Вы находитесь на странице: 1из 120

WDM Network Elements

Based on: Optical Networks a Practical Perspective (2nd Edition) –


Chapter 7, by R.Ramaswami, K.N.Sivarajan
Architectural Aspects of
Network Elements
! Optical Line Terminals (OLT) – widely
deployed today
! Optical Add-Drop Multiplexers (OADM)
– some deployment has been done
! Optical Crossconnects (OXC) –
deployment just starting
WDM Network Architecture

! Optical line amplifiers are not shown (all elements may include optical amps)
WDM Network Architecture
! OLTs are placed either at the end of links or
in point-to-point configurations
! OADMs are used at places where some
fraction of the wavelengths need to be
terminated and others need to be added and
are typically in linear or ring topologies.
! OXCs enable mesh topologies and switching
of wavelengths.
! Clients of these networks can be ATM,
SONET, IP switches using the optical layer.
Important Features of WDM
Architecture
! Each link can support a number of wavelengths (physical
limitations!)
! Wavelength Reuse: Multiple lightpaths can use the same
wavelength in the network as long as they do not overlap on
the same link.
! Wavelength conversion: lightpaths may undergo conversion
along the lightpath for better utilization/adaptation of signals.
! Transparency: optical layer is protocol insensitive.
! Circuit Switching: lightpath establishment on demand. (no
packet switching at the optical layer!)
! Survivability: in event of link/node failures lightpaths can be
rerouted (resiliency!)
! Lightpath topology: graph representation of nodes and
links/lightpaths between them (the view of the higher layer)
Optical Line Terminals (OLTs)
Elements inside OLTs
! Transponders
! Wavelength multiplexers (demultiplexers)
! (Optical amplifiers)

Non ITU λ Transponder ITU λ Mux/demux


1

IP router O/E/O
Non ITU λ ITU λ2
O/E/O λ1 λ2 λ3
SONET λ1 ,
ITU λ3 λOSC
SONET Laser λOSC
Receiver

Optical line terminal


Transponders
! Adapts a signal to be transmitted in the WDM network.
My include a simple OEO conversion or optical
wavelength conversion (research labs). The interface
between the client and the transponder may vary
(depending on: bit-rate, distance, loss, etc.)
! Most likely SONET i-face is short-reach (SR) but can be
a very-short range (VSR) interface for >=10Gbps.
! The signal generated by the transponder shoguld
(optimally) conform to ITU standards.
! Transponders may add networking functionality such
as: overhead for management purposes or forward
error correction (FEC – OEO required!).
! May Monitor BER of signal.
Need for Transponders
! In some cases the clients can receive on ITU
wavelengths so transponders are only needed for
transmission
! In some other (fortunate) cases the adaptation is
done inside the client equipment reducing the cost of
the OLT, however these interfaces are not
standardized and are likely to be vendor dependent.
! The transponders create the bulk of the cost, size
and power consumption of OLTs, thus the number of
transponders should be kept minimal.
Optical Multiplexers
! Any multiplexing technology can be
used
! Optical amplifiers may be used to boost
signals in both directions (for reception
as well as transmission).
Supervisory Channel in OLTs
! Optical Supervisory Channel (OSC) is
carried on a separate wavelength
! OSC is used to monitor the performance
of amps on the links as well as for
management functions (performance,
fault, configuration, security,
accounting) .
Optical Line Amplifiers
Optical Line Amplifiers
! Placed in the “middle” of optical fibre with a distance
of about 80-120km.
! EDFA is currently the most used amp.
! Typical amps cascade two or more gain blocks with
mid-stage access.
! In the mid-stage compensating elements can be put
(e.g., chromatic dispersion compensators, flat gain
compensators, maybe OADMs).
! Amplifiers also contain gain control and performance
monitoring capabilities.
! Employment of Raman amplification has just started,
where a laser pumps light in the opposite direction of
the signal.
Optical Line Amplifiers
! The optical supervisory channel is terminated at the
input and re-injected at the output (OEO conversion,
electronic processing).
! In a system using both C and L band, bands are
separated and employ separate EDFAs.
Dispersion
compensator
λ1, λ2,…, λW
OADM
λOSC
Gain stage Gain stage

