Transcript Wavelength Flexibility in the networks

A Presentation on
Design and Implementation of
Wavelength-Flexible Network
Nodes
Carl Nuzman, Juerg Leuthold, Roland Ryf, S.Chandrasekar, c. Randy
Giles and David T. Neilson
By
Sudharshan Reddy .B
1
Contents
•
•
•
•
What is this presentation about ?
Node Architectures
Wavelength flexibility in the networks
Analytic Estimate of Converter
Placement
• A brief discussion on Implementation
details
• Conclusion
2
What is this presentation about?
• Analytically and Experimentally examination
of node architectures for wavelength routing
networks
• Wavelength flexibility simplifies network
management and increases network capacity
• In a sharable pool, with fixed number of
wavelength channels per fiber, the number of
WC’s required remains low as the overall
capacity is scaled up.
3
What is this presentation about?
• Wavelength- routing networks provide a flexible
optical network layer where light paths can be
dynamically provisioned.
• To what extent wavelength conversion be available
at the network nodes, and how might wavelength
conversion be implemented.
• More insight into the size of the optical cross
connects (OXC’s) needed to implement nodes of
different designs in a given network.
• Discussion on cross-connect and wavelength
conversion technologies that could be used at
wavelength flexible network nodes.
4
Node Architectures
• Most existing wavelength routing networks use digital
cross-connect switches.
• A node is made opaque in the sense that the optical
signals on every link are insulated and isolated from the
signals on other links by electronic equipment.
• Converters can be classified as fixed or tunable output
wavelength respectively.
• Wavelength converters can be classified according to the
level of generation they provide i.e. WCs based on
optical-electronic translation typically provide 3R
regeneration (re-amplification, reshaping, retiming), while
typical all optical converters provide 2R regeneration
5
(reamplification and regeneration)
6
Node Architectures
• There are many tradeoffs between different
designs of the nodes.
• The simplicity of the node designs results in
number of networking challenges like increased
complexity of routing and wavelength
assignment , increased sophistication of physical
layer engineering and performance monitoring.
• The regenerators have to be deployed on the
node output ports to extend the physical reach of
7
the signals.
Node Architectures
A
B
C
D
E
Degree of wavelength
conversion
Full
(DCS)
Full
(OXC)
Shared
Partial
None
Wavelength Blocking
NO
No
No
No
Yes
# of cross-connects
1
1
1
1
W
Add/drop cross connects
required
No
No
No
No
Yes
Routing and wavelength
assignment
Simple
Simple
Complex
Complex
comple
x
Physical layer network
engineering
Node-tonode
Node-tonode
End-toend
End-toend
End-toend
Blocking fairness w.r.to
hop length
good
good
good
good
poor
* Assuming sufficient WCs provisioned
8
Acronyms
DCS – Digital cross connect
OXC – Optical cross connect
TWC - Tunable wavelength converter
FWC –Fixed – output wavelength converter
W – Number of wavelengths per fiber
F – number of fibers
P {p=P/W} – Arrival rate through demands
A {a=A/W} – Arrival rate of local add
demands
{ Fractional Rate}
9
Node Architectures
• A network built without any wavelength converters are
best for localized demand patterns , because the
wavelength continuity affects long demands (in hop
count) much more severely than in short ones.
• Limited conversion designs use single large OXC with
very few converters than in full-conversion case, but
requires sophisticated network management.
• In another architecture, electronic wavelength
conversion is performed at local access station in such a
way that transmitters and receivers are shared by adddrop traffic and traffic requiring conversion.
10
Wavelength Flexibility in the
networks
• Wavelength Blocking --- Important parameters affecting
blocking is the number of hops covered by a typical light
path and blocking is nil in single hop and likewise little in
short lightpaths.
• Although the hop count is larger in ring networks,
wavelength blocking is less under probabilistic model,
because there are strong correlations between the
wavelength occupancies on adjacent links.
• Wavelength blocking is significant in networks with long
lightpaths and low interference lengths, such as torus
networks.
• If static demands are to be routed with off-line computation,
wavelength blocking is typically reduced.
*** No. of links shared by an interfering demand averaged over all interfering demands
11
Limited wavelength conversion
• How Much wavelength conversion is sufficient ?
Although the details vary with the topology and traffic model, in
general, the answer tends to be that the level of wavelength
conversion required is small relative to the full conversion.
