green wired networks - inria

HAL Id: hal-01196819https://hal.inria.fr/hal-01196819

Submitted on 12 Sep 2019

HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, estdestinée au dépôt et à la diffusion de documentsscientifiques de niveau recherche, publiés ou non,émanant des établissements d’enseignement et derecherche français ou étrangers, des laboratoirespublics ou privés.

Green Wired NetworksAlfonso Gazo Cervero, Michele Chincoli, Lars Dittmann, Andreas Fischer,

Alberto E. Garcia, Jaime Galán-Jiménez, Laurent Lefèvre, Hermann de Meer,Thierry Monteil, Paolo Monti, et al.

To cite this version:Alfonso Gazo Cervero, Michele Chincoli, Lars Dittmann, Andreas Fischer, Alberto E. Garcia, et al..Green Wired Networks. Large-Scale Distributed Systems and Energy Efficiency, Wiley, pp.41-80,2015, 9781118864630. �10.1002/9781118981122.ch3�. �hal-01196819�

https://hal.inria.fr/hal-01196819

https://hal.archives-ouvertes.fr

Chapter 3

GreenWiredNetworks

ChapterContributors:AlfonsoGazoCervero,MicheleChincoli,LarsDittmann,AndreasFischer,AlbertoE.Garcia,JaimeGalán-Jiménez,LaurentLefevre,HermanndeMeer,

ThierryMonteil,PaoloMonti,Anne-CecileOrgerie,Louis-FrancoisPau,ChrisPhillips,SergioRicciardi,RemiSharrock,PatriciaStolf,TuanTrinh,LucaValcarenghi

After highlighting the significant energy consumption of existing wiredcommunication networks, this chapter considers various means of operating suchnetworksmoreefficiently.Thechapterexaminesthecomponentsthatmakeupwiredcommunicationsnetworkandtheirdifferingcharacteristicsbetweentheaccessandcore,aswellaspatternsoftrafficbehaviour.Oncethis isdone,thechapterfocusesonstatic (networkplanning)anddynamic (trafficengineering) schemes thatcanbeusedtoreducetheenergyconsumptionofnetworks.Thechapteralsopaysattentiontoanumberofchallenges/openresearchquestionsthatneedtoberesolvedpriortothe implementation of such schemes. These include issues with migration andresilience.Finallyasummaryreviewsthekeythemesthathavebeencovered.We use the term “wired” to represent any network where the communicationchannelbetweennodesisconfinedtoaspecificconduitratherthanrelyingonfree-spaceradiationofthesignal.Thisincludessystemsbasedonmetalconductorsaswellasglassfibre,withandwithoutin-lineamplification.Converselywereservetheterm“wireless”communicationforsystemsinvolvingradioaccess.ImprovingtheefficiencyoftheselattersystemsisaddressedinChapter4.Theaimistomakewirednetworksoperate inanenergy-efficientwaysinceenergyexpenditure has become amajor concern with the continued development of theInternet[c1].Italsohasconsiderableimpactontheenvironment[c2].Inthelastfewyears, energy consumption issues have received much attention from thegovernment, general public and telecommunication operators [c3][c4]. Firstly,withthesustainedgrowthofthecustomerpopulationandgreateruseofhigh-bandwidthservices such as streaming video, Internet Service Providers (ISPs) and telecomcompanies have needed to increase network capacity and expand their reach tosupportthesedemands.Associatedwiththeuseofmorepowerfulandpower-hungrynetworkequipment,theoperationalexpenditure,particularlyintermsofthecostofelectricity,isasignificantfactorparticularlyduetotheincreasingenergyprices[c5].Increasing energy usage is also an environmental problem particularly due to therelease of carbon dioxide (CO2) into the atmosphere. It is also a political one,especially as politicians come to terms with the consequences of the anticipatedenvironmentalchanges.

The power consumption of Information and Communication Technologies (ICT)includes the power consumption of servers, network infrastructure and end-userdevices.Itaccountsforabout2%oftheworld’spowerconsumptionrisingto10%fordevelopedcountries like theUK [c6]. It isanticipated that thiswill increasenotablyoverthenextfewyears.Forexample,non-domesticenergyconsumptionisexpectedtogrowbyafurther40%by2020[c7].Althoughitseemsasmallfractionofoverallpower consumption, the absolute value of 900 Billion kWh/year, is equal to thepowerconsumptionofCentralandSouthAmericatogether[c2].Withinthefieldoftelecommunicationsresearch,energymanagementcanbedividedintotwosegments:wirelessnetworksandwired(fixed)networks.Sincetheoperationof (mobileor unattended)user equipment inmanywireless networks is limitedbybattery power, considerable effort has been devoted to devising new energymanagement schemes for this sector. For example,many energy-efficient schemeshavebeenproposedforwirelessadhocsensornetworks[c8][c9].Thereisalsomuchongoingresearchintoenergysavingwithincellularnetworkinfrastructure,includingtheuseoflowenergy(andlimitedfunctionality)basestations,coding&multiplexing,and exploiting residential wired access points [c10]. Conversely, it is generallyassumed that wired networks, including fixed user equipment, have access toabundant power, implying there is no need to save energy. Consequently, wirednetworks energy consumption is inefficient and, moreover, there is considerableroom for improvement [c11]. Furthermore, technological advances, particularly inrespecttoDense-WaveDivisionMultiplexing(DWDM),meanthatthebandwidthofphysicaldevicesisnolongerarestrictiononthecapacityoftheInternet;insteaditislimitationsintheachievableenergydensity[c1].Commonly the wired network is considered to be composed of several parts: theaccessnetwork,themetro(politan)networkandthecarrier(core)network.Fromthepointofviewofpowerconsumption,research indicatesthat inthenearfuture,thecorenetwork’spowerconsumptionmaybecomethemostsignificantelementoftheoverall communicationsnetwork infrastructure [c12]. Therefore a lot of research isnow focused on the regional Tier-2 IP core networks employing various DWDMarchitectures.TheremainderofthechaptercommencesinSection3.1withanassessmentofsocio-economicmechanismsthatcanbeusedtopromotemoreenergy-efficientuseofthewirednetworkinfrastructure.Thisisfollowedbyabriefreviewofprinciplenetworkcomponents and typical wired-network architectures in Sections 3.2 and 3.3,respectively.Section3.4thengivesconsiderationtohowtrafficisevolvingbothduetotheincreasingnumberofusersonwirednetworksandalsoasaresultofadditionalservices being deployed across the Internet. Understanding these changing trafficpatterns isessential ifnetworksare tobeoperatedefficiently.Then,weconsideranumberofenergy-savingmechanisms.TheseareseparatedintostaticmechanismsinSection 3.5.1 and dynamic ones in Section 3.5.2. Static mechanisms or networkplanning[c13][c14],considertheenergyconsumptionwhileplanningandconfiguring

the network. Dynamic, or traffic engineeringmethods,make use of a prearrangedinfrastructure in an energy-efficiencyway. These approaches include infrastructuresleeping, rate-adaptation and network virtualization. Infrastructure sleeping letsrouters and links be switched off or enter a low power consumption state whennetwork traffic demand is light [c15]; rate-adaptation [c16] adjusts the powerconsumption according to the required transmission rate of the traffic. Finally,networkvirtualization[c17],ornetworkmachinemigration[c18][c19],allowsseveralnetwork component instances to run on the same physical platform whenperformance requirements permit the saving of energy through consolidation.Despitethesevariousschemes,significantchallengesremain.TheseareconsideredinSection3.6.Finally,Section3.7providesabriefsummaryofthekeyfindingsandtheissuesthatstillneedtobeaddressed.3.1 Economic Incentives and Green Tariffing WhereaslatersectionsinChapter3considerdesignandtrafficmeasurestoachieveenergy savings in communications networks, the present section deals witheconomic, regulatory and business related methods. It is estimated that suchapproaches,ifadoptedbyusers,couldachieveaverylargeimpact.3.1.1 Regulatory, Economic and Microeconomic Measures Various measures are available to influence energy usage. We consider severalsignificantexamples.3.1.1.1 Regulatory Dimensions Communications carriers are universally subject to operating licenses, specified,awardedand controlledby communications regulators [c20]. Electricity andenergyproductionisalso,insomecountries,subjecttoregulations,specified,awardedandcontrolledbyenergyregulators[c21].Policiesbybothcategoriesofregulatorsmustabidetonationalandinternationallawsortreatiesdealingwithclimatechange[c22];however,atthisstage,nocommunicationsregulatorsareknowntohaveincludedyetany mandatory requirements with “green” objectives in their license awardingschemes. Energy reduction objectives are therefore “de facto” implemented bycommunications equipment manufacturers, to help their clients, the operators,achieve operating cost reductions as well as “green objectives” of their own (e.g.:environmentalreportstoshareholders, regionalpolicies,commitmentstowardsuseofspecificenergysources).Therefore,ineffect,regulatorymeasuresaretodaydrivenmuchmorebythecommunicationsequipmentmanufacturersandoperatorsthanbynational regulators. Industry consortia such as the GSM Association (GSMA) haveintroduced best practices for energy savings leading later to an InternationalTelecommunicationUnion(ITU)standard[c23].

This is however bound to change, as some communications regulators plan tointroducemandatory requirements, especially on ISPs and on public wireless basestationoperators,wherebytechnologymigrationsetinthelicenses,mustbelinkedtoa specifiedpower consumption reduction for a given traffic capacity. Likewise, in anumberofcountries,somecommunicationsoperatorsareseekingenergyproductionand distribution licences, to be allowed to own and operate energy productionfacilitiesservingnotonlytheirownneeds,butalsothirdparties.Onemotivationistolimit the electricity and cooling prices charged to communications operators bypower utilities. Another is to capitalize on the preferential amortization rates forselectedgreenenergyproductionfacilities.Athirdmotivation is toearnsometimesincentive-led income from selling excess renewable energy to third parties. Finally,somecommunicationsoperatorsareinacompetitivestrategicpositioninggamewithpowerutilitiesaroundsmartgridmeteringandservices[c24].Theneteffect is thatcommunications operators interactmuchmorewith energy regulators, an offer totheseadiversificationandincreasedcompetitionamongstmarketplayers.3.1.1.2 Economic Dimensions Traditionally ineconomic terms, telecommunicationswasaboutuser-led trafficandservicedemands,besidesmanufacturingoutput(partofthesecondeconomicsector)[c25].Theuser-leddemandasstudiedineconomicsdidnotincludecontentcreation,allthisbeingaggregatedintoutilityservices,andthusthethirdsectoroftheeconomy(services).Also,inmacroeconomicstherearenospecificmetricsforsustainability.Asaresulttheenergyconsumptiondrivenbycommunicationstrafficandservices,isstilltreatedasapuresubstitutionbetweenutilityservicesinsidethethirdsector,andnotrenderedexogenous.Theconsequenceisthat,atamacroeconomiclevel,thereisnoincentiveorreadymeanstoestimatethecross-effectsbetweencommunicationsservicevalueandenergypricing.Anotherconsequenceisthattherelativereturnsoninvestmentsincommunicationsystemsandservices,andinenergysavingscapacity,cannot be assessed. This is bound to evolve, if indeed the knowledge andcommunicationseconomywillbetreatedasthefourthsectoronitsown[c25].3.1.1.3 Microeconomic and Pricing Measures The strict definition of rationality is on an individual’s preference: the preferencerelation is rational if it possesses the properties of completeness and transitivity[c26]. It means the individual is able to compare all the alternatives and thecomparisonsare consistent. Furthermore, rationality implies that the individualhascomplete informationof all alternatives andknowsabout the consequencesof thechoice;theuseralsohasunlimitedtimeandunlimitedcomputationalpowertopickthemost preferred option. In reality, such a perfectly rational user does not exist.Over the past decades, a largemass of empirical data has shown violations of therationalityassumption[c27].

