to send control packet


When burst is ready (size is known)
In anticipation of burst
• Determine the time offset between transmissions of control
packet and burst


Should account for processing and setup delays
Need to consider number of hops
• Resource scheduling for burst switches
Revision: date
87 ( )
OBS network
B. Mukherjee, ”Optical WDM Networks”, Springer
Revision: date
88 ( )
OCS, OBS, OPS: Comparison
Property
Wavelength
routing
OCS
Optical Burst
Switching
OBS
Optical Packet
Switching
OPS
Granularity
Large
Middle
Small
Hardware limitations
Low
Low
High
No
No
Yes
Wavelength converter
Yes/No
No
Yes
Electronic bottleneck
Yes/No
No
Yes
Control overhead
Low
Low
High
Scalability
Low
High
High
Flexibility
Low
High
High
Cost
Low
Low
High
Optical buffer
Revision: date
89 ( )
Summary
• Switched networks
• Photonic circuit switching (Solving LTD and RWA problems)
• Photonic packet switching

Makes networks more efficient
•

Lowers blocking in RWA
Many technical challenges for switch design
•
Buffering for contention resolution
•
Scheduling for contention resolution with possible deflection
•
Switching speeds
• Optical burst switching

Slower switching speeds than packet switching

Allows time division of resources

Promising technique with possibility of realization
Revision: date
90 ( )