Professor Z Ghassemlooy
Northumbria Communications Laboratory
School of Computing, Engineering and
Information Sciences
The University of Northumbria
U.K.
http://soe.unn.ac.uk/ocr
Prof. Z Ghassemlooy 1
Network Systems
Network Trends
Switch Fabric
Type of Switches
Optical Cross Connects
Optical Cross Connects Architecture
Large Scale Switches
Optical Router
Applications
Prof. Z Ghassemlooy 2
2004 International Engineering Consortium
Prof. Z Ghassemlooy 3
Network Connectivity
– Point to Point: one to one
– Broadcast: one to many
– Multicast: many to many
Network Span
– Local / Metro Area Network
– Wide Area Network
– Long Haul Network
Data Rates
– Voice 64kbps
– Video 155Mbps, etc.
Service Types
– Constant or Variable bit rate
– Messaging
– Quality of Service
Prof. Z Ghassemlooy 4
Ports Ports
Problem
- limited and could not scale to thousands or millions of users
Solution
switched network
Prof. Z Ghassemlooy 5
Pervasive, high-bandwidth, reliable, transparent
Prof. Z Ghassemlooy 6
Capacity
2.5 Gb/s 10 Gb/s 40 Gb/s Larger
Control (switching)
– Electronics
• 10 Gb/s (GaAs, InP) can deliver low order optical cross connects (16 x 16)
• > 10 Gb/s ??(mainly power dissipation)
– Optical
Reconfiguration:
– Static or dynamic
Prof. Z Ghassemlooy 7
Dense Wavelength Division Multiplexing
Optical Add/Drop Multiplexers (OADM)
Optical Gateways:
– A critical network element.
– A common transport structure to cater for
• variety of bit rates and signal formats, ranging from asynchronous legacy networks to 10 –Gbps SONET systems,
• a mix of standard SONET and ATM services.
Prof. Z Ghassemlooy 8
Right now, the optical switches have electrical core, where
– Light pulses are converted back into electrical signals so that their route across the middle of the switch can be handled by conventional ASICs (application specific integrated circuits).
This has a number of advantages:
• Enabling the switches to handle smaller bandwidths than whole wavelengths, which fits in with current market requirements.
• Easier network management, because standards are in place and products are available. Optical equivalents are not, at present.
But, there are concerns that electrical cores won’t be able to cope with the explosion in the number of wavelengths in telecom networks (deployment of DWDM).
Until recently, state-of-theart ASIC technology wouldn’t support anything more than a 512-by-512-port electrical core, and carriers demanding for at least double this capacity.
Prof. Z Ghassemlooy 9
Optical Bidirectional Line
Switched Rings
Optical Cross-Connect
(OXC)
– Efficient use of existing optical fibre facilities at the optical level becomes critical as service providers started moving wavelengths around the glob. Routing and grooming are key areas, and that is where OXCs are used .
International Engineering Consortium, 2004
10 Prof. Z Ghassemlooy
• To provide high switching speed
• To avoid the electronics speed bottleneck
• I/O interface and switching fabric in optics
• Switching control and switching fabric in optics
• Switches act as router s and redirect the optical signals in a specific direction.
• It uses a simple 2x2 switch as a building block
Main feature: Switching time (msecs - to- sub nsecs)
Prof. Z Ghassemlooy 11
That’s the theory. But, things are turning out a little different in practice.
– Vendors are finding ways of building larger scale electrical cores, with switch of many thousands of ports.
– This may encourage carriers to put off decisions on moving to all-optical switches.
Does this mean that is the end of the idea of alloptical networks?
– Well, not really. All that it might do is delay things.
Prof. Z Ghassemlooy 12
M C Wu
1024
512
256
128
64
32
16
8
• High power
•
• Large switches
• Need OE/EO conversion
•
Electrical Limits consumption: e.g. 1024x1024: 4 kW
Jitter: very large
Bipolar or GaAs
10 MHz 100 MHz
Electrical
1 GHz
DS3 OC3
10 GHz
OC12 OC48 OC192
Prof. Z Ghassemlooy
Optical
100 GHz
Data rate
13
Circuit Switching: E.g. Telephone
– Continuous streams
• no bursts
• no buffers
– Connections are created and removed
- Buffering does not exist in circuit-switches
Packet Switching: Uses store & forward
- The configuration may change per packet
- Switching/forwarding is based on the destination address mapping
- Switching table is used to provide the mapping
- Switching table changes according to network dynamics (e.g. congestion, failure)
Prof. Z Ghassemlooy 14
Electro-optical 2 x 2 switching elements are the key devices in the fabrication of N x N optical data path.
