impairment-aware routing and wavelength assignment (ia-rwa )
TRANSCRIPT
Impairment-AwareRouting and Wavelength
Assignment (IA-RWA)
WDM link
Generation of multiple streams of light each at a different wavelength Combination of the streams into an optical fiber (Single Mode) Amplification of the optical signals as required Separation of the multiplexed stream into its component streams Reception of the optical streams by wavelength specific receivers
Point-to-Point vs Wavelength Routed
Point-to-Point WDM Electrical Packet Switching Packet processing overhead Efficient bandwidth utilization Poor scalability, Good flexibility High energy consumption
Wavelength RoutedCircuit switching (end-to-end) No packet processing Inefficient bandwidth utilization Good scalability, Mediocre flexibility Low energy consumption
P2P WDM P2P WDM
P2P WDM
AB
C
D
P2POpaque
OXC OXC
OXC
AB
C
D
Transparent
Wavelength-Routed WDM networks Physical topology: A set of routing nodes connected by fiber links Optical Cross-connect - OXC: No O-E-O conversion Lightpath: A lightpath has to be setup before the data transmission. A
Lightpath remains in the optical domain from src to dst Logical topology: The set of src-dst pairs connected through lightpaths
1
2
3
46
5
7
1
2
3
W ave l e ng th R o ute r s :
L i g htpaths :
OXC
OXC
OXC
OXC
OXCOXC
OXC
Wavelength reuseWavelength reuse
# wavelengths # wavelengths << << # OXCs# OXCs
Routing and Wavelength AssignmentRouting and Wavelength Assignment
OXC Architecture
3x3switch
3x3switch
1
1
1
1
1
1
2
2
2
2
2
2
1
2
1
2
1
2
1
2
1
2
1
2
Optical Cross Connect
(OXC)
Protection Mechanisms
The 1+1 protection. No protocol is neededWorking fibre
Protection fibre
The 1:1 and 1:N protection. Signaling protocol is needed
1+1 is faster than 1:1 but in the latter case the sparefibre could be used for low priority traffic (extra Tx, Rx)
Wavelength Routing Pros and Cons
Setting up a lightpath is like setting up a circuit (a 2-way process
with Req and Ack): RTT = tens of ms
Pros: good for smooth traffic
Mature OXC technology (msec switching time)
QoS guarantee due to fixed BW reservation
Cons: BW inefficient for bursty (data) traffic wasted BW during off/low-traffic periods
very coarse granularity (OC-48 and above)
limited # of wavelengths (thus # of lightpaths)
RWA: Routing and Wavelength Assignment Definition
Given: network topology, end-to-end connection requests
Problem: Determine routes and wavelengths for the requests
Offline RWA (network planning phase) The entire set of requests are given in advance (traffic matrix).
Online RWA (network operation phase) Requests arrive randomly over time and are served one-by-one
Objective: Minimizing the Overall Blocking Probability
Transparent wavelength routed networks
All-optical transparent networks: advantages in capacity, cost and energy The transmission quality is affected by physical layer impairments (PLIs) Physical layer blocking: the signal detection at the receiver may be
infeasible Impairment aware (IA)-RWA algorithms
Pure RWA - problem definition
Input:
Network topology: connected graph G=(V,E)
V: set of nodes, assumed not to be equipped with wavelength converters
E: set of point-to-point single-fiber links
Each fiber is able to support a set C={1,2,…,W} of W distinct wavelengths
A-priori known traffic scenario given in a matrix of nonnegative integers Λ
Output:
the RWA instance solution, in the form of routes and assigned wavelengths
the number of wavelengths required to route all the connections
Objective: minimize the number of used wavelengths
LP Formulation and Flow Cost FunctionFlow Cost Function
1
ll l
l
wF f w
W w
Increasing and Convex (to imply a greater amount of ‘undesirability’ when a link becomes congested)
Approximated by a piecewise linear function Integer break points (makes Simplex yield
integer optimal solutions with high probability)
Parameters: s,d V: network nodes wC: an available wavelength l E: a network link pPsd: a candidate path
Constant: Λsd: the number of requested connections from node s to d
Variables: xpw: an indicator variable, equal to 1 if path p occupies wavelength w, else 0 Fl: the flow cost function value of link l
RWA LP FORMULATION
minimize : ll
F
subject to the following constraints:
Distinct wavelength assignment constraints,
|
1,pwp l p
x
for all lE, for all wC
Incoming traffic constraints,
sd
pw sdp P w
x
, for all (s,d) pairs
Flow cost function constraints,
|
l l pwp l p w
F f w f x
The integrality constraint is relaxed to 0 1.p wx
We obtain integer solutions in 98% of the problem instances!
