Migration Planning for Optical Backbone Networks · • Usually SDH, combined with MPLS • DWDM...

Faculty for Electrical Engineering and Information Technology, Chair for Telecommunications

Migration Planning for Optical Backbone NetworksOptical Backbone Networks

ITG Workshop Stuttgart 2009ITG Workshop, Stuttgart 2009“Optimierung und Optimierungsverfahren, -algorithmen”

TU DresdenTU DresdenStefan Türk

D d 13 11 2009Dresden, 13.11.2009

Agenda

1. Introductiono Project 100Geto Optical core networkso Optical core networkso Techno economicso Migration

2. Migration/Planning Approacheso Traffic modelo Traffic modelo Cost modelo ILPo Heuristicso Grooming

3. Resultso Utilizationo Performanceo Publications o Publications

4. Summary

TU Dresden, 2/1/2010 Migration Planning for Optical Backbone Networks

Slide 2

Introduction


Slide 3

Project 100Get-E3

• EUREKA-Initiative CELTIC (Cooperation for a sustained European Leadership in Telecommunications)

• 100 Gbit/s Carrier-Grade Ethernet Transport Technologies (100GET)• NSN guided project part "End-to-End Ethernet (E3)„• TUD-TK contribution

o Network technology migrationo Multicast resilienceo u as s

• Sponsors

• Partners


Slide 4

Optical core networks

• Transport network between cities or countries• Usually SDH, combined with MPLS• DWDM technologygy

• In Germany:o Biggest provider DTAG

O Go 74 core POP in Germany

• Metro networks connected to backbone• Access networks connected to metro


Slide 5[No2]

Techno economics

“Calculation where in a production chain [netplanning] technical innovation will bring planning] technical innovation will bring significant cost-savings or return of investment”

• Usually working with cost functions• Considering:

o CAPEX Cost for device introduction

o OPEX Cost to operate the device

o IMPEX Cost of software implementationCost of software implementation

o Models for price developmento Models for penalty


Slide 6

Migration

• Transformation from an “old” network into a “new”• Our case: IP/MPLS/SDH/WDM IP/MPLS-TP/ETH/OTN

Questions: • Why migration?

o Cost reductiono Energy reduction

IP Router

ETH Switch

OXCo Energy reduction

• How to migrate cost optimal?o Node and link ordero Different layerso Constraintso Constraintso Time

• Possible approacheso Exact vs. heuristic

S l l lo Single vs. multi layero Incremental vs. all period


Slide 7

Migration

• Current reference scenario• IP/MPLS/SDH/DWDM

• Evolutionary scenario• IP/MPLS-TP/ETH/OTN

• Expected transition of demands from IP to ETHSS O C h d f d

[No2]

• WSS-OXC with defined AD-ratio


Slide 8

Single layer migration

IP layer 1

4

5

1

45dIP dIP

23

23

Transport layer

Step 0 Step 1


Slide 9

Multi layer migration

IP layer 1

4

5

1

45dIP dIP

dE

23

23

54

Ethernet layer

54

Transport layer

Step 0 Step 1


Slide 10

Multi period migration( )

• Incremental planning (“worst” case)

Costs

( )

P1 P2 P3 P4 P5time

Planning solution costs

• All period planning (best case)

P1 P2 P3 P4 P5time

[M ]


Slide 11

[Meu]

Migration/Planning Approaches


Slide 12

Phases Planning of Greenfield networks (Start/End)Planning of Greenfield networks (start/end)

ON

Constraint definition System model definition

Calculation of optimal migration steps

Heuristic approach for global optimum (change order)

ILP system description (CPLEX, AMPL)

Constraint definition System model definition

ILP system description (CPLEX, AMPL)

IGR

ATIO Calculate cost optimal step (local optimum)

Calculate routing

Shortest path, load distribution, …

Grooming, wavelength and fiber distribution, …W

OR

K M

I

Calculate migration order

According to constraint i.e. IP Transit Traffic

Using meta heuristic (GRASP, Ant, …)Using meta heuristic (GRASP, ACO, …)

P t ti f i ti lt

NETW Calculate migration characteristic

Using traffic model, system model, …

Migration doctrine (granularity, oversize)

P t ti f i ti ltPresentation of migration results

Visualization

Migration order (time)

Algorithm economic performance

Import/Export

Interface from Mig.-Alg.