The switching elements rely on the electro-optic effect (i.e., the application of an electric field to an electro-optical material changes the refractive index of the material).
The result is a 2x2 optical switching element whose state is determined by an electrical control signal.
Can be fabricated using LiNbO3 as well as other materials.
Electrical control Electrical control
Optical input
Optical output
Optical input
Prof. Z Ghassemlooy
Optical output
15
contd.
Input interface
Output interface
Switching fabric
Switching control
Prof. Z Ghassemlooy 16
contd.
...
...
1.3 m m intra-office
Optical
Crossconnect
(OXC)
Optical transport system
(1.55 m m WDM)
Terminating equipment
|
SONET, ATM, IP...
Prof. Z Ghassemlooy
Transponders
17
Since a switch work as a permutation that routes input to the outputs, therefore it needs to provide at least N ! different configuration
A minimum number of Log
2
( N !) is needed to configure N ! different permutation
Blocking
Non-Blocking
Prof. Z Ghassemlooy 18
Occurs when one reduces the number of crosspoints in order to achieve low crosstalk and less complexity.
In some switching architecture internal blocking may be reduced to zero by:
– Improving the switching control: Wide sense nonblocking
– Rearranging the switching configuration:
Rearrangeably non-blocking
Prof. Z Ghassemlooy 19
A new connection can always be made without disturbing the existing connections:
Strictly Non-blocking
– A connection path can always be found no matter what the current switching configuration is or what switching control algorithm is used
Wide-Sense Non-blocking
– A connection path can always be found regardless of the current switching configuration provided a good switching control algorithm is employed
– No re-routing of the existing connections
Rearrangeably Non-blocking
– The same as wide-sense, but requires re-routing of the existing connections to avoid blocking
– Use fewer switches
– Requires more complex control algorithm
Prof. Z Ghassemlooy 20
Interchanges sample (slot) position within a frame: i.e. time slot interchange (TSI)
– when demultiplexing, position in frame determines output link
– read and write to shared memory in different order
1
1
N
M
U
X
4 3 2 1 TSI
1
2
3
4
2 4 1 3
D
E
M
U
X N
Prof. Z Ghassemlooy 21
Simple
Time taken to read and write to memory is the bottle-neck
For 120,000 telephone circuits
– each circuit reads and writes memory once every 125 ms.
– number of operations per second : 120,000 x 8000 x2
– each operation takes around 0.5 ns => impossible with current technology
Prof. Z Ghassemlooy 22
Crossbar
Clos
Benes
Spank - Benes
Spanke
Prof. Z Ghassemlooy 23
Each sample takes a different path through the switch, depending on its destination
Crossbar:
– Simplest possible space-division switch
– Wide- sense blocking:
When a connection is made it can exclude the possibility of certain other connections being made
Crosspoints
– can be turned on or off
Input ports 3
4
1
2
Sessions: (1,4) (2,2) (3,1) (4,3) 1 2 3 4
Output ports
Prof. Z Ghassemlooy 24
Input channels
1
2
M inputs x N outputs
Switch configuration: “set of input-output pairs simultaneously connected” that define the state of the switch
For X crosspoints, each point is either ON or Off , so at most 2 X different configurations are supported by the switch.
3
N X N matrix S/W
4
Case 1:
- (3,2) ok
- (4,3) blocked
Optical switching element
1 2 3
Output channels - Cross
4
Prof. Z Ghassemlooy 25
Rule : To connect i th input to
Input channels the j th output, the algorithm
1 sets the
2
3 switch in the i th row and j th column at the “ BAR ” state and sets all other switches on its left and below at the “ CROSS ” state.
4 Case 2:
- (2,4) ok
- (3,2) ok
- (4,3) ok
1 2 3 4
Output channels
Prof. Z Ghassemlooy 26
Only uses 6 x 9 = 54 cross points rather than 9 x 9 = 81
Penalty is loss of connectivity
2
3x3
5
Prof. Z Ghassemlooy 27
1
2
3
1
2
3
4
5
6
7
8
9
7
8
9
4
5
6
Blocking still possible
Prof. Z Ghassemlooy http://www.aston.ac.uk/~blowkj/index.htm
28
*
1
2
3
7
8
9
4
5
6
1
2
3
Blocking
8
9
7
4
5
6
*
The first four connections have made it impossible for
3 rd input to be connected to 7 th output
The 3 rd input can only reach the bottom middle switch
The 7 th output line can only be reached from the top output switch.