Raman Receiver Laser


pump
laser
Optical Add/Drop Multiplexers
(OADM)
OADMs
! May be used in OLAs (seen previously).
! Can be used as stand alone network
elements.
! OADMs can save on costs significantly, by
reducing the number of point-to-point
connections (terminations), thus reducing the
number of OLTs (and transponders, that
generate most of the cost).
! In the first of the following pictures - at node
B six out of eight transponders are connected
back-to-back – what a waste!
OADM vs. OLT
(a)
Node A Node B Node C

OLT

Transponder
Add/Drop
(b) Node B Node C
Node A

OADM

Optical passthrough

Add/Drop

Note, that transponders can be skipped in the first picture, if those OLTs
Are engineered in that way. (remember power levels and required SNR!)
OADM Architectures
! OADMs have two line ports and a number of local
ports. Attributes are:
! Total number of λ-s supported
! Number of λ-s that can be added/dropped
(adropped).
! Any constraints on what λ-s can be adropped?
! How easy to upgrade (λ-s to be adropped)?
! Is it modular? Is cost proportional to the number of
λ-s to be adropped?
! Complexity of its physical layer. Is the pass-through
loss, crosstalk, etc. growing with additional λ-s?
! Is it reconfigurable (software control) on what λ-s
can be adropped? Today’s OADMs are rather
Parallel OADM Architecture
Demux Mux
λW

λ1, λ2,…, λW λ1, λ2,…, λW


λ2

λ1

Drop Add

! No constraints on what λ-s can be adropped (minimal constraints on planning


lightpaths).
! Loss is fixed (adropping additional channels is easy).
! Not cost effective if adropping small number of λ-s.
! Since all λ-s are always re-multiplexed, the tolerance of lasers/filters must be
stringent.
Modular Parallel OADM
Architecture
Demux Mux
Band 4

Band 3
Band 2
λ1, λ2,…, λW λ1, λ2,…, λW

Band 1

λ1,λ2

Drop Add
! Implies constraints on what λ-s can be adropped.
! Cost effective also if adropping small number of λ-s.
! The tolerance of lasers/filters can be higher
! Loss is fixed (adropping additional channels is easy).
! Modular multistage approaches are also used today
! Loss is not uniform for all λ-s.
Serial OADM Architecture

λ1, λ2,…, λW λ1, λ2,…, λW

λ1 λ2
Drop Add

! A single channel is adropped (SC-OADM). To drop multiple channels,


SC-OADMs can be cascaded.
! Adding additional SC-OADMs disrupts existing channels for a short
while, thus planning is needed ahead of time.
! Highly modular (cost is low for less λ-s).
! Loss increases with λ-s to be adropped which may require additional
OLAs.
Engineering Problems with
SC-OADMs
Band-drop OADM Architecture
λ1, λ2,…, λW λ1, λ2,…, λW

Drop Add

λ1, λ2, λ3, λ4

! Fixed group of λ-s is adropped and undergo a further level of


demultiplexing.
! Adropping additional λ-s does not effect loss (if these λ-s are in
the group).
! Complicates λ planning (constraints)
Reconfigurable OADMs
! Do not only reduce maintenance cost
but enable higher flexibility.
! Transponders need to be deployed
ahead (modular cost??) and tunable
transponders are yet expensive.
! Reconfigurable OADMs can be viewed
as OXCs.
Reconfigurable OADMs
Reconfigurable OADMs
Optical Crossconnects (OXC)
OXCs
! OXCs are required to handle mesh topologies
(OADMs have only two ports making them available for ring and point-
to-point only).