Interface from Greenfield (AMPL)

Presentation of migration results

Visualization

Migration order (time)

Algorithm economic performance

Import/Export

Interface from Mig.-Alg.

Interface from Greenfield (AMPL)


Slide 13

Algorithm controlAlgorithm control

Current migration approach

• Cost efficient insertion of OXCs (Single layer migration)

o Describe start/end backbone System model / yo Calculate demands Traffic modelo Calculate costs Cost model for

CAPEX, OPEX, IMPEXo Calculate realistic reference network ILP o Find most interesting migration nodes Routing, Groomingo Make heuristic approach for optimal costs GRASP, ACOo Budgeting Dept redemptiono Visualize results GUIo Visualize results GUI


Slide 14

Traffic model

• Reference German 17 node backbone network• Demand creation calculated through population density

o Currently: Layer 3 demandsy yo Todo: Layer 2 demands

• Directions via assumptions• Increase 40% per year

• Currently no directional changeschanges


Slide 15

Cost model

• CAPEX cost modelo According to Nobel2-Projecto Cost erosion 20% reduce per device per yearp p y

• IMPEX cost modelo Complicated to model due to:

Travelling costsTravelling costs Extra charges (i.e. holidays) Penalty fee

o Simplify: 20% of (initial) CAPEX of node+10% every yeary y

• OPEX cost modelo Complicated to model due to:

Energy/Cooling costs Security costs Rental costs

o Simplify: 2% of (initial) CAPEX of node


Slide 16

Cost model - IP/MPLS CAPEX

• Every node:• Basic node 640 GBit/s (16 Slots): 16,67 CU• Slot card 40 GBit/s: 9,17 CU/ ,

Filled depending on traffic (4x) Port card 4x STM16, LR (1550nm, 80km): (4x) 6,67 CU

• Example:• Example:• IP Router (max 40 GBit/s throughput)

costs:C[CU]=16,67+9,17+4*6,67=52,52 CU

• 1 CU == costs for 10G WDM Transponder


Slide 17

[No2]

Reference network generation - ILP

• Find appropriate start (and final) networko Give Locationso Give Demands

7 X

80Ch

3 X

40Ch

o Give (possible) Edges• Searched: Dimensions

o For minimal costso For working networko For working networko With given constraints

Problem formulation Mathematical model 1 X

Make AMPL formulation with model and data

Find a relaxation with CPLEX

640

4 X

4 X

Find integer problems with CPLEX

Solution

40Ch

10 X1280


Slide 18

[Hoe]10G

Migration heuristics

• Selective randomized search heuristic (SRSH)o Simple hardware insertion schemeo Follows predefined constraintsp

• Ant colony optimization (ACO)o Device granularities as possible stateso Partial deviation from initial patho Partial deviation from initial patho Best results to elite poolo What is most cost efficient granularity set

• Greedy randomized adaptive search procedure (GRASP)• Greedy randomized adaptive search procedure (GRASP)o Delivers potential migration ordero Construction phaseo Local search phase

( ) h l ko (ev.) Path relinking


Slide 19

Migration heuristics SPP - ACO

Minimize Calculation time (CNF)

Add or replace better results

STEP 4 …

5000 CU4900 CU

Total migration costs

best 2nd

Elite pool

STEP 2

STEP 3

…

…3rd

Actualize pheromone trails

G1 G2 G3 …G1 G2 G3 …(OXC) (IPR)

STEP 1

N1 N2 N3N3

…

50%

p

(O C) ( )

Initial random path

Ant follow path

20%

10%

20%


Slide 20

Grooming – Future Perspective2 * Kdemand < Chcapacity

• Link-by-Link Grooming

demands

IP Layer

Requirements:IP: 6 * Kdemand interface cardsPhy: 1 wavelengthsOptical Layer

• End-to-End Grooming

Requirements:

IP Layer

Requirements:IP: 4 * Kdemand interface cardsPhy: 2 wavelengths + 1 OXCOptical Layer

• Traffic Grooming

Requirements:IP: 4 * Kdemand interface cards

IP Layer


Slide 21

Phy: 1 wavelengths + 1 OXCOptical Layer

Grooming – Future Perspective

Grooming

Cost Energy Impairment TE/QoS Resilience

Energy

Costs

Lambda seta b c d e f g h i

Migration decision


Slide 22

Results


Slide 23

Results – Resource utilization

Component expenses per step:• CAPEX• OPEX

CAPEX per step:• Optical basic node (OPBN)• IP basic node (IPBN)