Prof. Z Ghassemlooy 29
Architecture:
Switch element:
Switch drive:
Switch loss:
SNR:
Wide Sense Non-blocking
N 2 (based on 2 x 2)
N 2
(2 N -1).
L se
+2 L fs
XT – 10log
10
( N -1)
Where XT ; Crosstalk (dB),
L se
; Loss/switch element
L fs
; Fibre-switch loss
Prof. Z Ghassemlooy 30
Advantages:
– simple to implement
– simple control
– strict sense non-blocking
– Low crosstalk: Waveguides do not cross each other
Disadvantages
– number of crosspoints = N 2
– large VLSI space
– vulnerable to single faults
– the overall insertion loss is different for each inputoutput pair: Each path goes through a different number of switches
Prof. Z Ghassemlooy 31
1
M
U
X
2 1 TSI 2 1 time 1 time 1
2
3
M
U
X
4 3 TSI 3 4
4
3
1
2
4
Each input trunk in a crossbar is preceded with a TSI
Delay samples so that they arrive at the right time for the space division switch’s schedule
Note: No. of Crosspoints N = 4 (not 16)
Prof. Z Ghassemlooy 32
Can flip samples both on input and output trunk
Gives more flexibility => lowers call blocking probability
TSI
TSI
TSI
TSI
TSI TSI TSI TSI
Complex in terms of:
- Number of cross points
- Size of buffers
-Speed of the switch bus (internal speed)
Prof. Z Ghassemlooy 33
1 n
32
n x p
1 k x k
1
33
64
2
32
993 k
N = 1024
Stage 1
2 p
64
Stage 2 p x
1
2 k n
32
Stage 3 n
•It is a 3-stage network
1st & 2nd stages are fully connected
- 2nd & 3rd stages are fully connected
- 1st & 3rd stages are not directly connected
Defined by: ( n , k , p , k , n )
e.g. (32, 3, 3, 3, 32)
(3, 3, 5, 2, 2,)
• Widely used
• Stage 1 ( n x p )
• Stage 2
( k x k )
• Stage 3 ( p x n )
Prof. Z Ghassemlooy 34
In this 3-stage configuration N x N switch has:
2 pN + pk 2 crosspoints (note N = nk )
(compared to N 2 for a 1-stage crossbar)
If n = k , then the total number of crosspoints =
3 pN , which is < N 2 if 3 p < N .
Problem:
Internal blocking
Larger number of crossovers when p is large.
Prof. Z Ghassemlooy 35
If p < 2 n -1, blocking can occur as follows:
- Suppose input 1 want to connect to output 1 (these could be any fixed input and outputs.
- There are n -1 other inputs at k -switch (stage 1). Suppose they each go to a different switch at stage 2.
- Similarly, suppose the n -1 outputs in the first switch other than output 1 at the third stage are all busy again using n -
1 different switches at stage 2.
- If p < n -1 + n -1 +1 = 2 n -1 then there will be no line that input 1 can use to connect to output 1.
If p = 2 n -1, then
– Total Switch Element: 2 kn (2 n -1) + (2 n -1) k 2
Prof. Z Ghassemlooy 36
If p = 2 n -1, then
– Total Switch Element: 2 kn (2 n -1) + (2 n -1) k 2
Since k = N / n , therefore
– the number of switch elements is minimised when n ~( N /2) 0.5
.
Thus the number switch elements =
4 (2) 0.5
N 3/2 – 4 N , which is less than N 2 for the crossbar switch
Prof. Z Ghassemlooy 37
If p
2 n -1, the Clos network is strict sense nonblocking (i.e. there will free line that can be used to connect input 1 to output 1)
If p
n, then the Clos network is re-arrangeably non-blocking (RNB) (i.e. reducing the number of middle stage switches)
Prof. Z Ghassemlooy 38
If N = 1000 and and n = 10, then the number of switches at the:
– 1 st & 3 rd stages = N/n = 1000/10 = 100
– 1 st stage = 10 x p
– 3 rd stage = p x 10
– 2 nd stage = p x k x k.
If p = 2 n -1 = 19, then the resulting switch will be non-blocking.
If p < 19, then blocking occurs.