H. Simon [c28] has pointed out that the behaviour of most individuals is partialrational.Muchofthetime,theindividualdoesnotknowallthealternatives.Neitherdoes s/he have perfect information regarding the consequences of choosing aparticularalternativesbothbecauseoflimitedcomputationalpowerandbecausetheuncertainty in the external world. The individual’s preferences do not possess therational prosperities when comparing heterogeneous alternatives. Simoncharacterized this as “bounded rationality”. Model construction under boundedrationalityassumptioncantaketwoapproaches.Firstistoretainoptimization,buttosimplifysufficientlysotheoptimumiscomputable.Secondistoconstructsatisfyingmodel that provides decisions good enough, with reasonable computational cost.Neitherapproachdominatestheother[c28].Following thepioneeringworkof Simon,KahnemanandTversky [c29] conductedaseries of research on various types of judgement about uncertain events. Theirconclusionwas thatpeople relyona limitednumberofheuristicsprinciples,whichreducethecomplextasksofassessingprobabilitiesandpredictingvaluestosimplerjudgementoperations.Arecentlyrevisitofthisanalysis[c30]proposedaformulationin which the reduction of complexity is achieved by an operation of “attributesubstitution”wherebyajudgmentissaidtobemediatedbyaheuristic.Costs, benefits and incentives linked to the adoption of energy savings incommunications networks, represent a typical attribute substitution, and thereforedominantly hinge onmicroeconomic measures at supplier level as well as at userlevel.Thenewframeworktobeconsideredhere isonewherethecommunicationssuppliersandusersalike,arenot justdealingwith trafficandservices,butarealsoand at the same time suppliers andusers of energy,with trade-offs between theirtworoles[c31].Ultimately,asseenabove,thecommunicationssuppliermayalsobeanenergysupplier,andvice-versa,sothatatafundamentallevelthemicroeconomicpricing and revenue are on combined “bundles” of communications and electricitywith“attributesubstitution”.Byacombinationofalargenumberofusers,technologyimprovements,andoperatorproductivity gains, the pure transport and access tariffs for communications haveplummetedtolowvalues,whilstenergycostsmaysoonexceedthem.Content-basedservices will generate certain revenue. Thus the very mix inside the combined“bundles” will change, with energy representing a large component, which onlyenergysavingscanreduce.3.1.2 Pricing Theory in Relation to Green Policies Inthissectionwenowreviewpricingprinciples,introducetheconceptofenergy-and-communicationsusebundlesandthen“green”tariffsasincentives.3.1.2.1 Pricing Principles

Theeconomictheoryofpricinghastraditionallybeenderivedforphysicalgoods,andfromdifferentangles:1. Either as a static equilibrium between supply and demand, including auctions

[c32];2. Ortakinginconsiderationrankedpreferencesforindividualpriceformation;3. Or reflecting price dynamics with endogenous fluctuations due to market

restructuring.Most conventional telecommunication pricing schemes have been variants of theabove,assuminglimitedcapacityineitherbandwidthortransmissioncapacity;thesetwo assumptions have beenmade largely erroneous with the advent first of fibreoptic transmission, next of advanced radio coding/modulation/spectrum usagetechniques, and finally of IP basednetworks [c25].More recently, operators,whilelargely regulated according to the above principles, have determined prices bysummingmarketing,customercare,andfinancialcosts,thusblurringthepictureforusersaswellas for themselves.Customersegmentationapproachesnowdominatethe pricing and revenue strategies for communications services. Likewise, mostelectricity pricing schemes have also been variants of the above, assuming limitedenergy production/generation capacity, and/or daily variation patterns of demand[c33].However,atthedawnoftelephonyservices(circa1880),therewereindividualtariffs[c34]! For quite some years, and in somemarkets, telephony servedmainly somehigh-level civil servants and privileged people (trade, news), and was a symbol ofwealthandsocialrank.Telephonesubscriberswerenotdesignatedbyanumber,butby theirname;pickingup thephonewouldget theoperator,whowould then ringandconnectthedesiredpartybyapolitesupportstaffprotocol.Asaluxuryserviceatthattime,thedemandfortelephonywaslimitedaswasnetworkcapacity.Individualdemands were price insensitive, which resulted in an inelastic demand curve asshown inFigure3.1.Thesupplywasscarceand inelastic toprices,as therewasnochoice,norcompetitivemechanisms:theaveragepricepercallwashighandanywaythe supply stalled because of limited exchange capacity and human operations.Althoughpriceswerefixedindividuallyorbundledintopackageswithfixednumbersofcalls,theywereextremelyhigh,whichresultedintheexclusionofthemajorityofpopulationfromaccessingtheservicesduetounaffordability.

Pricepercall

Averagepricepercall

Numberofvoicecalls0

A

B

C

Figure3.1:Servicedemandsatthedawnoftelephony

In Figure3.1A,BandC represent individualdemands,whichare limitedandpriceinsensitive.Theaveragetelephonydemandcurveattheintroductionoftelephonyisshownastheboldcurve,which is inelastic.Allvoicecallshereareassumedforthesamedestination/distance.Fromthebeginning,althoughsetsoftelephoneusersandsupplierswererestricted,charging patterns began to diverge between flat rates and individual usage-basedrates. The number of call attempts, the physical destination, and the call durationwereallmanuallyrecorded.However,thepricingofthecallswasamatterbetweenthe telephone company sales person and the customer (who was then not asubscriber); usage, rank, fame, location were all taken into account, and thesettlementwasaccomplishedbyabanknoteorcash.Thishistoricalrecollectionisveryrelevantnowadaysinrelationtogreenpoliciesduetoanumberof fundamental reasons. First, theoperatorsand theirusers that careabout energy savings are initially minorities, also called “early adopters”, and theprice settlement thus can/should be individualized. Next, as explained in Section3.1.1,theregulatoryvacuumregardinggreenpolicyimplementationsleavespaceforindividualnegotiations rather than fixed tariffbundles.One simple reason is thatagivenusermayhavea lowenergyprofileandahighcommunicationsprofile,whileanothermightshowthereverse,thusimpactingthecosts,benefitsandincentivesatan individual level. Likewise, a communications supplier may have from itsinfrastructureandserviceofferings,relativelyhighenergyconsumption,butaccesstolowutilitysupplierprices,whileanothermightshowthereversepicture.Thisleadstotheconceptof“individualtariffs”foracommunicationsandenergyconsumer[c34].An“individual tariff”assummarized inTable3.1meansthateach individual setsatariff for himself/herself for a specific set of communications and energy servicesprovided by a supplier or a community, whether this service is user-defined orcommunity-defined. Even if that individual belongs, say to an enterprise, themembersoftheenterprisemayhavedifferentindividualtariffs,simplybecausetheirservice(contributionsandreceipts)aredifferent.Even,differentusersofanidentical

service (e.g. voice)mayvalueandprice itdifferentlyas theydecide individually tochoose another supplier, or belong to different communities. Furthermore, asupplierorcommunityneednotownpartortheentiretransmissioninfrastructure,sourcedcompetitivelyfrominfrastructureowners,especiallyasthisisencouragedbyregulatorsunderthe“virtualoperator”concept.Indeed,thisdefinitionomitsserviceprovisioningduration,asthedurationoftheservicewillbejustoneattributeinthemulti-attributeservicedemand;e.g.sporadicusesarepossible,justasarelongtermones,butthedifferencewithtodayisthatservicedurationbecomesuser-defined.

! ConsideroneBuyer(B)andNSellers(S(i),i=1,..,N).TheBuyersignalshismaximumpricepandcontractdurationT

! BuyersignalshisservicemixandattributesMix=(A(j),q(j),j=1,..,M)where“A”denotesaserviceand“q”thequantityorQoS

! OneorseveralSellersrespond! Thereisanindividualtariff,wheneverBandoneSellerS(*)agreeby

contractontheprovisioningbyS(*)toBoftheserviceMixfordurationTatpriceequalorlowertop

! Thiscontractmaybedifferentfromthecontractforthesameservicemixforotherbuyersorsellersrespectively.

Table3.1:IndividualCommunicationsandEnergyBundleTariffMechanism

3.1.2.2 Mass Customized Communication and Energy-use Bundles Apertinentquestionis“howdoearlyadoptersevolveovertime?”Twentyyearsfromnow combined advanced communication technologies and smart grid technologieswill probably enable people to stay connected anytime, anywherewith access andenergynetworkalternatives.Userswillseamlesslyroambetweennetworksusingyetunknown new services, and behave as joint producers and consumers ofcommunications and energy. 100% penetration of access devices over the entirepopulationinmostcountrieswillrenderworkmoreflexible,andindividualizationoflabour,willproduceahighlysegmentedsocialstructure[c35].Thetotaldemandwillbelarge,aspeoplecanpersonalizetheircommunicationsandenergyservicesaccordingtotheirneeds.Furthermore,communityproliferationmaymultiply the effect. However, due to the highly segmented structure, mosttraditional flat rate tariffswill either not be transparent and appealing enough tousers,oriftheyaremadetoodiverse,userswill“churn”(changesuppliers)implyinghighswitchingcosts.Theseeffectswillallowflatratesforalltobereplacedbymasscustomized individual tariffs, provided each user and community has suitablemanagement and monitoring tools for both communications and energy use,allowing in real-time (as well as over contract durations) to quantify equilibriumbreak-even tariffs [c36]. This results in price insensitive individual demands.Although foreachspecificservice, thepreference fromeachuserwillbedifferent,the average demand across many users again leads to an inelastic price demand

curve as illustrated in Figure 3.2. Consequently, we appear to repeat history (seeFigure 3.1). However, on closer inspection the new demand curve ismore to therightintermsofthe“numberofservicerequests”andmuchlowerintermsofprice.

Priceofaspecificservice“S”

Averagepriceperservicerequestof“S”

Servicerequests

A

B

C

Figure3.2:ServiceDemandforaMass-customizedIndividualServiceBundle“S”

In Figure 3.2, the mass-customized individual service bundle “S” comprisescommunicationsandenergy.A,BandCrepresentindividualA’sserviceprofile:(S,A1,A2, etc.), individual B’s service profile: (S, B1, B2, etc.) and individual C’s serviceprofile: (S, C1, C2, etc.). The average demand for service “S” is shown by the boldcurve.Notethereisa“serviceproliferationeffect”drivingthecurveslope.Already today technical means exist to achieve individual tariffs, and theinfrastructureindustrytogetherwiththebillingplatformindustries,haveshownthatcomplexity and costs involved in individual user multi-service and multiple accessdemandmonitoringandcontrolaresurmountable. In the future, trafficandserviceaggregation,andfilteringcoupledwithservicecreationplatformswillallowthistobemorecapableandcheaperbothforcommunicationsandsmart-gridenergyservices.Althoughtechnicaldetailsoftheeventual implementationarenotstandardizedyet,we can identify possible architectural elements of such a solution using standards[c36],asfollows:

! The linking of user communications and energy service profiles to theadaptation functionality in thebilling/ratingsystem,wouldbecarriedoutatuserauthenticationtimeintheAAAserver;

! In accessnetworks, via theenergygridmeters, routers could recognizeandtagspecialpackets,andprocess themaccordingtoagivencode;within thistagmayresidea label representing the individual tariffandwhichwouldbetiedtotheaccessnodeatwhichsuchtariffsareapplied;

! Theflowlabel intheIPv6protocolusedtocarryQoSfeatures,couldencodeindividual tariff information for this packet. This approach offers theadvantagethatthepackettowhichthisindividualtariffcodeappliesdoesnothave to originate in some service node, but only in those who have thedecodingkeytothisfield.

! Using the SIP protocol, a P-Charging-Vector header is defined to conveycharging related information. The information inside the vector can bepreparedandretrievedbymultiplenetworkentitiesduringtheestablishmentof a dialogue or standalone transaction. In this way, charging relatedprocessingcanbesession-based.

3.1.2.3 Green Tariffs as Incentives Wedefineherea“greentariff”[c31],asthecontractualpricepaidbyasubscribertoan operator for a set of bundled communications and energy services at a givenqualityof service for a specifiedduration; thispricemust reflectbothparties’ besteffortstoreduceCO2emissions,andpassontothesubscriberashareoftheenergyandemissionssavingsfromthenetwork.Theoperatorcaneasilyoffer“greentariffs”by having a green tariff subscriber base segment, which, against committing tocontractforsuchbundles,allowstheoperatortoearmarkaspecifiedpercentageoftheir energy consumption at a network level to technologieswith fewer emissionsandpossiblytotheuseofrenewableenergysupplies.Furthermore, if such green tariffs are made individual (see Section 3.1.2), theyprovide an incentive for users to specify and adopt a communications and energyservicedemandprofileleadingtolesswasteorunutilizedcapacity.Asthis incentivehas monetary consequences, there is a strong likelihood that individualized greentariffswouldbecomevery important.Theirdeploymenthoweverrestsonachievingequilibrium (called a sustainable individual green tariff) between users requestingindividual green tariffs, and theoperators’ economic sustainability, as theyhave tomake new investments in communications technology and/or renewable powersourcestheycontrol.Usingarealoperatormodel, ithasbeenshown[c36], thatbyapplyingrecursiveStackelbergequilibriumcomputationstomulti-attributeindividualtariff contractscovering individualizedservices, thecommunicationsoperatorcouldstillachieveeconomicsustainability.This isonamass-customizedbasis,andapplieseven for complex services and extreme cases. In addition, the analysis shows thatoperator profits are not eroded for generic communications services; however, forvalue-addedcommunicationsservices,thesituationislessclearasinvestmentisalsorequiredincontent,storageandcompetence.3.1.3 COST Action Results ResearchcarriedoutduringtheCOSTActionIC0804focussedonthegreentariffsandindividualcommunications-and-energybundles,includingwirelessaccess[c37][c38].Theoperatorwasassumedtohaveanenergyproductionanddistributionlicensewithalternativerenewableenergysources(wind,solar),whereexcesssupplygotsold.Themainresultistoconfirmthepossibletariffreductiontotheuser,whilemaintainingoperator’sprofitmargin;thedifferencebetweentheindividualgreentariffandthepublictariff(forthesameservicebundle),whichisanindicatorofthenetincentivetotheuser,isintheorder10-20%[c25].Otherresultsonspecific

