Evaluation of a Centralized Solution Method for One … · Evaluation of a Centralized Solution Method for One-Step ... free transition between ... resolvable by improved heuristic

Universität Stuttgart

Institute of Communication Networks

and Computer Engineering (IKR)

Prof. Dr.-Ing. Andreas Kirstädter

Evaluation of a

Centralized Solution Method for One-Step

Multi-Layer Network Reconfiguration

TIWDC 2013, Genova, Italy

Frank Feller

[email protected]

2013-09-23

2© 2013 Universität Stuttgart • IKR F. Feller – One-Step Multi-Layer Network Reconfiguration Method

Outline

Motivation

Load-Dependent Core Network Operation

• resource adaptation assumptions

• multi-layer network reconfiguration

Periodic One-Step Network Reconfiguration

• constraints and resulting reconfiguration procedure

• optimization problem

• optimization-heuristic solution method

Evaluation

Conclusion


Motivation: Trends in Transport Networks

Traffic Evolution

exponential growth of traffic volume significant diurnal traffic variations

Access Technology Evolution

energy-efficient optical access technologies

→ power consumption in the core

gains importance

→ energy savings in the core by dynamic resource operation desired

cf. e

g.:

DE

-CIX

tra

ffic

sta

tistics

2009 2010 2011 2012 2013

traffic

vo

lum

e

traffic

vo

lum

e

00:00 12:00 00:00 06:0018:0006:0018:00

cf. e

.g.:

Hin

ton

, B

alig

a, F

eng,

Ayre

, T

ucke

r,

“Pow

er

con

sum

ption a

nd

energ

y

effic

ien

cy in the

Inte

rnet,”

IEE

E

Ne

two

rk,

vol. 2

5, 20

11.

pow

er

con

sum

ption

access

rate100Mbps1Mbps 10Mbps

optical access

core

total


Load-Dependent Core Network Resource Operation

Scenario: Multilayer Network (e.g. IP/MPLS over WSON)

Dynamic Resource Operation

• activation / deactivation of optical circuits

– along with line cards and transponders consuming largest share of energy

– switching times in the order of minutes due to interaction with fibre amplifiers

• power scaling in packet processors

enabled by sleep modes for parallel structures and frequency scaling

→ energy consumption scales closely with traffic load

→ energy savings by adapting network configuration to load

fibre amplifier

line card

port / transponder

packet-processing node

optical switch

electronically switched

transit traffic

optical

fibre


Multi-Layer Network Reconfiguration

Configuration Dimensions

Upper layer

• virtual topology

– realized by optical circuits

– independent of physical topology

• demand routing in virtual topology

Lower layer

• routing of optical circuits

• wavelength assignment

Reconfiguration Objective

Energy consumption – defined by upper layer

→ focus on upper layer



• time-triggered computation of low-energy network configuration for forthcoming traffic

time

tear-down

set-up

∆T ∆T

config A

config B

config C

compute config C compute config Dcompute config B




• prediction of peak traffic demands for future periods assumed to be available

config A

config B

config C

time

compute config C compute config D

period C peak trafficforecast

∆T ∆T

compute config B






• interruption-free transition between configurations (→ make-before-break)

amplifier transients require slow circuit setup and teardown

set up

circuits for

config B

tear down

circuits of

config A

set up

circuits for

config C

tear down

circuits of

config B

set up

circuits for

config D

timereroute to

config B

reroute to

config C



tear-down

set-upconfig A

config B

config C







traffic forecast horizon is limited (presumably) → one-step reconfig (no transition path)

set up

circuits for

config B

tear down

circuits of

config A

set up

circuits for

config C

tear down

circuits of

config B

set up

circuits for

config D

timereroute to

config B

reroute to

config C



tear-down

set-upconfig A

config B

config C








→ new circuits may only use resources not occupied in the previous configuration

tear-down

set-up

set up

circuits for

config B

tear down

circuits of

config A

set up

circuits for

config C

tear down

circuits of

config B

set up

circuits for

config D

timereroute to

config B

reroute to

config C



config A

config B

config C








→ new circuits may only use resources not occupied in the previous configuration

→ specific optimization problem – to be solved in less than the reconfiguration interval

tear-down

set-up

set up

circuits for

config B

tear down

circuits of

config A

set up

circuits for

config C

tear down

circuits of

config B

set up

circuits for

config D

timereroute to

config B

reroute to

config C



config A

config B

config C


One-Step Reconfiguration Problem

Finding a Network Configuration in terms of

• set of optical circuits including the resources they occupy

• routing of demands in resulting virtual topology

Constraints

• traffic demands (between all node pairs) to be satisfied

• installed resources (line card ports and fibre capacity)

• resource occupation in previous configuration

Objective Cost Function

Simultaneously minimize cf =

• energy consumption = α × # active optical circuits

+ β × electronically switched transit traffic volume

• changes to configuration + δ × # newly established or torn-down circuits

• traffic blocking + µ × # virtual links with insufficient capacity

+ ν × blocked traffic volume


Virtual Topology Centric Reconfiguration (VTCR)

Simulated-Annealing Based Solution Method

Heuristic optimization: randomized search of solution space

Optimization Procedure (Loop)

• perturbation of virtual topology

randomly add or remove one virtual link

• cost computation

– deterministically route demands on shortest path in virtual topology

– determine required circuits from traffic on each virtual link

• set up according circuits if feasible

• count blocking if insufficient resources

– evaluate cost function

Post-Processing

Drawbacks of deterministic demand routing

1. blocking on shortest-path links while spare resources on alternative paths available