For p = 19, the number of crosspoints are given as follow:-
Prof. Z Ghassemlooy 39
contd.
In the case of a full 1000 x 1000 crossbar switch, no blocking occurs, requiring 10 6 crosspoints.
For n = 10 and p = 19, the number of crosspoints at
– 1 st and 3 rd stages
= no. of stages x ( n x p ) x k
= 2 x (10 x 19) x 100 = 38,000 crosspoints
– 2 nd stage ( p = 19 crossbars each of size 100 x 100, because N/n =
100.
= p x k x k = 19 x 100 x 100 = 190000 crosspoints.
The total no. of crosspoints = 38000 + 190000 = 228000
Vs. the 10 6 points used by the complete crossbar.
Prof. Z Ghassemlooy 40
contd.
Since k = N / n , the number of switch elements k is minimised when n
~( N /2) 0.5
= (1000/2) 0.5
=~ 23 instead of 19.
then k = N / n = 1000/23 =~ 44 switches in the 1
2(23) -1 = 45. st & 3 rd stages, and p = the number of crosspoints at 1 st and 3 rd stages
= no. of stages x ( n x p ) x k
= 2 x (23 x 45) x 44 = 91080.
the number of crosspoints at 2 nd stage = p x k x k = 45 x 44 x 44 = 87120.
Since n = 23 does not divide 1000 evenly, we actually have 12 extra inputs and outputs that we could switch with this configuration ( 23x44=1012 and 1012 - 1000 = 12).
Thus the total number of crosspoints = 91090 + 87120 = 178200 best case for a non-blocking switch as compared with the:
1,000,000 for the complete crossbar and about 190,000 for n = 10.
This is a factor of over 11 less equipment needed to switch 1000 customers!
Prof. Z Ghassemlooy 41
2
2
N /2
N /2
Benes
N
N /2
N /2
Benes
2
2
N
N x N switch ( N is power of 2) RNB built recursively from
Clos network:
1st step Clos(2, N /2, 2, N /2, 2)
Rearrangably non-blocking
Prof. Z Ghassemlooy 42
contd.
Number of stages = 2.log
2
N - 1
Number of 2x2 switches /each stage = N/2
Total number of crosspoints ~ N.(log
2
N -1)/2
For large N , total number of crosspoint = N.log
2
N
Benes network is RNB (not SNB) and so may need re-routing:
Modular switch design
Multicast switches can be built in a modular fashion by including a copy module in front of the point-to-point switch
Prof. Z Ghassemlooy 43
contd.
1
2
3
4
5
6
7
8
•e.g. Connection sequence
2 to 1 1 to 5 3 to 3
Note there is no way 4 to 2 connection could be made
Prof. Z Ghassemlooy
X
4 to 2 Fails
44
3
4
5
1
2
6
7
8
contd.
• Now use different connections
• e.g.
2 to 1 1 to 5 3 to 3
Prof. Z Ghassemlooy
4 to 2 OK
45
International Engineering Consortium, 2004
Prof. Z Ghassemlooy 46
Control Signal
Input port I i
Optical Switch
I
1
I
2
Output ports
The input signal can be switched to either of the output ports without having any access to the information carried by the input optical signal
• In the ideal case, the switching must be fast and low-loss.
• 100% of the light should be passed to one port and 0% to the other port.
Prof. Z Ghassemlooy 47
contd.
The two-position switch requires three fibres with collimating lenses and a prism.
B
A
C
Fibre
Lens
Prisem
Light arriving at port A needs to be switched to port C.
B
A
C
Prof. Z Ghassemlooy 48
Provisioning: Used inside optical cross connects to reconfigure them and set-up new path. [1 - 10 msecs]
Protection Switching: To switch traffic from a primary fibre onto another fibre in the case of a failure. [1 to 10 usecs]
Packet Switching: 53 byte packet @ 10 Gb/s. [1 nsecs]
External Modulation: To switch on-off a laser source at a very high speed. [10 psecs << bit duration]
Network performance monitoring
Reconfiguration and restoration: Fibre networks
Prof. Z Ghassemlooy 49
Slow Switches (msecs)
– Free space
– Mechanical
– Solid state
Fast Switches (nsecs)
– LiNbO
– Non-linear
– InP
Prof. Z Ghassemlooy 50
Maximum Throughput:
– Total number of bits/sec switched through.