aspectsofthesameissuearereportedin[c37],[c38].Nevertheless,theoverridingissueistheright-sizingofthemutualdependence,andpossiblecombination,ofcommunicationsoperators,electricpowerutilities,andtheircustomers.Despitenowprovenadvantagesintermsofenergysavingsfromindividualizedenduserpricing[c39],throughtheco-creationofvaluewithcustomers,thereremainregulatorychallenges[c40-c42].3.2 Network Components Thissectiondrawsuponexistingpublicationsandresearchfindingstoassessthecurrentandnear-futureexpectedenergycharacteristicsofkeywirednetworkingcomponents.ItalsoconsidershowthisprofilemightchangeastheInternetcontinuestodevelopoverthenextfewyearsbasedonthevariousservicesitsupports.3.2.1 Router A router is adevice that selectively forwardspacketsbetweencomputernetworks.RoutersoperateattheNetworkLayer(Layer-3)oftheOSImodel.Inrelationtodata,they inspect the headers of data packets they receive on their ingress ports andselectivelyroutethemthroughtoasuitableegressportwithreferencetoaroutingtable.Router architecture has developed for several generations: from bus-based routerarchitectures with a single processor, through bus-based router architectures withmultipleprocessorsandswitch-basedrouterarchitectureswithmultipleprocessors,to switch-based router architectures with fully distributed processors [c43][c44].However, a router remains primarily composed of line-cards, switch fabric, amanagementsystemandaRouterProcessor,oftenlocatedwithinthesameLine-CardChassis(LCC)[c45].The line-card is responsible for processing and forwarding packets using its localprocessingsubsystemandbufferspacesforprocessingthepacketsarrivingalongtheingress interfaces (orports)andawaiting transmissionat theegress interfaces.Theswitch fabric provides the sufficient bandwidth for transferring packets amongdifferentline-cards.Itreceivesdatafromtheingressline-cardinterfacesandswitchestoappropriateegressone(s).TheRouterProcessorisresponsibleformaintainingtheoverallforwardingtableanddistributingdifferentpartsofthetabletodifferentline-cards.Themanagementsystemincludesfunctionalityforcooling,powercontrolandalarmhandling.A more powerful router is a multiple-shelf device, which has a similar functionalstructuretothesingleshelfrouter.TheprincipledifferenceisthatithasseveralLCCsandthatareconnectedwithoneormoreFabric-CardChassis (FCC)elements [c46].For example, the Cisco CRS-1 multi-shelf routing system typically has 72 LCCs

connected via 8 FCCs [c47]. There are both fabric switches on the FCC and LCC.Generally,thefabricswitchemploysthree-stageswitchingdevices,knownasSwitchFabricElements (SFEs).Theyareall identicalandtheiroperationandbehaviouraredetermined by their location. In a single shelf router, the SFEs are all in the samefabricswitch.Inthemulti-selfroutingsystem,thedifferenceisthattheswitchfabricon theFCConlysupportsstage-2switching functionalitywhilst theswitch fabricontheLCCisresponsibleforstage-1and3switchfunctionalities.The data-plane, also called the forwarding plane, is the part of router architecturethatisresponsiblefordecidinghowtohandletheingresspacketsbylookinguptheroutingtableandthensendingthemtotheappropriatetheegressinterfaces.Ontheother hand, the control-plane is used for routing related control functions, such asgenerating the network map, the way to treat packets according to the differentserviceclassificationsanddiscardingcertainpackets.Ingeneral,thedata-planerunson the line-cards and the control-plane operates on the CPU and main memory.Anotherimportantpartoftheframeworkisthedynamicinterfacebindingfunction,which allows different interface resources to be allocated the according torequirements.Energy efficient networking means that we aim to transport the same amount oftraffic(inbits)withlowerenergyconsumption(inJoules)thaniscurrentlythecase.Fromtheaspectofpowerconsumption,boththenodeandlineequipmentshouldbeconsidered.However,inpractice,theconsumptionofthelineequipmentistypicallyomitted [c48].WaveDivisionMultiplex (WDM) linksaccount forbetween1.2%and5.8%oftheInternet’spowerconsumption,varyingwiththeaverageaccessrate[c1].Therefore,thepowerconsumptionoftheroutertendstoreceivemostattention.The energy consumption of a router is made up of two parts: static and dynamicenergy.StaticenergyisconsumedbytheLine-CardChassis(LCC)anddynamicenergyisconsumedbytheline-cardsandisrelatedtothetrafficload.Forexample,in[c49],the power consumption of two standard router platforms has been tested withdifferentcombinationsofline-cardsandLCCs.TheexperimentalresultsshowthattheLCCconsumesmorethanhalfofthemaximumobservedpowerconsumptionforanycombination. From this experiment, we can conclude that switching off thewholerouter can save more energy than only switching off the line-cards since theconsumption of the base system is the major contributor to the overall energyconsumption.Asmonitoringandperforming “green”experimentsonawidenetwork is generallyunfeasible, somesimulation frameworkssuchasECofenhavebeenproposed [c50].Such frameworks allow the simulation of large-scale networks and the impact interms of energy consumption and quality of service from on/off or slowdown(AdaptiveLinkRate,LowPowerIdle)powersavingschemes.Nevertheless,hardwareexperiments have been made on a French regional platform called GridMip

composedof3backbone routers,3border routersand150nodes.Theplatform isillustratedinFigure3.3.

Figure3.3:GridMipEvaluationPlatform

Theaimoftheexperimentswastoevaluatepreciselywhatistheconsumptionofthenetworkequipmentduringthebootprocessandhowlongittakes.Thisisparticularlyrelevantwhenconsideringapproachesthataimtoswitchoffroutersinordertosaveenergy during “light” operational conditions as it is necessary to take into accountpenaltywhenswitchingtheroutersbackonagain.WithaPLOGGwattmeterthepowerconsumptionofaCisco7600BorderRouter(R2)composedof3moduleswasmeasuredduring thebootprocessas shown inFigure3.4.Thisrouterconsumes785watts.Ittakes400secondstoswitchonarouter.Fromthefigureitiscleartoseethattheenergyconsumptionisneitherconstantnorlinearduring the boot process. Instead we observe appreciable steps in the energyconsumption. This has implications for the accurate modelling of dynamic energysaving network systems. It also shows that using typical contemporary equipmentthereisanappreciabledelaybeforearoutercanagainbecomeoperationalonceitispowereddown.

Figure3.4:PowerConsumptionofaCisco7600BorderRouterduringtheBoot

Process

Afurtherexperimentexaminedthepowerconsumptionofaroutermodule.Forthis,a4x10GbpsmodulebelongingtorouterR2wasturnedoff.Itwasthenturnedbackon20slater.Whenamoduleisoffitconsumes195wattsless.Whenthemoduleisswitched on, there are two phases during the boot process: one from 40s to 80swhich consumes 100wattsmore and a second phasewhich consumes 100WattsmoreagainasshowninFigure3.5.

Figure3.5:Cisco7600BorderRouter4x10GbpsInterfaceModule:Powering

DownandUpafter20secWe notice that the powering down of a module is achieved in one step almostimmediately.Onemightalsoassumethatpoweringupthemodulewouldalsotakeone step. However, a module recovers in two phases, with the core moduleinitialisingaheadofswitchingontheports.Thepoweringonprocesslastsalmost50s.Finally the power consumption associated with connected Ethernet ports wasexaminedas shown inFigure3.6. For thisexperiment,18 computerswerepluggedinto the Ethernet module of the router. The right axis represents the number ofunplugged interfaces (called Nb Itf in the graph). The main objective of thisexperimentwastoobservetheenergyconsumptionwithasthenumberofconnected(plugged)Ethernetportswasaltered.

Figure3.6:Cisco7600BorderRouterPower:PhysicallyPluggingEthernetPorts

We see from this experiment that themean power consumption of a port is 0.94Wattsandisapproximatelylinearwiththenumberofconnectedports.Experiments such as these are a necessary precursor to the creation of accuratemathematical models such as work exploring energy-performance trade-off in thecontextofnetworkanddata-centreoptimization[c51].3.2.2 Network Interface Card The roleofaNetwork InterfaceCardhasalreadybeen introduced in theprecedingsection where it was referred to as a line-card, forming part of the routerarchitecture.However,thisfunctionalityactuallyexistswithinanydeviceattachedtoa network. In a “local area” access network it is more common to use the termNetworkInterfaceCard(NIC).Inwirednetworks,researchlike[c52][c53]showthatenergyconsumptionofnetworkdevicessuchasNICs/linecardscanbelowerifoperatingratesarereduced.Thisisduetothefactthatdevicesoperatingatalowerfrequencycanreducetheirenergyconsumption bymechanisms such as Dynamic Voltage Scaling (DVS) [c54][c55]. Byloweringtheclockrateofsynchronouslogic,operationcanbemaintainedatalowerpowersupplyvoltage.However,thislowerstheswitchingperformanceofdevicesandsomustcoincidewithsituationswhentheprocessingloadislighter.Theauthorsof[c52]trytofindadecisionareaforwhichhardware-basedtechnique(SleepingorRateAdaptation) can capitalizeonoperational conditions. To this end,theystudythetrade-offbetweenlinksratesandtheirenergyconsumption.Auniform(linearenergy function)andanexponential (10Gbps,1Gbps,100Mbps,10Mbps)distribution of rates are used. The latter approach is considered as hardware

technologiesfortheseratesalreadyexist(Ethernettechnology).Wecanextractfromtheirresultsthattheuniformdistributionofratesisthemostsuitableonetoachieveagreaterreductionoflinksoperationsratesand,consequently,areductionofenergyconsumption.Moreover,theystatethatbothSleepingandRateAdaptationaretwoeffectivetechniquesdependingontheenergeticprofileofnetworkingequipmentandnetworkutilization.One step further, in [c56] it is presented a study based on [c52] in which theycompare energy savings in wired networks after applying Sleeping and RateAdaptation techniques. In this case, they draw some conclusions according tonetwork devices individual baseline consumption, i.e. based only in the amount ofenergy consumedby a component regardless of its operation rate. For the case inwhich the baseline consumption of network components is relatively high, it isinterestingtokeepasfewactivelinksaspossible(Sleeping).Ontheotherhand,ifthebaselineconsumptionofnetworkdevicesis low,RateAdaptationtechniqueismoresuitable.Inthelattercase,authorsrelyonresultsfrom[c52].Thus,theyassumeanduseauniformdistributionofratesfortheirexperimentsaswell.Otherwise, authors in [c57] perform several experiments to measure energyconsumption of two different Cisco routers: GSR 12008 and 7507. Both of theminclude their base systems (chassis plus routerprocessor) and line cards.However,we can extract from the results obtained in this case that the energy functiondescribedbylinecardsconsumptiontakesdiscretevaluesandassumeitsadaptationtoalogarithmicdistributionofrates.Rodgers, in turn, uses an Intel 82579 Gigabit Ethernet PHY and obtains energyconsumption values for 100 Mbps and 1 Gbps links [c53]. In this case, severalmeasurementsareperformedforeachofthethreedifferentenergylevelsatwhichlinks can be configured: Active, Idle and LPI (Low Power Idle from IEEE 802.3azstandard).We can extract from these results the value of links consumptionwhenthere is no traffic flowing through them (LPI energy level) as a function of theirconsumption when they are active. Thus, it can be assumed that the energyconsumptionofalinkwithaLPIenergylevelorSleepingtakesvaluesbetween5-10%outofitsconsumptionifitisconfiguredwithanactiveenergylevel.Furthermore, someheuristicsarecommonlyexamined,as in [c58],with theaimofachievingsignificantenergysavingsinthenetworkbyprogressivelyswitchingoffbothnodesandlinksaftertheprioraggregationoftrafficflows.Otherworksindeedsolvetheproblemusing Integer ProblemFormulation (ILP) by considering that only links[c59]orbothlinksandnodes[c60]canbeswitchedoff.3.2.3 (Reconfigurable) Optical Add-Drop Multiplexer TheROADMisaknownasan“anytoany”component,whichmeansthatitprovidesa flexibleway of adding, dropping or switching anywavelength to any node [c61].