2. lowly utilized circuits while traffic could be accommodated by existing circuits on other path

→ greedy traffic rerouting heuristic to resolve such situations

solution space

initial

solution

accepted

solution

rejected

solution

best

found

solution

perturbations


Evaluation by Simulation

Scenario

• Géant reference network topology

from SNDLib (http://sndlib.zib.de)

22 nodes, 36 links, 462 traffic demands

• 14 days out of measurement-based

demand trace; scaled to vary average load

• reconfiguration every 15 minutes

• network resources dimensioned for peak of all demands

Baseline Case

resource scaling (RS)

– fixed virtual topology and fixed traffic routes (optimized for peak demands)

– load-dependent resource operation

Cost Parameters

• energy – per circuit: α = 1; per circuit equivalent of switched traffic: β = 4.3 · 10-5

• circuit modification: δ ∈ {0; 0.43}

• traffic blocking: µ = ν = 17


Evaluation Results

Energy Consumption

• VTCR reduces energy consumption by 25% to 35% over resource scaling

0 0.5 1 1.5 2average peak demand [circuit equivalents]

0

50

100

150

200

250

300

350

400

450

avera

ge n

orm

aliz

ed p

ow

er

consum

ption

always on

RS

VTCR

δ = 0.43

circuit modif. time

0∆T


Evaluation Results

Energy Consumption Configuration Changes


• positive reconfiguration penalty δ significantly reduces circuit changes


0

50

100

150

200

250

300

350

400

450

avera

ge n

orm

aliz

ed p

ow

er

consum

ption

always on

RS

VTCR

δ = 0.43

circuit modif. time

0∆T

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2average peak demand [circuit equivalents]

0

10

20

30

40

50

60

70

80

90

100

avg. num

ber

of circuit m

odif. per

inte

rval

RS

VTCRδ = 0.43

VTCRδ = 0


Evaluation Results




→ increased energy efficiency if consumption of transient circuits considered


0

50

100

150

200

250

300

350

400

450

avera

ge n

orm

aliz

ed p

ow

er

consum

ption

always on

RS

VTCR

δ = 0

δ = 0.43

circuit modif. time

0∆T


0

10

20

30

40

50

60

70

80

90

100

avg. num

ber

of circuit m

odif. per

inte

rval

RS

VTCRδ = 0.43

VTCRδ = 0


Evaluation Results




→ increased energy efficiency if consumption of transient circuits considered

• traffic blocking in 8 / 21,504 settings due to sparse resources or suboptimal solutions


0

50

100

150

200

250

300

350

400

450

avera

ge n

orm

aliz

ed p

ow

er

consum

ption

always on

RS

VTCR

δ = 0

δ = 0.43

circuit modif. time

0∆T


0

10

20

30

40

50

60

70

80

90

100

avg. num

ber

of circuit m

odif. per

inte

rval

RS

VTCRδ = 0.43

VTCRδ = 0


Conclusion

Periodic One-Step Multi-Layer Network Reconfiguration

• reconfiguration procedure implied by technological and operational time constraints

→ optimization problem with resource pre-occupation constraints

• VTCR: an optimization-heuristic solution method for this problem

– optimization of virtual topology

– deterministic demand routing (shortest path & greedy heuristic)

• evaluation results

– VTCR reduces load-dependent energy consumption by 25% to 35% compared to

dynamic resource operation with static virtual topology and fixed traffic routing

– reconfiguration penalty significantly reduces number of circuits established and torn down

– traffic blocking is rare and resolvable by improved heuristic in practically relevant cases

Future Work

• accounting for resource hierarchy for power consumption (port – line card – rack)

• evaluation in further scenarios (e.g. different network size, resource dimensioning)

• MILP formulation and exact reference solution for small problem instances

• comparison with other network reconfiguration schemes


References

• K. Hinton, J. Baliga, M. Feng, R. Ayre, R. S.Tucker, “Power consumption and energy

efficiency in the Internet,” IEEE Network, vol. 25, 2011.

• A. Bononi, L. Rusch, “Doped-fiber amplifier dynamics: a system perspective,” Journal

of Lightwave Technology, vol. 16, 1998.

• P. N. Tran, U. Killat, “Dynamic reconfiguration of logical topology for WDM networks

under traffic changes,” IEEE Network Operations and Management Symposium

(NOMS), 2008.

• F. Idzikowski, S. Orlowski, C. Raack, H. Woesner, A. Wolisz, “Dynamic routing at

different layers in IP-over-WDM networks – maximizing energy savings,” Optical

Switching and Networking, vol. 8, no. 3, 2011.

• E. Bonetto, L. Chiaraviglio, F. Idzikowski, E. L. Rouzic, “Algorithms for the multi-period

power-aware logical topology design with reconfiguration costs,” Journal on Optical

Communications Netw., vol. 5, no. 5, May 2013.

• S. Orlowski, M. Pióro, A. Tomaszewski, R. Wessäly, “SNDlib 1.0 – Survivable Network

Design Library,” Networks, vol. 55, no. 3, 2010, http://sndlib.zib.de.

• S. Uhlig, B. Quoitin, J. Lepropre, S. Balon, “Providing public intradomain traffic matrices

to the research community,” SIGCOMM Computer Communication Review, vol. 36,

Jan. 2006.

Documents

Evaluation of a Centralized Solution Method for One … · Evaluation of a Centralized Solution Method for One-Step ... free transition between ... resolvable by improved heuristic