– To increase throughput:
• Increase the number of I/O ports
• Bit rate of each line
Maximum Switching Speed
– Important:
• Packet switched
• Time division multiplexed
Minimum Number of Crosspoints
– As the size of the switch increases, so does the number of crosspoints, thus high cost
– Multistage switching architecture are used to reduce the number of crosspoints.
Prof. Z Ghassemlooy 51
contd.
Minimum Blocking Probability: Important in circuit switching
– External blocking: when the incoming call request an output port that is blocked.
• Subject to external traffic conditions
– Internal blocking: when no input port is available.
• Subject to the switch design
Minimum Delay and Loss Probability
– Important in packet switching, where buffering is used, which will introduce additional delay.
Scalability
– Replacing an old switch with a new larger switch is costly.
– Incrementally increasing the size of the existing switching as traffice grows is desirable
Broadcasting and Multicasting
– To provide conferencing and multimedia applications
Prof. Z Ghassemlooy 52
contd.
• Optical switches with low insertion loss and low crosstalk are needed in broadband optical networks
– Restoration
– Reprovisioning
– Bandwidth on demand
• Conventional optical switches cannot satisfy all the requirements:
– Solid-state guided-wave switches (electro-optic, thermo-optic,
SOA): limited expandability due to high crosstalk, loss, and power consumption
– Optomechanical switches: excellent insertion loss and crosstalk , but are bulky, expensive, and suffer from poor reliability and scalability
Prof. Z Ghassemlooy 53
Waveguide
Electro-optic effect
- Semiconductor optical amplifier
- LiNbO
- InP
Thermo-optic effect
- SiO
2
/ Si
- Polymer
- Fast
- Complex
- Maturing
- Lossy
- Slow
- Maturity
- Reliable
Free Space
- Liquid crystal
- Mechanical / fibre
- Microoptics (MEM’s)
- Slow
- Low loss & crosstalk
- Inherently scalable
Prof. Z Ghassemlooy 54
Some materials have strong thermo-optics effect that could be used to guide light in a waveguide.
The thermo-optic coefficient is:
–
Silica glass dn/dt = 1 x 10 -5 K -1
– Polymer dn/dt = -1 x 10 -5 K -1
Difference thermo-optic effect results in different switch design.
+ v
Electrodes
Prof. Z Ghassemlooy 55
Mach – Zehnder Configuration
Input I i
Heater
I
1
I i
sin
2
(
/ 2 )
Directional coupler
Prof. Z Ghassemlooy
I
2
I i
cos
2
(
/ 2 )
Outputs
I
1
I
2
56
Y – Junction Configuration
I
1 P
H1
I i
P
H2
I
2
• If P
H1
• If P
H1
• If P
H1
= P
H2
= 0, then I
1
= I
2
= I i
= P on
& P
= 0 & P
H2
H2
= 0, then I
= P on
, then I
1
1
/2
= 0, and I
2
= I i
, and I
2
= I i
= 0
Prof. Z Ghassemlooy 57
Parameters
No. of S/W
Insertion Loss (dB)
Crosstalk
S/W time (ms)
S/W power (W)
2 x 2
Si Poly.
1 1
2 0.6
22 39
2 1
0.6
0.005
Switch Size
8 x 8
Si Poly.
64 112
4 10
18 17
~3 1.5
5 4.5
16 x 16
Si
256
18
13
~4
15
Prof. Z Ghassemlooy 58
1 st Generation – Mid. 1980’s
Loss
Speed
Size
Low (0.2 – 0.3 dB) slow (msecs)
Large
Reliability
Applications:
Has moving part
- Instrumentation
- Telecom (a few)
Size:
Loss:
8 X 8
3 dB
Crosstalk: 55 dB
Switching time: 10 msecs
Prof. Z Ghassemlooy 59
Combines optomechanical structures, microactuators, and micro-optical elements on the same substrate
Made using micro-machining
Free-space: polarisation independent
Independent of:
– Bit-rate
– Wavelength
– Protocol
Speed: 1 10 ms
4 x 4 Cross point switch
Lens
Output fibres
Flat mirror Raised mirror
Prof. Z Ghassemlooy 60
Lightwave
This tiny electronically tiltable mirror is a building block in devices such as all-optical cross-connects and new types of computer data projectors.