Whenmultiplewavelengthsarriveatan input interfaceofanROADM,oneormorewavelengths can be dropped by pre-selection and in the output interface; one ormorepre-selectedwavelengthsareadded.Whatismore,anywavelengthcanbypassthe electronic processing function and pass through the ROADM unhindered tospecificports.Sincethesesignalsremainintheopticaldomain,thereislittlelatency.Furthermore, due to its “any to any” feature, ROADM technology allows DWDM-basednetworkstobetechnologyagnosticandthusmoreflexible[c62].Little work has focused on energy saving within ROADMs, perhaps because theytraditionallyconsume lesspower than routers.However,a recentexamplegiven in[c63]looksatusingROADMSmoreefficiently.3.2.4 Digital Subscriber Line Access Multiplexer Adigitalsubscriberlineaccessmultiplexer(DSLAM)isanetworkdevice,oftenlocatedin the telephone exchanges of the telecommunications operators. It connectsmultiple customer digital subscriber line (DSL) interfaces to a high-speed digitalcommunicationschannelusingmultiplexingtechniques.TheDSLAMequipmentcollectsthedatafromitsmanymodemportsandaggregatestheirvoiceanddatatrafficintoonecomplexcomposite"signal"viamultiplexing.TheaggregatedtrafficisthendirectedtoaServiceProviderbackboneswitch(BroadbandRemoteAccessServer(BRAS)atupto10Gbit/sdatarates.Dependingonitsdevicearchitecture and setup, a DSLAM aggregates the DSL lines over its AsynchronousTransfer Mode (ATM), frame relay, and/or Internet Protocol network (i.e., an IP-DSLAM using PTM-TC [Packet Transfer Mode - Transmission Convergence])protocol(s)stack.TheDSLAMactslikeaswitchsinceitsfunctionalityisatLayer2ofthe OSI model. Therefore it cannot re-route traffic betweenmultiple IP networks,only between BRAS devices and end-user connection points. From the BRAS theDSLAMtrafficisthenroutedacrosstheISPnetworktotheInternet.DSLequipmentcanbeasignificantsourceofenergyconsumptionforwirednetworkoperators.OnepossibleapproachtodecreasingtheoverallenergyconsumptionistoreplacelargecentralizedDSLAMslocatedattheexchangeswithsmallerremoteunitsclosertocustomers.Thisnotonlyreducesenergyconsumptionbutalsoincreasesthereachoftheaccessnetwork[c64].3.3 Architectures This section focuses on the architecture of communication networks and thecorresponding relationship with energy usage. However, a key dilemma whenexamining networks, and particularly the Internet, end-to-end is who pays for theelectricityandthepotentialsavingsexist.

At theperipheryof the Internet sit theendusers,be they residential customersorbusinesses. In both instances the Customer Premises Equipment (CPE) is typicallypowered by an electrical source provided by the end users, for which they pay.Conversely theaccessnetworksaretypicallyoperatedby InternetServiceProvidersusing their own equipment. These networks in turn are connected to carriernetworks,maintainedbyothertier-2andtier-1operatorsthatprimarilyconveytheinformation over large geographical distances using high performance (and power“hungry”)equipment.TheInternetisthusafederationofnetworkswithaplethoraofprovidersandconsumers.Interestinglygiventhehierarchical“pyramidal”structureoftheInternet,thenumberofnetworkingdevicesateachtierincreasesbyoneormoreordersofmagnitudeaswedescendfromthetier-1operatorsatthetipofthepyramid.ThismeansthatthegreatestpotentialenergysavingbyvolumeexistsattheCPE,wheresmallpercentagereductions in energy consumption can amount to substantial savings at a nationallevelduetothevastnumberofdevicesinvolved.Paradoxically thisarrangement leadstotwo issueswhen itcomestosavingenergy.Firstly,theamountofenergysavedataparticularCPEmayberegardedbytheowner/end-userasmarginal.Secondly,astheelectricitytopowerthesedevicesispaidforby theend-users there is less incentive tomotivate theaccessandcarriernetworkoperators tochange theirbehaviour.Despite this,weargue thatembracingenergysavingmechanismsdetailedinSection3.5canstillimpactonoperationalexpendituretoadegreeworthyofconsideration.3.3.1 Access Networks An access network is that part of a telecommunications network which connectssubscriberstotheirimmediateserviceprovider.Traditionally,accessnetworksconsistlargelyof pairs of copperwires, each traveling in adirectpath typically betweenaDigital Subscriber Line AccessMultiplexer (DSLAM) and the customer. The DSLAMconnectsmultiple customer digital subscriber line (DSL) interfaces to a high-speeddigital communicationschannel in theexchange,whichprovidesanaccesspoint totherestoftheInternet.Recently,accessnetworkshaveevolvedtoincludemoreopticalfibreprimarilyduetoitshighbandwidth,lowlossandnoiseimmunitycharacteristics.Opticalfibrealreadymakesup themajorityof corenetworksandhas started to “rollout” closer to thecustomer,untila full transition to fibre isachieved, inorder todelivervalueaddedservicesoverFibretotheHome(FTTH).Onegroupofoptical fibretechnologiesarePassive Optical Networks (PONs), whereby point-to-multipoint fiber links multiplehomestoacommonOpticalLineTerminal(OLT)attheserviceprovider'sexchange,using unpowered optical splitters to create a tree-like structure. Indeed EnergyEfficient Ethernet Passive Optical Networks (EPONs) are now being deployed,providing cost efficiencyandhighdata rate for the lastmile access [c65].A typical

EPONisaPoint-to-Multipoint (PMP)networkwithatree-basedtopology,whereanOLTconnectsmultipleOpticalNetworkUnits(ONUs)viaopticallinks.TheOLTplaysarole of distributor, arbitrator and aggregator of traffic. In the upstream direction(fromONUstotheOLT),multipleONUsshareasinglelinkandtrafficmaycollide.TheOLT distributes the fibre capacity using an upstream bandwidth arbitrationmechanismtoavoidcollisions.Inthedownstreamdirection(fromtheOLTtoONUs),data frames are broadcasted to all ONUs. ONUs filter and accept data that areaddressed to them. However, ONUs have to constantly listen and examinedownstreamtraffic,whichresultsinwastingsignificantenergyintheONU.3.3.2 Carrier Networks Generally,theunderlyingtransporttechnology(Layer1)ofthecurrentInternetusesSynchronous Optical NETworking (SONET) or Synchronous Digital Hierarchy (SDH);botharchitecturestransportvoiceanddataincontainersinanefficienttime-divisionmultiplexed manner over optical fibre. SONET/ SDH provide advanced supportservices such as linemonitoring to provide signals for protection switching and soforth. SONET/ SDH also allow Time division Multiplexing (TDM) circuits to bemultiplexedfromlow-speedtohighspeed intodifferentsizedcontainers.However,withincreasingrequirementsforserviceflexibilityandreliabilityandthedemandforlower expenditure, Internet Service Providers (ISPs) are not satisfied with this“traditional” architecture and are looking for more efficient ways to handle thegrowing volume of IP traffic. One promising architecture for the Next GenerationNetwork (NGN) is IPoverDWDMsince iteliminates the legacySDH transport layerandsosavesequipmentcostsandreducesenergyconsumption.An IP over DWDMnetwork is composed of two layers: the IP-layer and an opticallayer.Thiscanbeabstractedasastructurewherebyanodeiscomposedbyasetofnetwork equipment, e.g. a router and an optical transport switch, typically in theform of a Reconfiguration Optical Add-Drop Multiplexer (ROADM), that are theninter-connected by an Optical Line System (OLS), which includes optical fibre andamplifiers. In thismodel, thewavelength continuity constraint is often considered,wherebywhenanopticalconnectionpassesthroughtheopticaltransportswitch,itisrequiredtobeonthesamewavelengthchannelfromtheingresstotheegress.ThechannelremainsintheopticaldomainandOptical-Electrical-Optical(OEO)conversionisavoided.ThenodeinanIPoverDWDMarchitectureiscomposedofarouterandanROADM,which connected to similar devices via a transmission line system. The router isresponsible forstoringandforwardingpacketsandtheROADMisusedtoremotelyswitch the traffic allowing for adding, dropping or bypassing the router at thewavelength level without OEO conversion. An example is the Cisco IPoDWDMsolution where network nodes are composed of a Cisco CRS-1 multi-self routingsystemandanROADM[c66].Theoriginaltranspondershavebeencombinedintothe

routing system. The ROADM only manages add, drop and bypass in the opticaldomain.3.3.3 Grid Overlay Networks Grid computing via overlay networks enables a federation of computer resourcesfrommultipleadministrativedomainstocoordinatetoachieveacommongoal.Gridsareaformofdistributedcomputingwherebya“supervirtualcomputer”iscomposedof many networked loosely coupled computers acting together to perform largetasks. For certain applications, “distributed” or “grid” computing, can be seen as aspecialtypeofparallelcomputingthatreliesoncompletecomputers(withonboardCPUs, storage, power supplies, network interfaces, etc.) connected to a network(private, public or the Internet) by a conventional network interface, such asEthernet.Ananalysisoftheusageofanexperimentalgridhasbeenconductedoveraone-yearperiod. Based on this analysis, [c67] propose a resource reservation infrastructurethattakes intoaccounttheenergyusage.The infrastructure isvalidatedona large-scale experimental “Grid5000” platform and the gains in terms of energy arepresented.3.4 Traffic Considerations There is little published data concerning the traffic characteristics of commercialnetworksas it is regardedas sensitive.However, somepublisheddatasetsexist fortheAbileneandGéantnetworks[c68][c69].Figure3.7providesjusttwoexamplesoftraffic flow between specific source-destination pairs in the case of the Abilenenetwork.Thetrafficwassampledevery5minutesfor24weeks,omittingweekends.Asseenfromthefigure,trafficvariesconsiderablyovertime.Itisthisvariationthatpermitsnetworkresourcestobeswitchedoffwhentheyarenotneeded.

Timex104(5minuteinterval)

TrafficRate(kbp

sx105)

Timex104(5minuteinterval)

TrafficRate(kbp

sx105)

(b)TrafficDataforRoute96

(a)TrafficDataforRoute63

0 0.5 1.0 1.5 2.0 2.5 3.0 3.50 0.5 1.0 1.5 2.0 2.5 3.0 3.5

0 0.5 1.0 1.5 2.0 2.5 3.0 3.50 0.5 1.0 1.5 2.0 2.5 3.0 3.50

2

4

6

8

10

12

0

2

4

6

8

10

12

0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

4.5

5.0

0

0.5

1.0

1.5

2.0

2.5

3.0

3.5

4.0

4.5

5.0

Figure3.7:TypicalTrafficDataover24Weeks

The variation in traffic over time allows for opportunities when resources can beplaced into a sleeping state. Furthermore, the regularity in the autocorrelationpattern lends weight to the idea that the flow of data between given source-destinationpairsistemporallypredictable[c70].Inadditiontoloadvariations,trafficcanalsobedifferentiatedintermsofwhereitiscoming from and going to. For example, within the context of IP over DWDMarchitectures,therearetwotypesoftrafficineverynetworknode:router-terminatedtrafficandopticalbypasstraffic[c71][c72].Intheformercasethetrafficisterminatedat the router. The router looks up the routing table at IP-layer to decidewhere toforward the packet. In this scenario, since the packet may be transported in thephysical fibreusingadifferentwavelength, the trafficneedsanoptical toelectricalconversion before it processed and forwarded by the router. In fact, the powerconsumption of optical and electrical transfer in the IP router ports is a majorelement of the overall power consumption. Therefore, some researchers areconsideringdesigninganenergy-efficiencynetworkbyminimizingIProuterports.Foropticalbypasstraffic,incontrast,therouterdoesnotprocessthetraffic.Insteaditistransporteddirectlythroughtheintermediaterouter(s)viapre-setlightpaths.Sincethetrafficacross thecorenetwork isbecomingmoredistributedduetochanges inapplication service deployment, the proportionof bypass traffic is increasing [c73].For example, it is possible that up to 70% to 80% [c66] traffic passing through anopticalnodeisofthebypasstypeanddoesnotrequireterminationattherouter.3.5 Energy Saving Mechanisms Energy saving inwired networks can be generally divided into static and dynamicmechanisms [c4].Staticonesconsidernetworkdesignandconfiguration in respectto power. The approaches are to design and place network resources in a more

energy-efficiencyway.Incontrast,dynamicapproachesattempttomakebetteruseoftheexistingnetworkresourcesbytrafficengineeringinresponsetovaryingtrafficload through infrastructure sleeping, rate adaptation and so forth. The followingsectionsummarisesthestate-of-the-artinthesetwoareas.3.5.1 Static Mechanisms Staticmechanismsmeantodesignandconfigurethenetworkfromtheperspectiveofpowerconsumption.Thatistosay,toplanpower-awarenetworks.Oneapproachistoconfigurethecombinationofchassisand linecards for lowingtheoverallpowerconsumption. In [c49], theauthor indicates that thebasesystem(includingchassis,switch fabric and router processor) consumesmore than half of thewhole powerconsumption of a router by testing two generic router platforms. Therefore, onemethod toachievepower-awarenetworkingat thedesign stage is tominimize thechassis number and to maximize the number of line cards per chassis. AnothermethodisplanningthenetworkistoreducethetrafficprocessedbytheIProuters.Generally, a packet is transported via 14 routers on average across the networks[c66].Therearetwomeansoftransportingpacketsintheopticallayer,lightpathnon-bypass and bypass. The non-bypass lightpath is always terminated at the IP routerports and then transformed into an electrical signal for processing. In contrast thelightpath bypass approach transports packets through intermediate nodes directlywithoutoptical-electrical-opticalconversion.ReducingthetrafficprocessedbytheIProutersmeans that theuseof bypass lightpaths alsominimizesO-E-O conversions.[c71] indicates the power consumption of these two different methods, router-terminated and bypass traffic, for packet transportation. In the former case, itconsumes about 10nJ/bit. Conversely the traffic processed in the WDM layer,consumeslessthan1nJ/bit.Thus,muchresearchhasbeendevotedtoexploitingthischaracteristic.In [c12], the IP over WDM network achieves energy efficiency by minimizing thenumber of IP router ports/interfaces since the ports/ interfaces play an importantroleinthepowerconsumptionofnetworks.Itisalsothefirsttimethatthelightpathbypassapproachistakenintoconsiderationforenergysaving.Theenergy-minimizednetwork design uses the lightpath bypass approach tominimize the number of IProuterportforreducingthepowerconsumptionviaO-E-Oconversions.In[c13],thepaperconsidersthemulti-shelfroutingsysteminanIPoverWDMnetworkforsavingtheenergy.BothpapersemployMixed-IntegerLinearProgramming(MILP)tobuildamodelforminimizingthepowerconsumption.Another approach is energy-efficient network design. Here the aim is to deviseenergy-efficient architectures during the network design state. For example,researchers have proposed an energy-aware design to minimize the energyconsumption of all the network components in conjunction for an IP-over-WDMnetwork[c13].Twoalternativeswereconsidered:lightpathnon-bypassandlightpathbypass.Intheformercase,IProutersprocessandforwardallthedatacarriedbythe