! OXC are also key elements for


reconfigurability.
! Some ports are connected to other OXCs and
some are terminated by optical layer client
equipment (SONET,ATM, etc.).
! OXCs usually do not contain OLTs (separate
products).
A Typical OXC

OLT
OXC

IP SONET ATM
SDH
Key Functions of OXCs
! Software controlled service provisioning
! Protection of lightpaths against defects
! Bit rate transparency is a desirable attribute
! Performance monitoring capabilities (testing,
non-intrusive troubleshooting)
! Wavelength conversion may be incorporated
! Multiplexing and grooming of STS signals can
be incorporated (electrical domain)
OXC Main Parts
! Switch core – the switch that actually
performs the crossconnect function
! Port complex – port cards or interfaces to
other equipment

! OXCs and OLTs can be interconnected in


different ways:
! Opaque configurations: the signal is converted in
the E domain
! All-optical: signal remains in the optical domain
Optical vs. Electrical Core
! Electrical core has a total switch capacity e.g.,
2.56Tbps. This can be used to switch e.g., 1024 OC-
48 or 256 OC-192 signals.
! Optical core cannot offer grooming or switching at
lower signal speeds
! Optical core is bit-rate independent (and transparent)
the cost is the same no matter what signal it is
switching (unlike electrical ports).
! Optical core is more scalable thus future proof and
may allow to switch groups of wavelengths
inexpensively.
Different OXC Architectures

! The size, power consumption, cost improve down the figures.


Properties of Architectures
! Electrical cores rely on the WDM signals to be demultiplexed
and fed in using e.g., SR optical interfaces (1310nm range).
Wavelength conversion is easy!
! The first two architectures can easily do embedded performance
monitoring, BER can be used e.g., for protection switching. In-
band overhead signalling channels may be used.
! The bottom two architectures cannot do that, they need out-of-
band signalling channels
! All-Optical OXCs from different vendors are unlikely to
cooperate (different link-engineering).
! Migration from the second to the third architecture is relatively
easy but it will require equipment from the same vendor.
Comparison of Architectures
Attribute Opaque Opaque Opaque All-Optical
Switch-core Electrical Optical Optical Optical
Figure a) b) c) d)
Grooming Yes No No No
Capacity Low High High Highest
λ conversion Yes Yes Yes No
Switching BER BER Optical Power Optical Power

Cost/port Medium High Medium Low


Power cons. High High Medium Low
Footprint High High Medium Low
All-Optical OXCs
! The bottom architecture of the last picture.
! No:
! Low speed grooming
! Wavelength conversion
! Signal regeneration
! One solution is to combine the optical core
(groups of wavelengths or entire fibres can be
switched at once) with an electronic core.
! Since there is no wavelength conversion,
architecture can be simplified – wavelength
planes.
Combined Optical – Electrical
Core
Wavelength Planes
WDM Network Design

Based on: Optical Networks a Practical Perspective (2nd Edition) –


Chapter 8, by R.Ramaswami, K.N.Sivarajan
Example of a Design Problem
A B C
(a)
Router

Router

Router
•••
•••

•••
•••
••• (b)