I/O Fibers
Reflector
MEMS 2-axis
Tilt Mirrors
Imaging
Lenses
Prof. Z Ghassemlooy 61
Monolithic integration --> Compact, lightweight, scalable
Batch fabrication --> Low cost
Share the advantages of optomechanical switches without their adverse effects
General Characteristics:
+ Low insertion loss (~ 1 dB)
+ Small crosstalk (< - 60 dB)
+ Passive optical switch (independent of wavelength, bit rate, modulation format)
+ No standby power
+ Rugged
+ Scalable to large-scale optical crossconnect switches
– Moderate speed ( switch time from 100 nsec to 10 msec)
Prof. Z Ghassemlooy 62
Switch sizes > 2 X 2 can be implemented by means of cascading small switches.
Used in all network control
Bit rate at which it functions depends on the applications.
– 2.5 Gb/s are currently available
Different sizes are available, but not up to thousands (at the moment)
Control
1
2
1
2
N N
N X N Cross Connect
Prof. Z Ghassemlooy 63
Prof. Z Ghassemlooy 64
Optical switching and optical cabling, clocking and synchronization are not significant issues because the streams are independent.
Inputs come from different clock domains, so the switch is completely timing-transparent.
Prof. Z Ghassemlooy
Electrical switching and optical cabling: inputs come from different clock domains resulting in a switch that is generally timing-transparent.
65
For a given switch size N ,
– the number of 2x2 switches should be as small as possible. When the number is large it will result in:
• high cost
• large optical power loss and crosstalk.
A switch with reduced number of crosspoints in each configured path, can have a large internal blocking probability
In some switching architectures, the internal blocking probability can be reduced to zero by:
– using a good switching control
– or rearranging the current switch configuration
Prof. Z Ghassemlooy 66
In the core large optical-switching elements have already started to appear to handle optical circuits,
Large, centralized IP routers are also appearing, because they're an extremely efficient solution to IP routing.
There are a variety of technologies and issues that influence the architecture for these types of network elements.
To transport Tbps, new optical technologies have emerged to enable the economic transport of incredible bandwidth over single-mode optical fibrer, including DWDM and
OTDM. That means individual optical links can sustain the enormous traffic needed to support the continuing growth of IP data.
Prof. Z Ghassemlooy 67
High-power, low-noise optical amplifiers-or erbium-doped fiber amplifiers (EDFAs)-and pulseshaping technologies mean the high-bit-rate optical signals do not require electronic regeneration except on the very longest fiber spans.
New fibres with larger cross-sectional areas mean a large number of high-bit-rate signals can be wavelength-multiplexed onto a single fiber.
Thus, it is becoming affordable to actually construct links that can support Tbps of capacity between routing and switching centres.
Prof. Z Ghassemlooy 68
The bottleneck at the core of the expanding network is at the junction points of the fibre bundles: I.e the switching and routing centres. With Tbps links, a huge amount of data converges into a single central office (CO) (see Figure 1).
New routers emerge only to be swamped with traffic within months.
Prof. Z Ghassemlooy 69
Solution:
Use of cluster of several routers (or crossconnects).
However, clustering is not a good long-term solution, because:
• a cluster of crossconnects requires interconnecting links between the crossconnects. As the number of switches in the cluster grows beyond about 4 or 5, the interconnecting links consume most of the ports. Clustered routers have the same problem.
• the IP traffic must transit more and more devices, and the latency (the delay of IP packets) and jitter (delay variance) of the cluster grow quickly.
• the hot-spot problem, where one of the small routers in a cluster can be overwhelmed by temporary traffic dynamics in the network that do not exceed the combined node capacity.
This swamping effect also increases the delay of that saturated small router .
Prof. Z Ghassemlooy 70
Current trend in XCs is to use large microelectromechanical systems (MEMS)-based OXCs for core node protection and grooming of DWDM traffic.
Similarly, large centralized routers are an efficient alternative to solving bottleneck problems:
– by avoiding the hot-spot problems of distributed routers,
– eliminating clustering problems, and
– permitting global scheduling.
A centralized (single-hop), synchronous, large nonblocking switch fabric has the best latency and throughput performance of all router topologies. It also scales better than a clustered system-and it results in less complicated system software for the network element.
Prof. Z Ghassemlooy 71
Router
Router
Router
Router
ONE
ONE
Optical Network
ONE
Router
End Customer
A V Lehmen, Telecordia Tech.
Prof. Z Ghassemlooy 72
IP
Router
IP
Router
IP
Router
IP
Router
OXC - A OXC - C
OXC - B
IP
Router
OXC - D
Crossconnects are reconfigurable:
Can provide restoration capability
Provide connectivity between any two routers
A V Lehmen, Telecordia Tech.