lightpaths,whilstthelatterallowsIPtraffictodirectlybypassanintermediaterouter(whichisnotthedestinationnode)viaacut-throughlightpath.Resultsshowedthatlightpath bypass could save significant energy comparedwith the non-bypass case(between25%to45%)duetothereductionofthenumberofIProutersneeded.Theyalso confirm that the total energy consumption of IP routers ismuch greater thanthatoftherestofopticalnetworkequipment,suchasErbium-DopedFiberAmplifiers(EDFAs) and transponders, accounting for over 90% of the total network powerconsumption.3.5.2 Dynamic Mechanisms Dynamicmechanismsusetraffic-engineeringmethodstomakebetteruseofexistingresources instead of designing and configuring a network in respect of the powerconsumption.Theyreact"quickly"tochangingconditionssuchasvariationsintrafficload.Methods that belong to this class are rate adaptation, infrastructure sleepingandgreenrouting.Wealsoconsidervirtualisation.3.5.2.1 Rate Adaptation Rate adaptation is a mechanism by which a communications device can adjust itsprocessing rate based on the currentworkload conditions.Many devices have thecapacitytooperateatarangeofprocessingrates.Whentheworkloadislow,adeviceoperatesatthelowspeed,whichconsumeslesspoweratthesametime.Therefore,when the traffic load is low, lowering the transmission speedof device reduces itspowerconsumption.Rateadaptationisdescribedin[c15]whereitiscomparedwiththe infrastructure sleeping. The conclusion is that the both approaches are usefuldepending on the hardware capabilities of device and network utilization. Forexample,anISPcoulduserateadaptationinthedaytimeandinfrastructuresleepingatnighttosatisfythedifferenttrafficloadrequirements.Nevertheless,theabilitytoperformsleepingandrateadaptationarenotsupportedinexistingnetworkdevices.Evenso,theseapproacheshavebeenincorporatedintothefutureEthernetprotocolstandard as: Energy Efficient Ethernet,whichwas approved inOctober 2010 [c74].Thus, we should see these approaches receive broad adoption throughout theInternetoverthecomingyears.Akeyissueisthatcurrentequipmentisconstrainedbytechnologylimitations.Someresearchershaveexploredthetrade-offbetweentheamountofenergythatcouldbesaved inwirednetworksand thediscretenumberofoperational ratemodes tobeimplementedinlinecards[c75].Resultsshowthatitisnotnecessarytomanufacturelinecardsthatsupportalargenumberofdifferentoperationalratemodes,butthatalimitednumber such as four different operational ratemodeswouldbe enough toachievesignificantreductionsinenergyconsumption.Forexample,fourenergylevelsareenoughtosavesignificantamountsofenergy(around22%).Agreaternumberofenergy levels do not result in large savings when compared to the case when aninfinitenumberofenergylevelsareassumed.

Other researchers have also considered the concept of multi-rate adaptationwhereby physical interfaces adapt their rates accordinglywith the typical Ethernetvalues. For example, a 10Gbps card is considered capable of intermediatetransmission speeds of 1Gbps, 100Mbps, 10Mbps. Consumption of theseintermediateratesarethenconsidered,accordingly[c76].Theuseof"adaptive”linkrates can then be applied in concert with "energy aware routing" and "multilayertrafficengineering"asdescribedin[c77][c78],respectively.Theapplicationofadaptivemulti-ratetechniquesandmultilayertrafficengineeringinhourlyintervalscanreducetotalconsumptionintoawiredbackboneto7%simplybyadaptingthecapabilitiesofexistingsystemstotherealneedsoftraffic.Thisresultisobtainedfrom[c79]usingatheoreticalscenarioformedofawiredbackbonenetworkdeployedoverfivecrossconnectionlevels(3aggregationand2core)asillustratedinFigure 3.8. It included a physical coverage 40000km2 (equivalent to the size ofSwitzerland) and a total volume of 6.3 million connected devices, distributedaccordingtothetypicalcharacteristicsofurban,suburban,andruralareas.

Figure3.8:ExampleMultilayerBackboneNetwork

Thedeployednetwork,basedonOTNtechnology,includes2000cabinetswithnearly4000communications interfacecards.Total consumptionof53MWatts (perhour)wasestimatedwithoutconsideringadditionalcoolingsystems.Theadaptationoftherouting and capacity of the links reduces consumption to nearly 2.5 MWatts onaverage,withpeaksofupto4MWatts.Giventhe1:1ratiointermsofrefrigerationequipment energy consumption, a decrease in networking energy consumptionwould see a similar reduction in the consumption of these ancillary systems.However,asindicatedin[c80],areductionof1wattinthesystemcouldreducethe

actualoverallconsumptionby2.42watts,sothetotalsavingcouldbeasmuchas5%onaverage,withpeakscloseto8%asshowninFigure3.9.

Figure3.9:TypicalAveragedInternetTrafficLoadandConsumptionReduction

(usingEETechniques)3.5.2.2 Infrastructure Sleeping Infrastructure sleepingmeans that when equipment is idle, it enters a low powerconsumption state called standby/sleep whilst retaining the information formaintaining its operational state when it again wakes up. An example is the“hibernation”stateofpersonalcomputers.Sincea routerstill consumesabout90%of its full-loadenergywhen it is idle [c81],approaches that turn off unneeded devices or permit equipment to enter acoordinated sleep state have more potential for saving energy in a network.Furthermore,duetothewell-knowndaily trafficvariations,where loadvaries inanalmost sinusoidmanner over the course of each day in response to people’s dailyactivities, the infrastructure sleeping approach can achieve a significant savings.Consequently,considerableresearchhasbeendevotedtotheinfrastructuresleepingprinciple.In[c11],twotypesofsleepingareidentified,uncoordinatedsleepingandcoordinatedsleeping. The former lets the equipment enter a sleep state based on localinformation, e.g. during the inter-arrival time between packet transmissions. Theothertakesaglobalperspectiveandreroutestrafficviasomeequipmentinordertoletotherdevicesenterasleepstate.Forexample,in[c16],theauthordiscussestwomechanismsforsleeping:traditional“wake-on-arrival”andaproposed“buffer-and-burst”.Theformeronewakesupthesleepingdevicewhenthenewpacketarrives.

Sinceittakestimeforadevicetoreturntotheactivestate,ifthepacketinter-arrivaltimeissmallerthanthetransitiontime,theenergysavingislimited.For“buffer-and-burst”sleeping the traffic is shaped intoburst-by-burstunitsandstored inabufferthatisprocessedwhenbufferisfull.Thisprovideslongersleepingperiodswhilethetraffic load is low. The results show that the proposed mechanism can obtainsignificantenergysavingswhilsthaving little impactonthepacket lossanddelay.Asimilar approach to buffering or (re)scheduling bursts to improve opportunities toshutdown(sleep)inthecontextofgrids[c82]andnetworks[c83][c84]havealsobeenconsidered.In[c85],LChiaraviglioetal.modelledanetworkformaximizingtheenergysavingbyswitching off idle nodes and idle links. The problem isNP-hard, so in [c15] severalsimpleheuristicalgorithmsareemployedwhichsortallthenodesdependingonthenumberoflinks,theamountofflowstheyaccommodate,orusingarandomstrategyto switchoff for savingenergy. In the simplenetwork scenario,which includes thecore,edgeandaggregatenetworks,itispossibletoswitchoff30%oflinksand50%ofnodes. The same author extends theirwork considering a real-world case in [c60].Theproblemforminimizingthenumberofnodesandlinksforsavingenergygivenacertain traffic demand has been solved by Integer Linear Programming (ILP) forsimple networks. The algorithm switches off devices that consumemore power indescendingorder.Theresultsshowthatitispossibletoreduceenergyconsumptionbymorethan23%, i.e.3GWh/year inarealnetworkscenario. Ina furtherexample[c86], the author took the Italian telecommunications network as referencearchitectureandevaluatedtheextentofenergysavingifgreennetworktechnologieswere adopted, such as power scaling and standby exploitation. It assumed thenetwork equipment, such as routers, home gateways and Digital Subscriber LineAccessMultiplexers (DSLAMs) have three power consumption states: the full load,idleandstandbyaftermakinguseofgreentechnologies.Thepowerconsumptionoffullloadstateandtheidlestatevarieslinearlyaccordingtotheactualtrafficload.Thepowerconsumptionof thestandbystate isasmall fractionof the idlestatepower.Theconclusionofusingdifferentgreentechnologieswithinthereferencescenarioisthat their adoption may save up to 68% power consumed relative to the currentapproach.However,theseapproachesallsufferfromtheproblemthatwhenarouteroralinkisswitched off in a network, they lose the present state of the network and so willtrigger routingprotocol reconvergenceevents. The traffic is re-routed along longerpathsthatmaybenotacceptableduetocongestion,qualityofserviceandtheextradelay. Therefore, when making use of the infrastructure sleeping, networkconnectivity and Quality of Service (QoS) must be considered to guarantee thereliabilityofthenetwork.Recently,therehasbeenanewdirectionininfrastructuresleeping:toswitchofftheline-card insteadofthewholerouter.Sincethedisappearanceofwholeroutermaytrigger reconvergence events and may thus lead to network disruption and

discontinuitiessomeresearcherschoosetoswitchoffindividuallinecardsandremapthelinkstootherones.Thisavoidsdiscontinuitiesandsavespowerwhenthetrafficloadislight.In[c87],routingre-configurationatdifferentlayersiscompared:theIP-layer(thevirtuallayer)andWDM(thephysicallayer)forthepurposeofswitchingoffline-cardsforsavingenergy.Theschemewhichre-routesdemandsinthevirtuallayerachieves thebestsaving.Similarly, in [c88], italsoproposesaschemetoswitchoffline-cards when traffic load is low. Fisher et al. [c89] propose another form ofinfrastructure sleeping where they shut down the redundant cables and the line-cardsinsteadofthewholerouterduringperiodsoflowutilization.Theswitchingoffline-cardapproachmaybeagooddirectiontopursue.Neverthelessinarouter,themajor source power consumption is the base system (chassis, processor andswitchingfabric),whichconsumesabout50%ofthetotal[c49].Therefore,toswitchoffthewholerouterstillyieldsmoresubstantialenergysavings.[c51] haveworked on introducing the ”self-optimizing” autonomic property at thehardwarelevelbyapplyingittotheoptimizationofdata-centresenergycosts.Theyhaveproposed an approach to describe a compromisebetween, ononehand, thepower consumption of the network infrastructure and on the other hand, thedeteriorationoftheQoSfortheapplicationsusingthisnetwork.Beingabletocontrolthe dynamic reconfiguration at two levels: at the application level by dynamicallyreconfiguringQoSprofiles and at thehardware level by switchingon andoff links,modulesorroutersallowstohaveaglobalmanagementofthedata-centre.Indeed,the use of an autonomic manager allows the administrator to control the energycosts or the performance by varying only one parameter that handles a high-levelmanagementpolicy.The goal was to deal with the performance/electric consumption dilemma for thenetworkpart.This isachallengeforyearstocomeastheperfectsystemmusttakeinto account the energy consumption of the machines and also the networkequipments,theQoSandfinancialcostsforexample.3.5.2.3 Sleep-Supporting Protocols Traditionally, network equipment does not typically possess sleeping state(s).However, recent industry standardizationefforts havebeenproposed for enablingtheequipmenttooperateinamoreenergy-efficiencymanner.Forinstance,theIEEE802.3az Energy-Efficient Ethernet (EEE) standard has been approved [c74] byInstituteofElectricalandElectronicEngineers(IEEE)inOctober2010.EEEprovidesahardwaresupportforenergysavingwhichdefinesamechanismtoreducethepowerconsumptionwhenthetrafficloadislowforaphysicallayerprotocol.Furthermore,theInternationalTelecommunicationUnion(ITU)hasorganizedaseriesofsymposiaon ICT and climate change [c90]; the Alliance for Telecommunications IndustrySolutions (ATIS) also set up a specific committee called Network Interface, PowerandProtectionCommittee(NIPP)[c91]workingondevelopingstandardsfornetwork