Random perturbation
In the general multicommodity problem, a flow
that is served by more than one paths has equal
sum of first derivates over the links of those
paths and also these paths are of equal length
In our problem a request that is served by more
than one lightpaths has equal sums of first
derivates over the links of these paths
To avoid such cases, we multiply the slopes of
each variable on each link with a random
number that is close to 1
In this way, the cases that two variables have
equal derivates over the links that comprise a
path are reduced, and thus we obtain more
integer solutions
X2=0.5X1=0.5
Handling non-integer solutions
Make Simplex yield integer optimal solutions
Piecewise linear cost functions
Random perturbation technique
Still the solution may be non-integer
Iterative fixings Fix the integer variables of the solutions and solve the remaining (reduced) LP problem
The objective cost does not change if we get to an integer solution it is optimal
When fixing does not further increase the integrality, we proceed to the rounding process
Iterative rounding Round a single variable, the one closest to 1, and continue solving the reduced LP problem
Rounding helps us move to a higher objective and search for an integer solution there
If the objective changes we are not sure anymore that we will find an optimal solution
Pure RWA algorithm
Use a pure RWA algorithm that is based on a LP-relaxation formulation
The algorithm consists of 4 steps
1. We calculate a set of candidate paths
2. Using the set of candidate paths we
formulate the RWA instance as a LP problem
and use Simplex to solve it
3. We handle a fractional (non-integer) solution,
by applying iterative fixing and rounding
methods
4. We handle non infeasible instances (when
the RWA instance cannot be served with the
given number of wavelengths)
Traffic Matrix ΛNetwork Topology G=(V,E)
Number of Available Wavelengths W
k
RWA formulationLP relaxation
RoundingRound a fractional variable to 1 and re-execute Simplex
SolutionRouted lightpaths, blocking
Integer Solution?
Simplex
yes
no
Candidate paths
Calculate the k-shortest paths for all connections (s,d) for which Λsd≠0
Feasible?yes no
Integrality is not further increased
Increase the number of available wavelengths and go to Phase 2Once the solution has been found we remove the additional
wavelengths (blocking >0)
Phase 1
Phase 2
Phase 3
Phase 4
FixingFix the integer variables up-to now and re-execute Simplex
IA-RWA problem
IA-RWA objective: minimize the number of wavelengths used (network layer)
and also select lightpaths with acceptable transmission quality (physical layer)
For IA-RWA algorithms we classify physical layer impairments (PLIs) into:
1st class PLIs: generated by the same lightpath (ASE, CD, PMD, FC, SPM)
2nd class PLIs: generated due to inter-lightpath interference (XT, XPM, FWM)
PLIs of the 2nd class make routing decisions for one lightpath affect and be
affected by decisions made for the other lightpaths
Solution:
1. Worst case interference assumption
2. Actual interference: cross layer optimization
Worst Case and Actual Interference
Worst case interference algo: Consider PLIs that do not depend on interference (1st class PLIs)
Assume all wavelengths active (2nd class PLIs)
Prune candidate lightpaths that do not have acceptable QoT
Hamburg
Berlin Hannover
Bremen
Essen
Köln
Düsseldorf
Frankfurt
Nürnberg
Stuttgart Ulm
München
Leipzig
Dortmund
Actual interference: cross layer optimization algo: Consider PLIs that do not depend on interference (1st class PLIs)
Prune candidate lightpaths that do not have acceptable QoT
Formulate the interference among lightpaths into the RWA
Illustrative example: DTnet topology - single connection request between all (s,d) pairs
The reduction in the solution space can deteriorate wavelength performance
Physical layer evaluation: Q-factor
Use the Q factor to estimate the feasibility of a lightpath The Q factor is related to the BER Analytical formulas can be used to calculate the Q factor
I’1’
I’0’ σ’0’
σ’1’
Proposed IA-RWA algorithms
Indirect IA-RWA algo: Constrain the impairment generating sources
1. the length and the number of hops of a path
2. the number of adjacent (and second adjacent) channels over all links of the
lightpath
3. the number of intra-channel generating sources (lightpaths crossing the same
switch utilizing the same wavelength) along the lightpath
Direct IA-RWA algo: Use the definition of Q factor and noise variance
related parameters to define physical layer constraints into the RWA
(Soft) constrain the number of adjacent channel interfering sources on lightpath (p,w)
B is a large constant used to activate/deactivate the constraint
Similarly we constrain the second-adjacent channel interfering sources
Indirect (Parametric) IA-RWA algo
Number of active adjacent channels(Affected PLIs: Intra-XT, XPM and FWM)
(Soft) constrain the number of intra-XT interfering sources on lightpath (p,w)
B is a large constant used to activate/deactivate the constraint
Similarly we constrain the second-adjacent channel interfering sources
Number of intra-channel XT sources
n0 n1 n2 n3 n4
w w
ww w ww
pp’ p’’’
l1 l2 l3 l4
p’3
ending here
p’’
wsource
pdestination
w
Curry the surplus variables in the minimization objective
Direct (Sigma Bound) IA-RWA algo
For each candidate lightpath (p,w) inserted in the RWA formulation, we calculate an
upper bound on the interference noise variance it can tolerate, after accounting for
the impairments that do not depend on the utilization of the other lightpaths (account
for 1st Class PLIs).
