impact of tra c randomness on restoration coe cient in
TRANSCRIPT
DSTIS 2009 - VIII International Conference on Decision Support for
Telecommunications and Information Society - Coimbra, September 4-7, 2009
Impact of Tra�c Randomness on
Restoration Coe�cient in Optical
Transport Networks
Claunir Pavan, Rui Manuel Morais, Armando Nolasco Pinto
c⃝2005, it - instituto de telecomunicações
Motivation
∙ The dimensioning of optical networks consists in determining the quantitiesof resources necessary to transport a given tra�c.
∙ Commercial tools are available to help us on the dimensioning task.
However, designers depend on a variety of inputs.
∙ Computational e�ort grows rapidly with network size and complexity. For
large networks the time required may be prohibitive.
∙ We aim to develop a method for quickly quantifying network elements
and CapEx with incomplete information.
Introduction
∙ Restoration coe�cient is
the ratio between working and
spare capacity.
∙ ⟨kr⟩= ψminW o
∙ We focus on survivable networks.
∙ Shared mesh restoration
mechanism.
∙ Single link failures.
1
2
4
3
5
1
2
4
3
5
1-3,2-32-4
1-4,3-54-5
1-2,1-32-5
1-4,1-52-5
2-4,3-43-5
1
2
4
3
5
1-32-3
1-5,2-53-5,4-5
1-4,1-52-4,2-5
1-2,1-32-4,2-5
3-43-5
1
2
4
3
5
1-2,2-42-3,2-5
2-5,3-54-5
1-2,1-31-41-5
1
2
4
3
5
1-3,2-33-4,3-5
4-5
1-2,1-32-4,2-53-4,3-5
1-5,2-53-5
1-4,2-43-4
1
2
4
3
5
1-3,2-33-5
1-5,2-53-5,4-5
1-4,2-44-5
1-2,1-32-4,2-53-5
3-4
1
2
4
3
5
1-2,2-32-4,2-5
3-54-5
1-3,1-42-3,2-41-5
2-51-3,2-33-4,3-5
1
2
4
3
53-54-5
1-4,2-4
1-5,2-5
1-3,2-31-2,1-32-4,2-5
3-4,3-5
a) b)
c) d) e)
f) g) h)
6 4
4 6
4
4
1-2,1-32-4,2-53-4,3-5
Calculating Restoration Capacity
∙ The time consuming for this task grows rapidly.
200
250
300
350
400T
ime
(m
inu
tes)
Working capacity
Restoration capacity
0
50
100
150
200 300 500 700 1400
Tim
e (
min
ute
s)
Number of demandsNumber of demands
∙ How to calculate ⟨kr⟩ without extensive simulation?
Calculating Restoration Capacity
∙ In [1] the author published a semi-empirical expression to estimate ⟨kr⟩.
∙ The expression is function of the mean nodal degree.
∙ Indeed, we verify that ⟨kr⟩ depends also on the number of nodes.
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0 10 20 30 40 50 60 70 80 90 100 110
Re
sto
rati
on
co
eff
icie
nt
Number of nodes
⟨δ⟩≈ 2.5 ⟨δ⟩≈ 3.5 ⟨δ⟩≈ 4.5
[1] S. K. Korotky, "Network Global Expectation Model," J. Lightw. Technol. 2004.
Our Approach
∙ We proposed an approximation based on statistical methods.
∙ We developed a model to generate realistic transport network
topologies2.
∙ Wrote a software tool to generate a meaningful data set.
∙ Obtained ⟨kr⟩ from more than 50 thousand transport networks.
∙ Made linear and non-linear regressions.
[2] C. Pavan et al., "Generating Realistic Survivable Transport Network Topologies,"submitted to IEEE/OSA J. Opt. Commun. Netw., 2009.
The Proposed Approximation
∙ From the least square method we obtained a variety of curve �t equations.
∙ Using ⟨δ ⟩ as dependent variable we obtained:
⟨kr⟩= bo+b1
⟨δ ⟩(1)
∙ From (1) and using N as dependent variable we found:
∙ b0 = 1.01−0.21ln(N) and b1 = 0.45N0.32.
∙ And �nally:
∙ ⟨kr⟩= 1.01−0.21ln(N)+ 0.45N0.32
⟨δ ⟩ .
Impact of Tra�c and Randomness on ⟨kr⟩∙ Simulations were conducted over a variety of networks and tra�c.
1
1.1
1.2
1.3
1 2 3 4 5
Re
sto
rati
on
Co
eff
icie
nt
Simulations over RNP topology
Uniform
0 - 4
0 - 10
0 - 20
0 - 50
0.7
0.8
0.9
1
1 2 3 4 5
Re
sto
rati
on
co
eff
icie
nt
Simulations over EON topology
Uniform
0 - 4
0 - 10
0 - 20
0 - 50
0.7
0.8
0.9
1
1 2 3 4 5
Re
sto
rati
on
co
eff
icie
nt
Simulations over OMNICOM topology
Uniform
0 - 4
0 - 10
0 - 20
0 - 50
0.7
0.8
0.9
1
1 2 3 4 5
Re
sto
rati
on
co
eff
icie
nt
Simulations over USA100 topology
Uniform
0 - 4
0 - 10
0 - 20
0 - 50
Impact of Tra�c and Randomness on ⟨kr⟩∙ For networks with N > 20 the maximum di�erence was less than 5%.