•••

Router
Router

•••
•••

••• •••
Router
(c)
! Line speed: 10Gbps
! 50Gbps is needed between each of the nodes.
Example of a Design Problem
! 10 router ports can be saved with the
employment of an OADM.
! This may cost more but the number of router
ports (and their cost), and transponders can
easily satisfy the employment of an OADM (today
and in the future).
! The lightpath topology encountered by the
routers is significantly different although both
scenarios are valid.
! If the required bandwidth is significantly less
than the line speed, then a router may be
beneficial.
Formulation of Design Problems
! Given: the fiber topology and the traffic matrix (traffic
requirements).
! Task: designing the lightpath topology that is superposed
on the fiber topology while satisfying the traffic
requirements (lightpath topology design problem).
! The routing and wavelength assignment problem
(RWA) is similar to the lightpath topology problem,
except that the design has to be done within the optical
layer (as in the second example).
! Grooming of traffic may be necessary to come up with
efficient traffic matrices.
! In general it is a good idea to separate this problem (like
we did before) since solving the two problems together is
quite hard.
Cost Trade-offs
Measurement of Cost
1. Higher layer equipment cost is determined by
the number of ports required (e.g., IP ports).
2. At the optical layer an important cost factor is
the number of transponders needed. Since
every port requires a transponder (and every
lightpath two), this cost can be coupled to that
of 1.
3. Optical layer equipment cost is estimated by
the number of wavelengths employed on links.
2-connected
! There should be 2 (node-wise) disjoint routes
between every pair of nodes.
! 2-connected (but arbitrary) mesh topologies
are more cost effective for large networks,
but ring topologies (always 2-connected) are
good for networks with less geographic
spread.
! A ring topology has the minimum number of
links (N) connecting N nodes that keeps a
topology 2-connected, thus it has a low fiber
deployment cost
Assumptions for Example
! Topology is a ring.
! N: number of nodes
! t: is the (units of) traffic generated at each node
to be routed to remote nodes.
! Destination distribution is assumed to be uniform
resulting in t/(N-1) units of traffic routed
between every pair of nodes in the ring (thus
there must be a way to reach each router from
the others).
! The capacity of each λ is assumed to be 1 and
there are no wavelength conversion capabilities.
Important Metrics
! Router ports: should be kept at a
minimum
! Wavelengths: should be kept at a
minimum
! Hops: the maximum number of hops
taken by a lightpath
! There is a trade-off between this
parameters
LTD: Point-to-point WDM Ring
(PWDM)
! Lightpath topology is a ring with several links (several
wavelengths)

D B

C
LTD: Hub Design
! All routers are connected to a central hub (packet travel
over two lightpaths: source->hub and hub-
>destination)

A C

Hub

B D
LTD: Mesh topology
! All-optical design: a lightpath between all pairs of routers.

A C

B D
RWA for PWDM Ring Lightpaths
λ2

IP 1 λ1 IP 4

OLT 1 OLT 4

λ2 λ1 λ1 λ2
OLT 2 OLT 3

IP 2
λ1 IP 3

! Single-hop lightpaths λ2
! W denotes the number of wavelengths on each link.
RWA for PWDM Ring
! Assuming N is even, and that each stream is
routed on the shortest route, the traffic load
(L) on each link is: 1
N +1+
L= N −1 t
8
Thus: W = L 

The number of router ports required per node


is: Q=2W
Hops: for a PWDM ring each lightpath is exactly
one hop
RWA for Hub Architecture
λ2 IP λ2
hub
OADM
IP 1 λ1 hub
λ1 IP 4
OADM OADM
1 4
λ2 λ1
OADM OADM
2 3
IP 2
λ1 IP 3

λ2
RWA for Hub Architecture
!
t  lightpaths are needed from each node to
the hub for transmission and another t  for
reception from other nodes.
! The number of router ports Q = 2 t 
! If wavelengths are reused as seen in the
picture, the number of wavelengths is
N
W = t 
2
! The worst-case hop length is: H=N-1
RWA for All-optical Design

λ1 IP 4
IP 1
OADM OADM
1 4
λ2 λ1 λ1 λ3
OADM OADM
2 3
IP 2 IP 3
λ1

λ3 λ2
RWA for All-optical Design
Between each router
!
2t /( N − 1)
lightpaths are needed
! Thus the number of router ports
 t 
Q = 2( N − 1) 
 N − 1

! The number of wavelength depend on how


the lightpaths are routed. Is is possible to
find WA, so  t  N
2
N
W =   + 
 N − 1 8 4
Comparing Architectures –
Number of Ports
! A lower bound can be calculated for the
number of ports required:
! Q=2t(N-1)/N
Comparing Architectures –
Number of Ports
Comparing Architectures –
Number of Wavelengths
! Lower bound for number of
wavelengths:
! Hmin is the average number of hops
between nodes (on a minimum route),
in a ring: H min = N + 1 + 1
4 4( N − 1)