Then using noise-variance related parameters per link we can constrain the
interference (due to 2nd Class PLIs) accumulated on lightpath (p,w)
If the selected lightpaths satisfy these constraints they have, by definition,
acceptable quality of transmission
XPM from adjacent channels XPM from second adjacent channelsintra-XT
2 2 2, ', , ', 1 ', 1 2 , ', 2 ', 2
{ '| '} { '| '} { '| '}XT n p w XPM l p w p w XPM l p w p w
p n p p l p p l p
s x s x x s x x
2max
{ | endof }
( , )pw pl p n l
B x S p w B
2 2 2,'1' ,'1' max( , ) ( , ) ( , )XT XPMp w p w p w
Performance evaluation results
Simulation platform
Matlab + LINDO API
Generic DT network topology
Traffic Scenarios
Random traffic matrix generator
DTnet actual traffic matrix
Physical Layer Evaluation: Q-Tool
Developed within DICONET project
Uses analytical models to calculate the
Q factor of lightpaths
Realistic physical layer parameters
Hamburg
Berlin Hannover
Bremen
Essen
Köln
Düsseldorf
Frankfurt
Nürnberg
Stuttgart Ulm
München
Leipzig
Dortmund
Node
SMFPre-DCM DCF
Node
Post-DCMSMF
N-1 spans
N-th SMF span
NodeNode
SMFSMFPre-DCMPre-DCM DCFDCF
NodeNode
Post-DCMSMFSMF
N-1 spans
N-th SMF span
Pure RWA performance
100 RWA instances ILP min-max: optimality criterion LP min-max: running time & integrality criteria The proposed LP-relaxation+piecewise linear
costs has superior performance The performance is Improved with the
random perturbation technique
100 100 100 100
61 6359
46
87 88 87 88
100 98 98 98
0
20
40
60
80
100
0.5 1 1.5 2Load
Inst
ance
s su
re t
o ob
tain
opt
imal
sol
utio
nILP
LP-min_max
LP-piecewise
LP-piecewise+random_perturbation
7
92
840
3992
1723
3038
2
8
17
31
1
5
15
31
1.E+00
1.E+01
1.E+02
1.E+03
1.E+04
0.5 1 1.5 2Load
Ave
rage
run
ning
tim
e (s
ec)
ILP-min_max
LP-min_max
LP-piecewise
LP-piecewise+random perturbation
7.70
14.01
20.53
26.66
7.76
14.03
20.56
26.68
7.72
14.02
20.54
26.68
7.70
14.02
20.53
26.66
0
5
10
15
20
25
30
0.5 1 1.5 2Load
Ave
rag
e n
um
be
r o
f w
ave
len
gth
s
ILP-min_max
LP-min_max
LP-piecewise
LP-piecewise+random perturbation
Indirect and Direct IA-RWA 100 RWA instances
W=16 available wavelengths
Algorithms: Pure RWA
Indirect P-IA-RWA
Direct SB-IA-RWA
The proposed IA-RWA algorithms reduce
the (physical layer) blocking
Additional wavelengths are required to
spread the lightpaths and avoid interference
The direct SB-IA-RWA algo can find zero
blocking solutions
The direct SB-IA-RWA algo maintains zero
blocking up to ρ=0.8, after which the 16
available wavelengths are not enough
0
2
4
6
8
10
0.5 0.6 0.7 0.8 0.9 1
Load
Blo
ckin
g ra
tio
pure RWA
P-IA-RWA
SB-IA-RWA
Load ρ=0.5
Algorithm Average
number of wavelengths
% of optimal
solutions
Average number of fixing and
roundings
Average running
time Pure RWA 7.70 0.94 1 1.27
P-IA-RWA 10.32 0 23 112.2
SB-IA-RWA 9.27 0 15 20.4
Direct IA-RWA algo performance
Direct SB-IA-RWA algorithm, solved using
The proposed LP-relaxation technique
ILP
100 random RWA instances
Find zero blocking solutions
Using ILP we were able to solve all
instances within 5 hours up to ρ=0.7 load
Using the LP-relaxation the optimality is
lost in 2 or 3 instances but the execution
time is maintained very low
9.27
10.78
12.72
14.17
15.32
16.28
9.24
10.76
12.70
9
11
13
15
17
0.5 0.6 0.7 0.8 0.9 1Load
Wa
vele
ng
ths
to a
chie
ve z
ero
blo
ckin
g
LP SB-IA-RWA
ILP SB-IA-RWA
20.4
47.662.4
101
179252
40.7
227.4
2550
1.E+01
1.E+02
1.E+03
1.E+04
0.5 0.6 0.7 0.8 0.9 1Load
Ave
rage
run
ning
tim
e (s
ec)
LP SB-IA-RWA
ILP SB-IA-RWA
Realistic traffic matrix
Realistic traffic matrix
(381 connections load ρ=2.05)