Prof. Z Ghassemlooy 73
Access lines
A
Z
• All traffic flows through routers
• Optics just transports the data from one point to another
• IP layer can handle restoration
• Network is ‘simple’
Access lines
A V Lehmen, Telecordia Tech.
• But…..
- more hops translates into more packet delays
- each router has to deal with thru traffic as well as terminating traffic
Prof. Z Ghassemlooy 74
OXC
OXC OXC
OXC
• Router interconnectivity through OXC’s
• Only terminating traffic goes through routers
• Thru traffic carried on optical ‘bypass’
• Restoration can be done at the optical layer
• Network can handle other types of traffic as well
•But: network has more NE’s, and is more complicated
Prof. Z Ghassemlooy
A V Lehmen, Telecordia Tech.
75
IP
Router
IP
Router
IP
Router
IP
Router
OXC - A OXC - C
OXC - B
A V Lehmen, Telecordia Tech.
1. Router requests a new optical connection
2. OXC A makes admission and routing decisions
3. Path set-up message propagates through network
4. Connection is established and routers are notified
Prof. Z Ghassemlooy 76
1 1
N N
1 1
N
N
• Type I 1 x N & N x 1 optical switches
• Type II 1 x N passive optical splitter
N x 1 Optical switches
Prof. Z Ghassemlooy 77
Architecture
Switch Element
Switch Drive
Switch Loss
SNR
Type I
Strictly non-blocking
TypeII
2N(N-1) N(N-1)
N log
2
N 2 N log
2
N
( 2 N log
2
N ) L se
+ 4 L fs
2XT-10log
10
(log
2
N ) log
2
N ( 3+L
XT-10log
10 se
) +
(log
2
2
L fs
N )
Where XT ; Crosstalk (dB),
L se
; Loss/switch element
L fs
; Fibre-switch loss
Prof. Z Ghassemlooy 78
Prof. Z Ghassemlooy 79
Characteristic
Switching Speed
Multicast
Insertion loss
Cross talk
Scalability
Traditional Optical
Switches
>1ms
Not available
Next Generation
Optical Switches
<1µsec
Dynamic power partition between ports
High dynamic range VOA Integrated VOA functionality
Reliability
Not available
~10 Million cycles (Mech.dev.) ~10 Billion cycles (Optoelect.)
Low Low
High
Low
Low
Medium-High
Prof. Z Ghassemlooy 80
Performs digital grooming, traditional multiplexing, and routing of lowerspeed circuits in mesh or ring network configurations. Specifically, it brings in lower rate SONET/SDH layer OC-3/STM-1, OC-12/STM-4 and OC-
48/STM-16 rates and electrical DS-3, STS-1 and STM-1e rates and grooms them into higher rate optical signals.
Alcatel. 2001
Prof. Z Ghassemlooy 81
IP-router with Tb/s throughput can be built with fast tunable lasers & NxN optical mux
From Input Port
Buffer
Output
40G Rx
40G Rx
40G Rx retiming
Scheduler
T-Tx
T-Tx
T-Tx
T-Tx
40G Rx
Yamada et al., 1998
Clock
Prof. Z Ghassemlooy 82
CHIARO- OptIPuter Optical Switch Workshop
Prof. Z Ghassemlooy 83
Services are consolidated onto a single access line at the user site and fed into a Sonet multi-service provisioning platform at the carrier’s
POP (point of presence). Several
POPs feed traffic into a terabit switch capable of handling all traffic— including IP, ATM and TDM. The terabit switches sit at the edge of a three-tier network of optical switches—local, regional and long distance-each of which has a mesh topology. DWDM is used throughout the network and access lines. Where fiber is scarce, FDM (frequency division multiplexing) is used to pack as much traffic as possible into wavelengths. Light signals no longer need regeneration on long distance routes.
Prof. Z Ghassemlooy 84
Separate access networks carry telephony and data into the carrier’s point of presence. Voice traffic runs over a TDM (time division multiplexer) network running over a Sonet (synchronous optical network) backbone. IP traffic is shunted onto an ATM backbone running over other Sonet channels. The Sonet backbone comprises three tiers of rings at the local, regional and national level, interconnected by add-drop multiplexers and cross-connects.
DWDM (dense wave division multiplexing) is in use in the regional and national rings, but not the local rings. Light signals need regenerating on long distance routes.
Prof. Z Ghassemlooy 85