TotalTraffic
Average load: Lavg = H min *
Numberoflinks
Comparing Architectures –
Number of Wavelengths
A Lightpath Topology Design
Problem
LTD Assumptions
! We assume that there are no constraints on the
physical layer (maximum length of lightpaths or W).
! We assume that lightpaths are bi-directional (two
fibers).
! At each node a maximum of ∆ ports can be used
(this constrains the number of lightpaths to N∆/2).
! We assume that the cost of a lightpath does not
depend on its length (may not hold in long-haul
networks).
! By designing the lightpath topology we also have to
solve the routing problem.
LTD Assumptions
! IP traffic is assumed, with an arrival rate of
packets between source s and destination d
of λsd (packets/second).
! bi,j are (n2-n) binary variables (one for each
possible lightpath; i≠j). bi,j=1 if the design
has a lightpath from i to j.
LTD specifies the values for {bi,j}.
! We assume that we can arbitrarily split traffic
over different paths between the same pair of
nodes.
LTD Assumptions
! Let the fraction of traffic between s and d
that is routed over (i,j) be: ij
sd
a
! Then the traffic in (pkts/sec) for s, d over
λ
(i,j) is: ij
sd
= aij λ
sd sd

The total traffic over (i,j) is: ij ∑sd ij


λ = λsd
!

! Congestion parameter: λmax = max ij λij


! The lambda values have to be determined!
! Let’s assume a Poisson process for arrivals
and exponential distribution of transmission
times with mean: 1/µ seconds.
LTD Problem
! By modeling the traffic generation with a
Poisson process, each link can be modeled as
a Markovian M/M/1 queue.
! The average queuing delay is then: d = 1
µ − λij
ij