0.0
0.2
0.4
0.6
0.8
1.0
0
20
40
60
80
100
120
140
160
uniform 0-5 0-10 0-20 0-50
Re
sto
rati
on
co
eff
icie
nt
Nu
mb
er
of
cha
nn
els
(th
ou
san
ds)
Traffic model
Restoration channels
Working channels
Restoration coefficient
Final Remarks
∙ Tra�c randomness does not impact signi�cantly the restoration coe�cient,
mainly for networks larger than 20 nodes.
∙ Results show that we can estimate the restoration coe�cient with good
accuracy, without complete information about the network.
Contact:
Backup slides
Application for ⟨kr⟩
∙ Estimation of CapEx in Optical
Networks.
∙ Estimation of the number of
network components.
∙ Number of transponders.∙ Number of ports on EXCs.
IP Router
SDH EXC
WDM Terminal OLA
10 GE Cards
STM-64 PoS Cards
STM-64 Gray Cards
10 G Transponders
Cost Model
∙ Total cost of the network is:
∙ Costs of Transmission + Costs of Bandwidth Management
∙ CT =CT RANS +CBWM
∙ Costs of Transmission (links):
∙ WDM terminals + transponders + ampli�ers
∙ CT RANS = L(2γt0+ ⟨tk⟩γt1+ ⟨a⟩γt2)
∙ Costs of Bandwidth Management (nodes):
∙ IP routers + tributary ports + line ports + EXCs + EXC ports
∙ CBWM = N(γi0+ ⟨PIPT ⟩γi1+ ⟨PIPL⟩γi2+ γe0+ ⟨PkEXC⟩γe1)
Cost Model
∙ Normalized cost for the network components4.
Component Variable Cost
IP/MPLS Router Basic γi0 16.67
10 GE Interface γi1 1.05
STM-64 PoS Interface Card γi2 4.58
SDH/OTN EXC Basic γe0 26.67
STM-64 Gray Interface Card γe1 0.67
WDM Terminal, LH (40 channels) γt0 4.17
WDM Transponder, LH (10 Gbits/s) γt1 1.00
Optical Line Ampli�er, LH (100 km span) γt2 1.92
[4] R. Huelsermann et al., "Cost modeling and evaluation of capital expenditures in opticalmultilayer networks" J. Opt. Netw., 2008.
Cost Model
Variable Expression
Average nodal degree ⟨δ ⟩= 2LN
Average link length ⟨s⟩= 1L ∑
Ll=1 sl
Average OLAs ⟨a⟩= 1L ∑
Ll=1
⌈sl
span
⌉−1
Average tra�c demand ⟨d⟩= 2N ∑
N−1i=1 ∑
Nj=i+1 di j = N−1
Average number of hops ⟨h⟩= 1D ∑
N−1i=1 ∑
Nj=i+1 hi jdi j
Cost Model
Variable Expression
Average number of demands ⟨W o⟩= ⟨d⟩⟨h⟩⟨δ ⟩
Restoration coe�cient ⟨kr⟩= ψ
⟨W o⟩L
Average number of IP Ports PIPT = PIPL = ⟨d⟩
Average number of EXC Ports (surv.) ⟨PkEXC⟩= ⟨d⟩[1+ ⟨h⟩](1+ ⟨kr⟩)
Average number of transponders (surv.) ⟨tk⟩= 2⟨W o⟩(1+ ⟨kr⟩)
Required Approximations
∙ In the lack of topological information, some variables must be estimated.
Variable Expression
Average link length ?
Average number of hops ?
Restoration coe�cient ?
Actual Approximations
Cost Expression
Average link length1 ⟨s⟩ ≈√
A√N−1
Average number of hops1 ⟨h⟩ ≈√
(N−2)(⟨δ ⟩−1)
Restoration coe�cient ⟨kr⟩ ≈ 1.01−0.21ln(N)+ 0.45N0.32
⟨δ ⟩
[1] S. K. Korotky, "Network Global Expectation Model," J. Lightw. Technol. 2004.
Experiments
∙ European Optical Network, with A=11.180.000 km2.
∙ Routing through shortest path strategy.
∙ Restoration through the shared path technique.
Results
∙ Quantities of networkcomponent.
∙ EXC gray ports with
error of 10%.
∙ Transponders with error
of 2%.
∙ OLAs with error of 0%.
1000
1500
2000
Qua
ntiti
es o
f com
pone
nts OPNET
Approximation10%
2%
0
500
OT
N E
XC
s B
asic
OT
N E
XC
Gra
y P
orts
IP R
oute
r
IP R
oute
r G
E
Car
ds
IP R
oute
r P
oS
Car
ds
WD
M T
Ms
Tran
spon
ders
OLA
s
Qua
ntiti
es o
f com
pone
nts
Network componentsOT
N E
XC
s B
asic
OT
N E
XC
Gra
y
Network components
Results
∙ CapEx for BWM with
error of 3.5%.
∙ CapEx for TRANS with
error of 1.5%.
∙ CapEx for TOTAL cost
with error of 2.5%.
4000
5000
6000
7000
Nor
mal
ized
cos
ts
OPNET
Approximation
3.5%
2.5%
0
1000
2000
3000
CBWM CTRANS CTOTAL
Nor
mal
ized
cos
ts
C C C
1.5%
CBWM CTRANS CTOTAL
Network costsCBWM
CTRANS
CTOTAL