interfaces,withpowerandprotectionincludingreducingthepowerconsumptionofequipment.3.5.2.3.1Energy-EfficientPassiveOpticalNetworksPassiveOptical Networks (PONs) are currently themajor contributor to the powerbudgetoffixedopticalaccessnetworks.InaPON,thelargestenergyconsumptionisatthecustomerpremises,i.e.,bytheOpticalNetworkUnits(ONUs)thataccountforover 65% of the overall power consumed by PONs [c92]. Methods both at thephysical layerandattheMAClayerhavebeenproposedtoreducetheONUenergyconsumption [c93]. MAC layer methods are mainly based on the possibility ofswitching to sleep mode the ONU during low traffic period. Therefore strategiesbasedoncyclesof sleepandwakeperiodshavebeen standardized [c94].Howeversuch methods trade energy reduction for increased frame transfer delay. Forexample,whentheONUissleepingitcanneitherreceivenortransmitpackets.Forthisreason,theconceptofvariablesleepperiodswasintroducedin[c95]tosaveenergyattheONUswhileavoidingservicedegradationinthepresenceoftrafficwithdemandingdelayrequirements.Therationaleissimple:eachclassofservice(CoS)isassignedaspecificsleepperiodsuchthatthedelayconstraintofthespecificclassismet. If multiple CoS are received by one ONU, the sleep period of the mostdemandingclassisselected.Thelengthofasleepperiodcanbefixed,(i.e.,setapriorito a predefined value [c96]) or it can vary. In fact, users connected to ONUsmaydynamically modify their service subscriptions (e.g., request a video streaming towatchamovieforsomehours),possiblychangingtheoveralldelayrequirements.Ontheotherhandsuchchangeshaveaperiodofhours,thereforethesleeptimedoesnotneedtobeupdatedveryoften.Insuchasituationitmightbeconvenienttobeabletochangethe lengthof thesleepperiod,byeithertying it tothedownstreamframe statistics [c97], or by having it varying based on the service class delayrequirementsandtothepredicteddelaystatistics[c95][c98].When focusing on strategies based on variable sleep periods it becomes crucial todevelop effective decision mechanisms for the sleep length. One possibility is tomodel the implementation of cyclic sleep in a PON using queuing theory. Forexample,theauthorsin[4b]proposedamodelwhereanOpticalLineTerminal(OLT)-ONU system is assimilated to a polling system with gated service, exponentiallydistributed frame inter-arrival time, and generic service time. This model wasimplementedinOPNET,anditsperformancewasassessedthroughsimulationswithascenarioconsistingofa10Gigabit-PONwithoneOLT linked toconnected tosingleONU.TheOLTisawareoftheservicessubscribedbytheONUintermsofQualityofService (QoS) Class, IP packet Delay Variation (IPTD) and IP packet Transfer Delay(IPDV) constraints [c99], and bandwidth (i.e., data rate) [c100]. The values for thedelay constraints are related to a generic end-to-end IP network composed of nspans,contributingequallyandadditivelytotheconsideredend-to-endparameters,wherethePONistheaccessspan.Eachserviceisdefinedwithitsownframepayload

sizeanddatarate,definedbystudyingtherealtrafficdistributioninanetwork[c100-c104].TheresultsshowtheeffectivenessofthemodelinchoosingavalueofsleeptimethatguaranteestheIPDTconstraintforalltheservices.ByfocusingonlyontheIPDTvalueit isnotpossibletoguaranteetheIPDVrequirementsforthestrictestservice inthefirstcombination;however,theobtainedenergyefficiencyishigh,i.e.slightlybelow90%. When the sleep time was calculated as a function of both IPTD and IPDVconstraints,theIPTDconstraintisagainsatisfiedand,thistime,thesimulatedIPDVisbelowthemaximumallowedtime.However,theenergyefficiencyislower,butisstillarespectable50%.3.5.2.4 Green Routing Inthecurrentcorenetworkarchitecture, IProutersconsumea largerproportionofenergy as compared with the underlying Synchronous Digital Hierarchy (SDH) andWDM layers [c12][c13][c105].With the Next Generation Network (NGN), currentlythe industryvision is toemployan IPoverWDMarchitecturewhicheliminates theoriginal SDH transport layer because this lowers the equipment costs and energyconsumption[c106][c107].However, inanIPoverWDMarchitecture,sincethere isno traffic grooming in the SDH layer, IP routers are required to perform more IPpacket processing, e.g. sub-wavelength grooming, and thus consumemore energy.Therefore, in this scenario, the switching off routers or selective interfaces whentraffic demand is light has been regarded as a promising strategy and exploredintensivelyinrecentyears[c15][c16].Energy-awareroutingschemesusuallyfollowtheform:

! Aphysicalnetwork topology constructedof routersand links, inwhich linkshaveaknowncapacity

! Knowledge of the amount of traffic exchanged by any source/destinationnodepair

! Thepowerconsumptionofeachlinkandnode! Find the setof routersand links thatmustbepoweredon so that the total

powerconsumptionisminimisedsubjectto:! Flowconservationandmaximumlinkutilizationconstraints! PossibleQoSconstraints–suchasend-to-enddelay

Theseapproachesnaturallyleadtosomeresourcesbeingheavilyutilisedinorderforothers tobe“turnedoff”.Toavoidcongestion it isnormal to represent linksbyan“effectivecapacity”.A typical example is the Green Distributed Algorithm (GRiDA), which exploits linkstate routing protocol to disseminate load and power consumption of links [c108].Being reactive, it does not require any knowledge of the traffic demandmatrix. Adistributed algorithm switches off links when they are under-utilized, and theirabsenceinthenetworkdoesnotaffectthenetworkfunctionalities,andswitcheson

idle links when capacity is required to guarantee a proper reaction to faults andchangesinthetrafficdemand.Apenaltymechanismreducestheriskofundesirableactions.However,themainproblemofgreenroutingprotocolsisthatwhenroutersortheirinterfacesareswitchedoff,theylosetheabilitytoexchangeroutingprotocolsignalling messages. In other words, the logical IP-layer topology changes when anodedisappearsfromthenetwork.Thiscantriggeraseriesofreconvergenceeventsthatcancausenetworkdiscontinuitiesanddisruption.NeverthelessGRiDAiscapableofreducingenergyconsumptionby50%.Routers are the most power-hungry devices, followed by the electronic signalregenerators and ROADMs, with the all-optical devices such as OXC and opticalamplifiers being less power consuming. In the context of wide areatelecommunications networks (WAN), inwhichnetworknodes are spreadout longdistances, such a knowledgemay be exploited by a properly crafted energy-awarerouting algorithm that routes connections through the NEs which are currentlypowered by green energy sources, minimizing the overall GHG emissions of thenetwork.Sincetherenewableenergysourcesvarywithtime,itisnecessarytohavethis information updated for each node. In themodern GeneralizedMultiprotocolLabelSwitching (GMPLS)controlledopticalnetworks, suchupdatesmayberealizedbyOpenShortestPathFirst–TrafficEngineering(OSPF-TE)extensions.In[c109],suchanextensionispresentedandusedtospreadtheenergy-relatedinformationthroughthenetworkonafixedtimeinterval.Greenawarenessisenabledbyfloodingenergysource informationoverthenetwork,which isused inOSPF-TEroutingdecisionstolowertheGHGemissions.Observingthebehaviouroftheproposedalgorithmunderdifferentscenarios,itisseenthattheproposedalgorithmcansaveupto27%oftheGHGemissions (in termsof costunit) at theexpenseof amarginal increase in thepathlength,comparedtotraditionalshortestpathroutingalgorithm.Furthermore,in[c110], GMPLS extensions are used to integrate energy efficiency and networkresilience.Resultsshowthattheelectricalportusagecanbesignificantlyreducedbyusing intelligent regenerator andwavelength converter placement strategies. Thus,the scarce usage of electrical ports can help to reduce the power budget of theoverallcommunicationsystem.Anothermethodforachievingenergysavingsinexcessof30%inWideAreaNetworksispresentedin[c111].Theapproachappliesa limitedsetofpre-calculatednetworktopology configurations derived via a Genetic Algorithm across the day. The GAdeterminestheminimumsetofresourcesrequiredinordertosupportagiventrafficdemand. Informationgleaned fromSNMP trapmessages, triggeredby theuseofalink utilization threshold, determine when to switch between configurations. Thethresholdemploysmovingaverage smoothingand isdiscretely readjustedover thecourseofadailycyclebasedonanticipatedbasal loadvariations.ByexploitingMT-OSPF this approach provides a scalable and flexible means of reconfiguring aninfrastructurethatavoidsroutingdiscontinuities,excessivecomputationaleffortandtheexchangeofconsiderablevolumesofcontrolinformation.

Despitethepromiseofgreenroutingprotocolsthereremainsignificantimpedimentsto change. For example, Operators have Service Level Agreements with theircustomersandarewaryofadoptinganytechnologythatincreasestheriskofoutagesorlowersQoS.Furthermore,resiliencetypicallyequatestoredundancy.Redundancyimpliesadditionalequipmentisin“hot”standbyforfastprotectionswitch-overintheevent of a failure. Therefore, although switching off these “spare” resources savesenergy,theymaybeneededand,ifso,theyhavetobeavailableatveryshortnotice.Consequentially, the overriding commercial interests of the Operators may causethemtoshyawayfromnewroutingprotocolsandfavourenergy-savingtechnologiesthataretransparenttotheIPlayer(suchasrateadaptation).3.5.2.4.1Energy-AwareDynamicRWAAlgorithmThe energy consumption and Greenhouse Gas (GHG) emissions of the networkelementscanbetakenintoaccountwhendefiningthemetricofroutingalgorithms.In particular, inWDM optical networks, the GreenSpark [c112] heuristic has beendevelopedtoaddressthemulti-objectiveRoutingandWavelengthAssignment(RWA)problemofminimizing either thepower consumptionor theGHGemissionswhilstmaintaining the connections blocking probability as low as possible. GreenSparkoperates online, routing the connections as they arrive. At each new connectionrequest, two stagesareperformed: in the first stage, thebest load-balancedpathssatisfying the connection request and QoS constraint are selected (typically, threepartiallydisjointpathsareidentified).Atthesecondstage,amongsuchfeasiblepaths,the one thatminimizes the power consumption (MinPower) or theGHGemissions(MinGas)isfinallychosentoroutetheconnection,accordingtotheobjectivefunctionselected. Apart from minimizing the power consumption or the emissions,GreenSpark is even able to reduce the blocking probabilitywhen comparedwith atraditionalroutingalgorithm,suchasShortestPathorMinimumInterferenceRoutingAlgorithm(MIRA).ApartfromdefininganenergyconsumptionmodelforanIPoverWDM network, one of themost significant added values of the framework is theincorporation of both physical layer aspects, such as power demand of eachcomponent, and virtual topology-based energymanagementwith integrated trafficgrooming,soastoadverselyconditiontheuseofenergy-hungrylinksanddevices.3.5.2.4.2Energy-AwareOSPF-TEExtensionsTheinformationconcerningenergyconsumptionofthenetworkelementshastobedisseminated across the network routers that can thus determine the best orgreenestrouteforeachdestinationbyprovidingeachrouterwithcompleterelevantknowledge of the network. In an Autonomous System, a dynamic routing protocolsuch as Open Shortest Path First with Traffic Engineering extensions (OSPF-TE) isemployed to build the complete map of the network (link-state protocol) in eachrouter. However, OSPF-TE does not carry in its Link State Advertisement (LSA)messages any information about the current energy consumption of the networkelements.Therefore,energy-awareOSPF-TEprotocolextensionshavebeenproposed