! The (max) throughput is the minimum


amount of offered load, where the delay
becomes infinite! (when λmax=maxi,jλij=µ)
! Thus the congestion has to be minimized.
Hardness of LTD Problem
! Objective function and the constraints are linear
functions of different λ-s and b-s.
! A mathematical program with these properties is
called a linear program (LP) if all variables are
real.
! If all variables are integers, it is an integer linear
program (ILP).
! If some variables are integers and some are real
it is a mixed integer linear program (MILP).
! Although LPs are P problems, ILPs and MILPs
have been show to be in general NP-hard.
Hardness of LTD Problem
! Our problem is MILP and is sometimes called the
LTD-MILP.
! General MILPs may be solved by heuristics
(exhaustive search of variable space is too
expensive).
! Heuristics can be based on LPs (LP relaxation and
rounding) or other paradigms (simulated
annealing, genetic algorithms, etc.)
LTD Heuristic based on LP
! If we relax the bij constraint to: 0≤bij ≤ 1 then
the LTD-MILP becomes an LP.
! The value of the solution of the LP is a lower
bound to LTD-MILP called the LP-relaxation
bound.
! So first let us solve the LTD-LP problem and
assign values to bij carefully not to violate the
degree constraints.
! The with the fixed bij values rerun the now
modified LTD-LP and obtain a sub-optimal
solution.
Behavior of LTD-LP-relaxation
For a 14 node example:
Degree LP-relax MILP LP rounding
2 248 388 440
3 189 189 194
4 142 142 142
5 113 113 113
6 94 94 94
Routing and Wavelength
Assignment
RWA
! Given a network topology and a set of lightpath
requests (which can be obtained by an LTD
solution), let us determine the route and the
minimum possible wavelengths.
! If no wavelength conversion is available, then
lightpaths must use the same wavelength over
all links (two lightpaths cannot use the same
wavelength over a common link)
Wavelength conversion
Wavelength Conversion
! Full wavelength conversion can be done in
opaque switches.
! Fixed conversion: is fixed at the time of
deployment (a lightpath entering in port-a on
λi will always exit on port-b λj).
! Limited conversion: a signal is allowed to be
converted from one λ to a limited set of other
λ-s.
! The latter two can save cost on switches but
not transponders (thus it is a theoretical
problem when to employ them, except if the
switch is all optical)
Fixed Wavelength Conversion
Limited Wavelength Conversion
Limited conversion with a degree of 2
Multiple Fiber Pairs
! Multiple fiber pairs may be deployed between nodes to
increase capacity.
! This can be seen as having one fiber pair with limited
wavelength conversion capabilities (the two networks can be
shown to be equivalent).
WA and λ conversion
! Depending on the λ conversion capabilities
(NC,FC,LC,C) the WA problem changes.
! In case of C, the assignment of the
wavelengths is trivial (given the routing and
LTD) and W (the required number of
wavelengths) is equal to the number of
wavelengths required on the link requiring
the most wavelengths to be fit (L). If C is not
available, the question is: how much lager W
is than L.
Graph Coloring and WA
! WA relates to graph coloring:
! Graph representation of the network: G
! Convert it into P(G), where each lightpath is a
vertex and vertexes are connected by edges if they
share a common link in G.
! We have to assign a color to each node in P(G)
so that neighboring nodes do not share the
same color while minimizing the number of
colors.
! The general coloring problem is NP-complete !
Dimensioning Wavelength-Routed
Networks
Wavelength Dimensioning
! The number (and set) of wavelength on each
WDM link have to be determined.
! Today most networks are designed to support a
fixed traffic matrix that can be given in terms of
lightpaths (only RWA) or higher layer traffic (LTD
and RWA).
! The solution of the RWA solves the dimensioning
problem (offline-RWA in the design phase).
! For a network already in operation lightpaths are
added one-by-one and thus the so-called online
RWA has to be solved. Today it is important to
have good solutions for this problem.
Wavelength Dimensioning
! Determining the number of wavelengths per
link also determines how many ports are
needed to be terminated thus determines the
sizes of OXCs and OLTs.
! Today a forecast is made every half a year on
the required traffic matrix to be supported
and networks are upgraded accordingly.
! Capacity planning can also be done by
considering the statistical properties of the
traffic.
Dimensioning of Wavelengths
Based on Statistical Traffic Models
Two Cases for Statistical
Models
1. First-passage model: networks starts with
no lightpaths at all, as requests arrive
(according to a statistical model), they have
to be set up. Wavelengths also depart, but
in average there are more arriving than
terminating requests.
2. Blocking-model: lightpath requests are
treated in the same way as in PSTN
networks, the arrival and termination rate is
assumed to be equal.
First-passage Model
! Network is divergent in the long run.
! The question is: when (T) will a new request
first be rejected?
! Dimensioning: first rejection should happen
with high probability after T.
! Today, since lightpaths are more permanent
this is a good model. Operators should not
reject any requests – upgrading
(dimensioning) is needed. T should be
determined!
Blocking Model
! Network is in equilibrium.
! Request should be honored with a high
probability thus blocked with a very-low
probability.
! This is for the future, when lightpaths
will be provisioned on-demand.
Hardness of Problems
! In general, the analysis problem is
easier to solve than the design (or
dimensioning) problem.
! The easiest way to solve the design
problem to iteratively solve the analysis
problem.
! Thus the analysis of these networks can
be a major focus.
First-Passage analysis
! This is a relatively new field, and
publications just start appearing.
! The mathematical background is rather
complicated thus we will omit it.
Blocking Model analysis
! Offered load: arrival rate of lightpath
requests * average duration.
! Reuse factor: offered load per wavelength
! If the maximum blocking probability is set,
what is the maximum offered load or reuse
factor? Depends:
1. Network topology
2. Traffic distribution
3. RWA algorithm used
4. W (number of wavelengths available)
Blocking Model analysis
! The general problem is hard to solve
even analytically if routing is fixed and
RWA is random.
! The problem becomes even more hard
if routing is not fixed (only simulation
approach), however it is possible to
calculate offered load for small
networks with large W.
Blocking Model - analysis
! Poisson process of arrivals with exponential
holding times.
! RWA: If wavelengths are numbered from 1 to
W, a new request should be assigned the
first available wavelength on the shortest
path over all links on this path.
! Figure shows results for a 32 node random
graph with average nodal degree of 4.
Blocking Model - analysis
The Impact of Wavelength
Conversion on Blocking Probability
for Lightpath Requests
Assumptions
! We assume that the route for each lightpath is
specified.
! With NC the network assigns arbitrary but
identical wavelength over each link on the route
(if available).
! With FC the network assigns an arbitrary
wavelength on each of the links on the route (if
available).
! The probability that a wavelength is used on a
link is π, and π is independent of the same
probability of other wavelengths on the same link
and other links.
Blocking Probability with NC
! π: probability that a λ is used on a link
! W: number of λ-s on each link.
! H: a new lightpath request with H links
in its route.