• IMPEX( )

• IP port card (IPPC)

to be published …


Slide 24

Results – Algorithm performance

Integral CAPEX per node:• Optical basic node (OPBN)• IP basic node (IPBN)

Algorithm result improvement:• ACO• Selective random( )

• IP port card (IPPC)

to be published …


Slide 25

Results – Scientific publications

S. Türk, S. Sulaiman, A. Haidine, R. Lehnert, and T. Michaelis,“Approaches for the migration of optical backbone networks towardscarrier ethernet,” in 3rd IEEE Workshop on Enabling the Future, p gService-Oriented Internet - Towards Socially-Aware Networks (Globecom), Honolulu, Hawaii, USA, 2009.

T. Michaelis, M. Duelli, M. Chamania, B. Lichtinger, F. Rambach and S. Türk,a s, u , C a a a, g , a ba a d S ü ,“Network planning, control and management perspectiveson dynamic networking,” in 35th European Conference on OpticalCommunication (ECOC), Vienna, Austria, 2009, p. 7.7.2.


Slide 26

Summary


Slide 27

Project progress

• Tasks done:o Network topology definition for GER17 and EUR67o Routing: 3 grooming algorithms (SH,MH,CO)o Sorting: IP transit traffico Sorting: GRASP, PR, ev. PRo Hardware insertion doctrine (OXCs and IPRs)

o Randomly selectiveo ACO GRASPo ACO, GRASP

• Future work to be done or under investigation:o Adaption of the GRASP to real network situationso Improvement of current routing strategyp g gy

Implementation of intelligent groomingo Improvement of current hardware insertion

Forecasting (All-Period-Planning) Granularity

L 2 d i Layer-2 deviceso Introduction of Layer-2 demands

Change of traffic modelo …


Slide 28

References

[Big] Cezary, Wojciech : “Optimized Modeling and Design of Multilayer IP over Optical Transport Network Architectures”

[Ble] Bley, Andreas: “Multi layer network design a model based optimization approach”[Gli] www.glimmerglass.com: “Intelligent Optical Switches”[Hoe] Höller, Holger: „Schichtenübergreifende Planung von Hochgeschwindigkeits-Telekommunikationsnetzen“[Hon] Hongyue, Zhu : “Cost-Effective WDM Backbone Network Design with OXCs of Different Bandwidth

Granularities”[Kam] Kaman, V.: “Comparison of Wavelength-Selective Cross-Connect Architectures for Reconfigurable All-

Optical Networks”[Meu] Meusburger, Clara; Schupke, Dominic A.: “Opex Considerations for Optical Switch Upgrades: A Multiperiod[Meu] Meusburger, Clara; Schupke, Dominic A.: Opex Considerations for Optical Switch Upgrades: A Multiperiod

Point of View”[No2] Nobel 2 Project: “Migration Guidelines with Economic Assessment and New Business Opportunities

Generated by NOBEL phase 2”[Ver] Verbrugge, Sofie: “Strategic Planning of Optical Telecommunication Networks in a Dynamic and Uncertain

Environment”[Ve2] Verbrugge Sofie: “Capex comparision between link-by-link and end-to-end grooming in a European [Ve2] Verbrugge, Sofie: Capex comparision between link by link and end to end grooming in a European

backbone network”[Wes] Wessäly, Roland: “Two-layer Network Design by Branch-and-Cut featuring MIP-based Heuristics ZIB-

Report”

Student work:[G ] G H C i d i li i f k i i “[Gan] Gang, He: „Conception and visualization of network migration processes“[Hin] Hinz, Herwig: “Optimization of a network migration process”[Lar] Larrañaga, Javier: “Investigation of traffic grooming in optical backbone networks”[Loz] Lozano, Paloma: „Design of IP over WDM planning algorithms using ILP“


Slide 29

Thanks !!Thanks !!


Slide 30

Migration Planning for Optical Backbone Networks · • Usually SDH, combined with MPLS • DWDM...

Documents

Transcript of Migration Planning for Optical Backbone Networks · • Usually SDH, combined with MPLS • DWDM...