The propose IA-RWA algorithms
reduce the physical layer blocking
The direct SB-IA-RWA finds zero
blocking solution
with W=36
Running time: 20 minutes
acceptable for the realistic network
and traffic load
0
3
6
9
12
15
30 32 34 36 38 40Number of available wavelengths
Blo
ckin
g ra
tio
pure RWA
P-IA-RWA
SB-IA-RWA
Dynamic ΙΑ-RWA Algorithm
Input:
New connection request
Current network state
Objective: serve the connections and minimize blocking
over (infinite) time
We use a multicost algorithm with 2 phases
1. Calculate the set of non-dominated paths from the given
source to the given destination
2. Choose the lightpath that minimizes the objective function
(offline algos) (online algos)(offline algos) (online algos)
Calculating the Set of Non-Dominated Paths
Cost vector of link l:
Vector maps the utilization of wavelengths
The cost vector of path p can be calculated based on the cost vectors
of links l=1,2,...,m, that comprise it
The cost parameters of a path can be combined so as to calculate the
Q factors of the available lightpaths over that path
Prune the solution space
For each p, we check the Q factor of available lightpaths and we make
unavailable those that do not have acceptable performance
Stop extending the paths that do not have σταματάμε να επεκτείνουμε
μονοπάτια αν δεν έχουν τουλάχιστον ένα διαθέσιμο μήκος κύματος
Vl = (dl, lG , 2'1',l , 2
'0',l ,lW )
22 10100,1,
11
2 2
'1', '0 ',1 11 1 1
221010 (1,..., ),, , ,,( , , , ,* ) , &
mm
lli li l
iim mm m m
lll ll ll l l
GG
lp l l lWd mGV d G W p
lW
Q Qmax
dmin
Calculating the Set of Non-Dominated Paths
Domination relationship between two paths
p1 dominates p2 (p1 > p2) iff
Using the above definitions we use a multicost algorithm, which is a
generalization of Dijkstra algorithm, to compute the set of non-dominated
paths Pn-d from the given source to the given destination
By definition, the paths that are included in Pn-d have
At least one available wavelength
The available wavelength have acceptable transmission performance
(Q factor)
Q QQmax
dmin
1 2 1 2 1 2 and andp p p p p pd d W W Q Q
Optimization Policies
i) Most Used Wavelength (MUW)
We order the lightpaths to decreasing
wavelength utilization order and select the
one that is used more in the network.
ii) Better Q performance (bQ)
We select the lightpath with the higher Q
factor value
iii) Mixed better Q and most used wavelength
(bQ-MUW)
From the set of available lightpaths we select
those with Q values no less than 0.5dB than
the highest Q value and then apply the MUW
policy to this new set of lightpaths
1.E-03
1.E-02
1.E-01
1.E+00
100 120 140 160 180 200load (erlangs)
Blo
ckin
g pr
obab
ility
(%
)
MUW
bQ
bQ-MUW
`
1.E-03
1.E-02
1.E-01
1.E+00
100 120 140 160 180 200load (erlangs)
Ave
rage
num
ber
of r
erou
tings
MUW
bQ
bQ-MUW
`
We evaluated 3 optimization policies (that correspond to 3 different IA-RWA algorithms)
The “whole” picture
Continuing work to build the NPOT….
Network Planning and Operation Tool
(NPOT)
Network Planning and Operation Tool
(NPOT)
Planning ModePlanning Mode Operation ModeOperation Mode
Distributed Integration Scheme
Distributed Integration Scheme
Centralized Integration Scheme
Centralized Integration Scheme
•Offline IA-RWA
•Regenerator Placement
•Monitor Placement
•Offline IA-RWA
•Regenerator Placement
•Monitor Placement
•Online IA-RWA
•Failure localization
•QoT Degradation
•Online IA-RWA
•Failure localization
•QoT Degradation
•Online IA-RWA
•Failure localization
•QoT Degradation
•Online IA-RWA
•Failure localization
•QoT Degradation
Control