(
Pb , NC = 1 − (1 − π ) )
H W
Blocking Probability with FC
! π: probability that a λ is used on a link
! W: number of λ-s on each link.
! H: a new lightpath request with H links
in its route.

(
Pb , FC = 1 − 1 − π )
W H
Deriving π from Pb

( )
1/ W
1/ H P
π NC = 1 − 1 − P 1/ W
b , NC
π NC ≈ b , NC

1/ W
 Pb , FC 
π FC = (1 − (1 − Pb , FC ) )
1/ H 1/ W π FC ≈  

 H 
Comparing NC with FC
! Comparing for the same blocking
probability:
π FC
=H 1−1 / W

π NC
! For the NC case, the achievable link
utilization is lower by a factor of the
average hop of lightpaths than that of
FC.
Relaxing Independency
Assumption
! Lets relax the assumption that the probability that
one wavelength is used over a link is independent
of that of other links the following way:
! Interfering lightpath: a lightpath that has already
been established and shares links with the new
request.
! πl: probability that an interfering lightpath is using
let’s say link-i but not link-(i+1). (departure)
! πn: probability that an interfering lightpath is not
using let’s say link-i but will use link-(i+1). (arrivl.)
Relaxing Independency
Assumption
! P(λ is used on link i | λ is not used on link i-1)= πn
! P(λ is used on link i | λ is used on link i-1)= (1- πl)+πnπl
! Pb,NC and Pb,FC can be derived:

(
Pb , NC = 1 − (1 − π n ) )
H W

H
 (π i − π i π n )W π iW 
Pb , FC = 1 − ∏ 1 − 
i =1  1 − π W
i −1 
! Where πi is a function of πl and πn.
Comparing NC with FC
! Comparing for the same blocking probability:

π FC
= H 1−1/W (π n + π l − π nπ l ) if H >> 1 / π l
π NC

! The interference length is Li=1/πl an approximation


to the expected number of shared links between
the new and the interfering lightpath.
! The conversion gain is more in networks where
there is more mixing (dense mesh networks)
(unlike ring networks).
Wavelength Assignment and
Alternate Routes
Routings effect on WA
! Let’s consider four different RW algorithms:
1. Shortest path routing, random wavelength
assignment from available pool
2. Two shortest paths, random wavelength
assignment from available pool
3. Shortest path routing, assignment of wavelength
from the available pool that is the most used
already in the network
4. Two shortest path routing, assignment of
wavelength from the available pool that is the
most used already in the network
Routings effect on WA
! We look at simulation results of a 20 node
39 link network. If we fix the blocking
probability to 1%, the reuse factor can be
determined:
RWA alg. Reuse factor
Random-1 6.9
Random-2 7.8
Max-used-1 7.5
Max-used-2 8.3
Maximum Load Dimensioning
Models
Load Dimensioning
! In an all-optical wavelength routing
network, there are very limited
wavelength conversion capabilities.
! Offline case (all lightpaths are
predefined), if there is FC, then W=L,
otherwise W>=L.
! Online case (lightpath request keep
coming in) is more difficult to
determine.
Offline path request
! Theorem: Given a routing of a set of
lightpaths with load L in a network G
with M edges, with the maximum hops
in a lightpath being D, the number of
wavelengths sufficient to satisfy this
request is:
W≤min[(L-1)D+1,(2L-1)√M-L+2]
WA and Linear Networks
L
λ3
λ2
λ1
(a) (b)