in severalworks [c113][c114][c115][c116], togetherwithproperly craftedRWAandre-optimization algorithms that exploit such information. Specifically, opaque LSAshave been extended with new Type-Length-Value (TLV) fields, carrying power andCO2 emissions information. Such LSAs are spread in the network (among theDesignatedRouter/BackupDesignatedRouterandallthenodes)andareincludedinthe LSAs every time a significant change in the energy consumption or emissionsoccurs(e.g.,achangeintheenergysourcepoweringadevice,asanexample,whenswitchingfromsolarpanelsenergytotraditional“dirty”energysource).Resultsshowthat the proposed extension helps in reducing the overall network powerconsumption/GHGemissionswhen compared to traditional shortest path andpureload-balancing algorithms, at the expense of a limited increase in the blockingprobability. However, a network operator should also consider extra networkoverhead, and thepossible additional expenses for obtaining the information fromtheassociated“SmartGrid”powersupplynetwork.3.5.2.5 Network Virtualization: Virtual Router Migration Network virtualizationmay be seen as one of the basic building blocks of a futureInternet.ItisexpectedtoovercometheperceivedossificationofthecurrentInternetthrough more flexible management of network resources. Typical networkvirtualizationapproachespartitionphysicalroutersintoanarbitrarynumberofvirtualrouters [c117][c118]. A virtual router is a logical router that separates behaviouralfunctionalityfromthephysicalplatformthathoststheentity, includingmechanismsand tools for management, configuration, monitoring and maintenance. Virtualrouters provide a means of more effectively utilising a single physical router. Thefunctions of several small routers can be performed on a single large router thataccommodates the virtual router instances. Virtualization is already supported bysomecommercialrouters,suchastheCisconexus7000VDC[c119]andtheJuniperlogical router [c120]. They both belong to the class of logical routers. To bemoreprecise,theconceptofavirtualrouterandalogicalrouterarenotalwaysthesame.Avirtual router is a simplified routing instance,which has one routing tablewhilst alogicalroutercansupportmultipleroutingtablesatthesametime.Thatistosay,alogicalroutercancontainseveralvirtualrouters.Virtualroutersaretheninterconnectedbyvirtuallinks,whichmayspananarbitrarynumberofphysicallinksinthesubstratenetwork.Suchanapproachallowsnetworkadministrators to create virtual networks specially tailored towards a certainobjectives. These networks may have topologies that differ substantially from theunderlyingphysicalnetwork.Moreover, thesenetworkscanbecreatedon-demandand removed once they are no longer needed. Indeed, one particular advantagegained through virtualization lies in the mobility of virtual resources. Sincevirtualization itself provides an abstraction from actual hardware, virtual resourcesarenolongerboundtoaparticularhardwareinstance.Thus,avirtualroutercan,forexample, be moved from one physical router to another physical router. Thetechnicalbasisforthisisthelivemigrationofavirtualmachine–i.e.movingavirtual

machine from one hardware instance to another instance without suspending it[c121].Thisconcepthasalreadybeenused indata-centresandcloudenvironments[c123].Forrouters ithasbeenenvisionedasanewnetworkmanagementprimitive[c122]. Migrating routers, however, requires more sophisticated migrationmechanisms. Inparticular, some formof traffic redirection is necessary inorder topreservetheNetworkLayertopologyandkeepexistingconnectionsalive.Whileinasingledata-centretrafficredirectioncanbehandledtransparentlybytheLinkLayer(e.g., via artificial ARP updates), this is not easily possible for routers as primeNetworkLayerequipment.Migrationhastobeperformedacrossdifferentsubnets.This concept is commonly knownaswide-areamigration.Again, thishasbeen firstintroduced for data-centre scenarios with multiple, distributed data-centres[c124][c125].Amajoradvantageofvirtualroutermigrationisthepossibilitytoreactdynamicallytochangestothenetworkwherebyvirtualroutersaremovedamongdifferentphysicalhostswithoutchangingthenetworklogicaltopology.Forexample,migrationcanbeusedasaremediation/recoverymechanisminthefaceofpendinghardwarefailuresornaturaldisasters[c126]. Inthiscase,migrationcanincreasetheoverallresilienceof a virtual network by keeping the network operational even while part of thehardwarefails.Anotherapplicationofvirtualroutermigrationisthereconfigurationofavailablehardwareresources.Virtualnetworkscouldbecreatedanddestroyedina highly dynamicmanner. However, this is a particular challenge for the resourceallocationalgorithm,asitmustcopewiththedynamicreconfigurationwhilstavoidingfragmentationof theavailablehardware resources [c127]. Themigrationmustalsoavoidoscillatingstates,akintorouteflapping,wheretheactofmigrationitselfcausesachangeofnetworkstatesufficienttotriggerarapidmigrationbacktoaprecedingstate.Migration ismorecomplex thanroutervirtualizationsince itneeds tosolveseveralproblems. The first one is how tominimum the outages whenmoving the virtualrouter.Sincemovingavirtualroutertonewphysicalplatformalwaysleadstosomedelays, it may cause some network disruption and discontinuities. Specifically, thedataplane,whichisresponsibleforforwardingpackets,shouldnotbehaltedduringthistransferprocess.Ontheotherhand,thecontrolplaneusuallyhassomeformofretransmission mechanism that provides protection from interruptions. It is moretolerantofdisruptionthanthedataplane.Thesecondissueishowtorealizethelinkmigrationafterthevirtualroutermigratestothedestination.Inthepast,anIP-layerlink corresponds to a physical link. If so, it is hard to realize the virtual routermigration, as it is not possible for a physical link to follow the virtual router byunplugging and plugging-in the cable. Thanks to the recent new transport layertechnologies, such as the programmable transport network [c18], it is possible toallowthephysical linksbetweenrouterstobedynamicallyconnectedorsuspendedaccording to requirements. For example, when a virtual routermoves to the newphysicalplatform,thelinktothevirtualroutercanbeestablishedbythesettingupofanewlinkintheunderlyingopticaltransportnetwork.Athirdissueistoconsiderthe

facilities needed for supporting virtual routermigration. For example,whether thedestinationphysicalplatformhasenoughinterfacesforpacketdelivery,whethertheswitchfabric intherouterhastheabilitytoprocessall therequiredswitchingtasksafternewvirtual routerhavebeenaccommodated,andwhether the interfacescandynamicallybindwiththedataplane.Some researchers have devoted effort to considering virtual router migration forreducingtheimpactofplannedmaintenancesincedowntimeisaprimaryconcernforISPs. However, these approaches tend to perform the move whilst the router is“cold”, i.e. not actively transporting data traffic whilst the migration takes place.VROOM(VirtualROutersOntheMove),ontheotherhand,supportslivemigrationoflogical routers allowing them tomove among different physical platforms withoutcausing network discontinuities and instabilities [c19]. VROOMhas been tested onboth software and hardware. The results show that it does not interrupt the dataplaneandhasashortdowntimeperiodforthecontrolplane.Whennumberofroutesisabout15,000,themigrationtakesabout30secondsonasoftwaretestbedandlessthan 2 seconds on a hardware testbed. Fortunately, existing retransmissionmechanismswithin the routing protocols canmitigate the impact of control planedowntime.Nevertheless, the purpose of VROOM is to reduce the impact of plannedmaintenanceratherthantosaveenergy. Itdoesnotconsidertheeventsneededtotrigger the virtual router migration and the algorithm needed to determine theappropriatephysicaldestinationplatform for themigration; that is to say,when tomove the virtual router and how to choose the appropriate destination physicalrouter with the objective of saving energy. In response, a new dynamic energymanagementschemeemployinginfrastructuresleepingandvirtualroutermigrationhas been proposed [c128]. An alternative approach is considered in [c129]. It alsousesthevirtualmigrationconcepttomaintaintheIP-layertopologyunchanged.Thekeydifferenceisthatitmovesthefunctionalityofline-cardswithinaphysicalrouterinsteadofmovingtheentirevirtualrouterfunctionalitytoanotherphysicalplatform.3.6 Challenges Giventheuserandeconomicbenefitsofadopting“green”energy-savingmechanismsit is perhaps surprising the lack of uptake that has taken place. However, networkoperatorshavetobalancereliabilityagainstthepotentialcostsavings.Typically,foracarrier-gradenetwork,itisnecessarytorestoreservicedeliverywithin50ms.Evenforlessstringentservices,end-users(customers)areunlikelytotolerateoutagesofmorethan a few tens of seconds, particularly if this situation arises repeatedly. Inconsequence, operators design their networks with resilience, whereby variousbackup/alternativepathsexistbetweendifferentpartsofthenetwork.Intheeventof a failure, thepresenceof thesebackuppaths allows the traffic tobe redirectedalong them to the appropriate destination(s) soon after the failure is detected.

However,duringnormaloperatingconditions,thisequatestosparecapacity,whichisnonethelessconsumingresources.Operatorsarealsoaversetoembracingnewtechnologiesthatlackaclearmigrationpath from the existing deployed infrastructure and which may have an unproven“track-record”.Consequently, it is likelythattransparentenergy-savingtechnologieswillbeembraced,evenifthepotentialbenefitsaresomewhatmoremodest.Theseinclude the roll-out of more efficient routers that can interwork seamlessly withexistingequipment,thusprovidinga“plug-in”replacementasandwhenanupgradeis required. Other “straightforward” approaches include rate-adaptation wherebybothendsofapoint-to-point linkcansensethecapabilitiesoftheotherendand, ifsupported, adjust the transmission speed or related behaviour such as temporarysleepingtotakeplacewhentheloadconditionsaresuitable.Ofcourse,thereplacementofequipmentwithsimilardeviceswithabetter“EnergyStar” rating or dynamic link rate adjustment is only capable of delivering localisedimprovements.Betterenergysavingsarepossiblethroughtheglobalcoordinationofactivitiesacrossanetwork.Thisrequirestheadoptionofmoreradicalmechanisms,such as green routing and/or virtual router migration. However, at this time itremains unclear whether the added benefits will outweigh the required capitalexpenditure,particularly, ashasbeen shownmuch canbeachievedwithefficiencyimprovementswithinthenetworkingdevices,coupledwithtemporarysleepingandrateadaptation.3.7 Summary Designing energy efficient networking is a challenging task that must involveindustrial and academic researchers. Some joint projects like GreenTouch [c130]proposeto increasenetworkenergyefficiencybyafactorof1000atthehorizonof2015 compared to 2010 levels. Such aggressive approaches must deal with allnetworking levels from wireless to wired infrastructures and from core to accessequipment [c131]. Combined improvements in software and hardware must beexplored. Key research activities proposed in the area of energy-efficient wirednetworkshavebeendiscussedincludinganumberofcontributionsoftheEuropeanCOSTActionIC0804,however,researchinthisarearemainsverymuchongoing.

References [c1] J.Baliga,R.Ayre,K.Hinton,andR.Tucker,"Photonicswitchingandtheenergybottleneck,"

PhotonicsinSwitching,2007,pp.125-126.[c2] G.FettweisandE.Zimmermann,"ICTenergyconsumption-trendsandchallenges,"inProc.of

11thInt’l.Symp.WirelessPersonalMultimediaCommunications,2008.[c3] M.Pickavet,W.Vereecken,S.Demeyer,P.Audenaert,B.Vermeulen,C.Develderetal.,

“WorldwideEnergyNeedsforICT:TheRiseofPower-AwareNetworking,”2ndInt’l.Symp.ANTS’08,Dec.2008,pp.1-3.

[c4] R.Bolla,R.Bruschi,F.Davoli,andF.Cucchietti,"EnergyEfficiencyintheFutureInternet:ASurveyofExistingApproachesandTrendsinEnergy-AwareFixedNetworkInfrastructures,"IEEECommunicationsSurveys&Tutorials,2011,pp.1-22.

[c5] B.Talbot.(2009).Energypricerises.Available:http://www.energychoices.co.uk/partner-ec~cu/energy-price-rises-uk.html

[c6] GlobalAction.Plan(2007).“Aninefficienttruth”,.http://www.globalactionplan.org.uk/sites/gap/files/An%20Inefficient%20Truth%20-%20Executive%20Summary.pdf

[c7] N.E.F.a.1E.(2009).ThePCEnergyReport.Available:http://www.climatesaverscomputing.org/docs/1E_PC_Energy_Report_2009_US.pdf

[c8] S.JayashreeandC.S.RamMurthy,"ATaxonomyofEnergyManagementProtocolsforAdHocWirelessNetworks,"CommunicationsMagazine,IEEE,vol.45,pp.104-110,2007.

[c9] V.Prakash,B.Kumar,A.K.Srivastava,"Energyefficiencycomparisonofsometopology-basedandlocation-basedmobileadhocroutingprotocols,"Proceedingsofthe2011InternationalConferenceonCommunication,Computing;Security,Rourkela,Odisha,India,2011.

[c10] M.Deruyck,W.Vereecken,E.Tanghe,W.Joseph,M.Pickavet,L.Martens,andP.Demeester,"Powerconsumptioninwirelessaccessnetwork,"WirelessConference,2010,pp.924-931.

[c11] M.GuptaandS.Singh,"GreeningoftheInternet,"InProc.SIGCOMM’03,2003,pp.19-26.[c12] J.Baliga,K.Hinton,andR.S.Tucker,"EnergyconsumptionoftheInternet,"COIN-ACOFT07,

2007,pp.1-3.[c13] G.ShenandR.S.Tucker,"Energy-minimizeddesignforIPoverWDMnetworks,"Optical

CommunicationsandNetworking,IEEE/OSAJournalof,vol.1,pp.176-186,2009.[c14] L.Wang,R.Lu,Q.Li,X.Zheng,andH.Zhang,"EnergyEfficientDesignforMulti-shelfIPover

WDMNetworks."INFOCOM’11,WorkshoponGreenandCommunicationsandNetworking,2011

[c15] L.Chiaraviglio,M.Mellia,andF.Neri,"Reducingpowerconsumptioninbackbonenetworks,"ICC’09,2009,pp.1-6.

[c16] S.Nedevschi,L.Popa,G.Iannaccone,S.Ratnasamy,andD.Wetherall,"Reducingnetworkenergyconsumptionviasleepingandrate-adaptation,"inProc.NSDI'08,5thUSENIXSymposiumonNetworkedSystemsDesignandImplementation,2008,pp.323-336.