In worst case ring analysis --without WCs –Require 2W
with full WCs --- Require W
If equipped with simple, Fixed near neighbor wavelength conversion at a
simple node -- Require W +1
The number of WCs required to eliminate the wavelength blocking
depends on the routing and wavelength assignment algorithm used.
12
Analytical Estimate of Converter
Requirements
• The number of WCs needed in the network
depends on the wavelength assignment
algorithm used and trellis-based method is the
best.
• Random Local wavelength assignment.
• Though its simple, the analysis identifies a
number of qualitative factors affecting limited
share conversion and gives an upper bound on
the number of converters needed by other
methods.
13
Analytical Estimate of Converter
Requirements
• An analogous algorithm was analyzed in the
context of synchronous optical packet switching,
using a large deviations approach.
• Developed some simple fluid model
approximations to determine how many WCs are
needed at a given node using random
wavelength assignment.
• The results overestimate the number of
converters needed as compared to the other
methods.
14
Analytical Estimate of Converter
Requirements
• The upper bound will be loosest for very
sparsely networks, such as rings, because the
algorithm doesn’t take full advantage of high
interference lengths.
• Traffic demands arrive at times specified by a
homogeneous Poisson process and each
demand has a fixed (link and node) route.
• If the demand cannot be give a wavelength
assignment then it is blocked and disappears.
15
Analytical Estimate of Converter
Requirements
• The number of converters actually provisioned
can be chosen to keep the probability that all
converters are occupied below the given
blocking threshold.
• The number of new demands that arrive during
the average holding time in particular plays an
important role.
• The dynamic model broadly tries to capture the
variability arising from all the effects, without
being tied to a particular time scale.
16
Single input and output Fiber
•
•
•
•
•
P= rate of demands passing through the node.
A = Total number of demands being added.
Let mean holding time is 1 time unit.
X = active through light paths using converters.
Z = active light paths that are added locally.
17
Single Input fiber, Multiple
Output Fibers, Single output Link
• The need for wavelength conversion can be
greatly reduced.
• A channel is chosen randomly among the
available wavelengths when connections are
locally added and connections must be
converted.
18
Multiple Input Fibers, Single Output Link.
The through traffic from other fibers are
randomly distributed on the output fiber in
the same way as the add traffic,
regardless of whether or not this through
traffic uses conversion.
Similar description is done for multiple input
and multiple output links.
19
Maximum Number of Converters
needed
20
The analysis presented previously allows to determine
design parameters for an optical node with limited
wavelength conversion under random local wavelength
assignment.
21
• For full conversion (OXC-based) shared conversion, and partial
conversion , the number of ports required grows roughly linearly
with the total load.
• Discrete jumps occur at points where a new fiber must be added
to one o f the links surrounding the node.
22
For NODE 2
23
Implementation
• The digital cross-connect switches and
optical electrical optical conversion forms
the basis of the full conversion design.
• The principle challenges for the nodes are
limiting the cost and power consumption of
the node as the bit rates and aggregate
capacities in the network increase.
24
Implementation
• Mesh nodes with single fiber links and tens of
wavelengths per fiber require cross-connects
with 50-200 ports.
• Optical switches based on MEMS beam-steering
technology appear to be the most viable
solution.
• One of the primary relationships in the design of
beam-steering cross-connects is that between
the number of ports and the physical size of the
switch.
25
Considerations
1. The beam spots must be physically separated
on the micromirror array.
2. To maximize the port count, “D/s” should be
made as large as technologically feasible.
3. The micromirror diameter "d" should be chosen
at least 1.5 times larger than the spot size “D”,
in order to minimize clipping losses on the
mirrors and protect against small alignment
errors.
26
Conclusion
• Benefits of the wavelength flexibility in the
network
1.Improved network capacity
2.Improved fairness or the multi-hop demand.
Disadvantage: This need for WCs and large cross
connects.
• Although wavelength flexible node in the current
networks typically used digital cross-connects
and OEO conversion, the analysis shows that
design on all optical is also feasible.
27
Conclusion
• Optical degree of the wavelength flexibility
depends on many factors.
a. Network topology
b. Traffic assumptions
c. Network management considerations.
• The relative costs of the cross-connects, WCs
and the line systems are more important to
determine the degree to which wavelength
blocking may or may not be tolerated.
28
Discussion
29