! Line networks are closely related to rings.


! Algorithm for WA:
1. Number the wavelengths from 1 to L. start with the first
lightpath from the left and assign it to wavelength-1.
2. Go to the next lightpath and assign the next number until all
lightpaths are assigned
! In a ring there are two possible routes for each
lightpath
WA and Ring Networks
! If Lmin is the routing with the minimum
possible load (there is an algorithm for
that), then
! Given a set of request for lightpaths
and Lmin to satisfy these requests,
shortest path routing yields a load of at
most 2Lmin.
RWA and NC Ring Networks
! Given a set of lightpath requests and
routing on a ring with load L, WA-NC
can be done with 2L-1 wavelengths

(a)
Cut • •
λ3
λ2
• λ1

(b)
RWA and NC Ring Networks
! It can be shown with graph coloring,
that if no three lightpaths cover the
entire ring, then W ≤ 3/2L is sufficient
to perform wavelength assignment .
! Thus to support all (pathological)
requests we have to employ 1/2L more
wavelengths…
RWA and FC and LC Ring
Networks
! If FC is available, then as long as the
load is smaller then W, all requests can
be accommodated (L≤W)
! It can be shown that if only one node in
the ring has full wavelength conversion
all all others have none, then L ≤ W still
holds!
RWA and FxC and FC Ring
Networks

! If a node has fixed conversion like in the


picture at one node (where λi is converted
into λ(i+1)mod W), and no conversion at the
other nodes, then this network can support
requests with load L≤W-1
! Having two such nodes enables L≤W.
Comparison of the Impact of
Conversion on Diff. Topologies
Network Conversion type

None Fixed Full Limited

Arbitrary Min[(L-1)D+1, L L
(2L-1)√M-L+2)
Ring 2L-1 L+1 L L

Star 3/2L L L

Tree 3/2L L L
Online RWA in Rings
Online RWA in Rings
! Routing of wavelengths is already given
! Lightpaths arrive and depart
! In this situation it is much more hard to
determine L to a given W with NC (with
FC it is yet trivial)
Online Line Network Problem
! If W(N,L) denotes the number of
wavelengths required to support all
online lightpath requests with load L in
a network of N nodes with NC.
In a line network:
W(N,L) ≤L+W(N/2,L) (if N is a power of 2)
Online Line Network Problem
! If W(1,L)=0 thus W(2,L)=L:
! In a Line network with N nodes and
online requests of a load of L, all these
requests can be supported by at most
L*log2N wavelengths with NC.
Online Ring Network Problem
! In a Ring network with N nodes and
online requests of a load of L, all these
requests can be supported by at most
L*log2N+L wavelengths with NC.
Permanent Lightpaths
! If the lightpath requests are permanent,
then all requests can be supported by at
most 2L wavelengths with NC, and
! If there is a limited degree-d wavelength
conversion, then W≤L+max(0,L-d)
Comparing RWA for Rings
Conversion Degree Lower Bound on W Upper Bound on W
Offline traffic model
No conversion 2L – 1 2L – 1
Fixed conversion L +1 L+1
≥2 L L
Online model without lightpath terminations
No conversion 3L 3L
Fixed conversion L 3L
Full conversion L L
Online model with lightpath terminations
No conversion 0.5L[log2N] L[log2N]+L
Full conversion L L

Вам также может понравиться