[c17] Schaffrath,C.Werle,P.Papadimitriou,A.Feldmann,R.Bless,A.Greenhalgh,A.Wundsam,M.Kind,O.Maennel,andL.Mathy,"Networkvirtualizationarchitecture:Proposalandinitialprototype,"InProc.VISA’09,2009,pp.63-72.

[c18] M.Agrawal,S.R.Bailey,A.Greenberg,J.Pastor,P.Sebos,S.Seshan,K.VanDerMerwe,andJ.Yates,"RouterFarm:Towardsadynamic,manageablenetworkedge,"InProc.SIGCOMMworkshop,INM’06,2006,pp.5-10.

[c19] Y.Wang,E.Keller,B.Biskeborn,J.vanderMerwe,andJ.Rexford,"Virtualroutersonthemove:liveroutermigrationasanetwork-managementprimitive,"InProc.SIGCOMM’082008,pp.231-242.

[c20] E.M.Noam,Telecommunicationsregulationtodayandtomorrow.NewYork:Law&Business,1983.

[c21] G.Bertram,RestructuringtheNewZealandElectricitySector1984-2005‟,inF.P.SioshansiandW.Pfaffenberger(eds.)(2006),ElectricityMarketReform:AnInternationalPerspective,Oxford:Elsevier.,203-234.

[c22] DECC,TheUKLowCarbonTransitionPlan,2009,availableathttp://www.decc.gov.uk/en/content/cms/publications/lc_trans_plan/lc_trans_plan.aspx

[c23] ITU,StandardITU-TL.1410“MethodologyforEnvironmentalImpactAssessmentofICTGoods,NetworksandServices”,2011.

[c24] Towardssmartpowernetworks,EuropeanCommission,2012,ISBN:92-79-00554-5[c25] L-F.Pau,"Thecommunicationsandinformationeconomy:issues,tariffsandeconomics

researchareas,"JournalofEconomicDynamicsandControl,vol.26,pp.1651-1675,August2002.

[c26] A.Mas-Colell,M.D.Whinston,andJerryR.Green,Microeconomictheory.NewYork,N.Y.,[etc.]:OxfordUniversityPress,1995.

[c27] E.Shafir,andR.A.LeBoeuf,"Rationality,"AnnualReviewofPsychology,2002,53(1),491-517.

[c28] H.A.Simon,"RationalDecisionMakinginBusinessOrganizations,"AmericanEconomicReview,1979,69(4),493-513.

[c29] A.Tversky,andD.Kahneman,"JudgmentunderUncertainty:HeuristicsandBiases,"in,D.KahnemanandP.SlovicandA.Tversky(Eds.),1974,NewYork:CambridgeUniversityPress.

[c30] D.Kahneman,andS.Frederick,"Representativenessrevisited:Attributesubstitutioninintuitivejudgment,"inHeuristics&Biases:ThePsychologyofIntuitiveJudgment,T.GilovichandD.GriffinandD.Kahneman(Eds.),2002,NewYork:CambridgeUniversityPress.

[c31] L-FPau,Greennetworksandgreentariffsasdrivenbyuserservicedemand,in:A.Hadjiantonis,B.Stiller(Eds),Telecommunicationseconomics,SpringerLectureNotesincomputerscience,Vol.7216,2012,DOI:10.1007/978-3-642-30382-1,117-125

[c32] S.Meij,andL-FPau,AuctioningbulkSMS/MMSmobilemessages,Computationaleconomics,Vol27,no2-3,April-May2006,395-430

[c33] F.Lévêque,Competitiveelectricitymarketsandsustainability.2006,NewYork:EdwardElgar[c34] H.Chen,L-FPau,IndividualtelecommunicationstariffsinChinesecommunities,in:Huang,

WangandDay(Eds),Globalmobilecommerce:strategies,implementationsandcases,IGIGlobal(formerlyIDEAGroup),September2007,ISBN978-1-59904-558-0,http://www.igi-global.com/reference/details.asp?id=6980

[c35] M.Castells,Theriseofthenetworksociety,Vol.1,2nd.Ed,2000,Oxford:Blackwell.[c36] H.Chen,L-FPau,MassCustomizationinWirelessCommunicationServices:Individual

ServicesandTariffs,ERIMResearchreportAug.2007,http://hdl.handle.net/1765/10515;and:Proc.Intl.Conf.onMasscustomization,Cambridge,Mass,October2008;andChapterinF.T.Piller&M.M.Tseng(Eds),Handbookofresearchinmasscustomizationandpersonalization(in2Volumes:Volume1:StrategiesandConceptsVolume2:ApplicationsandCases),WorldScientific,2010,http://www.worldscibooks.com/business/7378.html

[c37] L-FPau,Optimizingapublic3G/LTEwirelessnetworkandassociatedservicesforminimumenergyconsumptionoremissions:analysisfor“greenwirelesstariffs”Jean-MarcPierson,HelmutHlavacs(Ed.),ProceedingsoftheCOSTActionIC0804onEnergyEfficiencyinLargeScaleDistributedSystems,IRITPubl.,Toulouse,ISBN:978-2-917490-18-1-EAN:9782917490181,87-91,2011

[c38] L-FPau,EnergyconsumptioneffectsofWiFioff-loadingaccessin3GorLTEpublicwirelessnetworks,inSpecialIssue(SI)ongreennetworkingandcomputing,InternationalJournalofBusinessDataCommunicationsandNetworking,Vol9(2),pp.1-10,april-june2013ISSN:1548-0631

[c39] F.A.Hayek,Individualismandeconomicorder,1980,UniversityofChicagoPress.[c40] W.J.Adams,andJ.L.Yellen,"CommodityBundlingandtheBurdenofMonopoly,"The

QuarterlyJournalofEconomics,1976,90(3),475-98.[c41] Y.Bakos,andE.Brynjolfsson,"BundlingInformationGoods:Pricing,ProfitsandEfficiency,"

ManagementScience,1999,45(12),1613-30.[c42] A.C.Pigou,"DiscriminationMonopoly(PartII,ChapterXVII),"in:TheEconomicsofwelfare.

1920,London:MacmillanandCo.[c43] J.Aweya,"IProuterarchitectures:anoverview”,JournalofSystemsArchitecture,Elsevier,

Vol46,1999,pp:483—511.[c44] “RouterArchitectures”,Available:http://www.cs.virginia.edu/~cs458/slides/module09b-

routers.pdf

[c45] Ciscodatasheet,“CiscoCRS-116-SlotSingle-ShelfSystem,"Available:http://www.cisco.com/en/US/prod/collateral/routers/ps5763/ps5862/product_data_sheet09186a008022d5f3.pdf.

[c46] Ciscodatasheet,“CiscoCRS-124-SlotFabric-CardChassis,"Available:http://www.cisco.com/en/US/prod/collateral/routers/ps5763/ps5862/product_data_sheet0900aecd80340baa.pdf.

[c47] Ciscodatasheet,"CiscoCRS-1CarrierRoutingSystemmulti-shelfsystemdescription,"http://www.cisco.com.

[c48] S.HuangandR.Dutta,"Dynamictrafficgrooming:thechangingroleoftrafficgrooming,"CommunicationsSurveys&Tutorials,IEEE,vol.9,pp.32-50,2006.

[c49] J.Chabarek,J.Sommers,P.Barford,C.Estan,D.Tsiang,andS.Wright,"Powerawarenessinnetworkdesignandrouting,"INFOCOM2008,2008,pp.457-465.

[c50] Anne-CécileOrgerie,LaurentLefèvre,IsabelleGuérin-LassousanDinoLopezPacheco."ECOFEN:anEnd-to-endenergyCostmOdelandsimulatorForEvaluatingpowerconsumptioninlarge-scaleNetworks",Sustalnet2011:FirstInternationalWorkshoponSustainableInternetandInternetforSustainability,Lucca,Italy,June2011

[c51] RemiSharrock,ThierryMonteil,PatriciaStolf,OlivierBrun.AutonomiccomputingtomanagegreenCorenetworkswithQualityofService(regularpaper).Dans/In:EnergyEfficiencyinLargeScaleDistributedSystemsconference(EE-LSDS2013),Vienna,22/04/2013-24/04/2013,Springer,2013.

[c52] S.Nedevschi,L.Popa,G.Iannaccone,S.Ratnasamy,D.Wetherall,ReducingNetworkEnergyConsumptionviaSleepingandRate-Adaptation,USENIXNSDI2008,pp.323-336,SanFrancisco(USA),Apr.16-18,2008.

[c53] J.Rodgers,“EnergyEfficientEthernet:Technology,ApplicationandWhyYouShouldCare”,http://communities.intel.com/community/wired/blog/2011/05/05/.Accessed31/10/13.

[c54] B.Zhai,D.Blaauw,D.Sylvester,K.Flautner,TheoreticalandPracticalLimitsofDynamicVoltageScaling,DAC2004,pp.868-873,SanDiego(USA),June7-11,2004.

[c55] M.Weiser,B.Welch,A.Demers,S.Shenker,SchedulingforReducedCPUEnergy,USENIXOSDI1994,pp.13-23,Monterey(USA),Nov.14-17,1994.

[c56] N.Vasic,D.Kostic,Energy-AwareTrafficEngineering,e-Energy2010,pp.169-178,Passau(Germany),Apr.13-15,2010.

[c57] J.Chabarek,J.Sommers,P.Barford,C.Estan,D.Tsiang,S.Wright,PowerAwarenessinNetworkDesignandRouting,IEEEINFOCOM2008,pp.457-465,Phoenix(USA),Apr.13-18,2008.

[c58] L.Chiaraviglio,M.Mellia,F.Neri,ReducingPowerConsumptioninBackboneNetworks,IEEEICCWorkshops2009,pp.2298-2303,Dresden(Germany),June14-18,2009.

[c59] W.Fisher,M.Suchara,J.Rexford,GreeningBackboneNetworks:ReducingEnergyConsumptionbyShuttingOffCablesinBundledLinks,1stACMSIGCOMMWorkshoponGreenNetworking,pp.29-34,NewDelhi(India),Aug.30-Sept.3,2010.

[c60] A.P.Bianzino,C.Chaudet,F.Larrocca,D.Rossi,J.-L.Rougier,Energy-AwareRouting:aRealityCheck,3rdInternationalWorkshoponGreenCommunications(GreenComm2010)inconjunctionwiththeIEEEGlobalCommunicationsConference,pp.1422-1427,Miami(USA),Dec.6-10,2010.

[c61] S.Tibuleac,"ROADMnetworkdesignissues,"inOFC/NFOEC2009,paperNMD1.[c62] J.HomaandK.Bala,"ROADMarchitecturesandtheirenablingWSStechnology,"

CommunicationsMagazine,IEEE,vol.46,pp.150-154,2008.[c63] Morais,R.Metal,“ImpactofNodeArchitectureinthePowerConsumptionandFootprint

RequirementsofOpticalTransportNetworks”,IEEE/OSAJournalofOpticalCommunicationsandNetworking,Vol:5Issue:5,May2013,pp.421–436,DOI:10.1364/JOCN.5.000421

[c64] S.Bhaumiketal,“Energy-EfficientDesignandOptimizationofWirelineAccessNetworks”,TechnicalReportarXiv:1101.2717v1,January2011.URL:http://arxiv.org/pdf/1101.2717.pdf(Accessed31/10/13).

[c65] YingYan,LarsDittmann,“EnergyEfficiencyinEthernetPassiveOpticalNetworks(EPONs):ProtocolDesignandPerformanceEvaluation,JournalofCommunications,Vol.6,No.3,pp:249-261,May2011.

[c66] Cisco(2007),whitepaper,“ConvergeIPandDWDMLayersintheCoreNetwork”,http://www.webtorials.com/main/resource/papers/cisco/paper114/ConvergeIPandDWDMLayersintheCoreNetwork.pdf,URLaccessed31/10/13

[c67] Anne-CécileOrgerie,LaurentLefèvre,Jean-PatrickGelas,“ChasingGapsbetweenBursts:TowardsEnergyEfficientLargeScaleExperimentalGrids”,20089thInternationalConferenceonParallelandDistributedComputing,ApplicationsandTechnologies.

[c68] “AbileneTM”Availableat:http://www.cs.utexas.edu/~yzhang/research/AbileneTM/(accessed31/10/13)

[c69] GéantNetworkWebsite,http://www.geant.net,(accessed31/10/13).[c70] CPhillips,AGazo-Cervero,JGalan-Jimenez,“Pro-activeenergymanagementforWideArea

Networks”,IETInternationalConferenceonCommunicationTechnologyandApplication(ICCTA2011);01/2011

[c71] G.J.Eilenberger,S.Bunse,L.Dembeck,U.Gebhard,F.Ilchmann,W.Lautenschlaeger,andJ.Milbrandt,"EnergyefficienttransportforthefutureInternet,"BellLabstechnicaljournal,vol.15,pp.147-167,2010.

[c72] W.Hou,L.Guo,J.Cao,J.Wu,andL.Hao,"Greenmulticastgroomingbasedonopticalbypasstechnology,"OpticalFiberTechnology,2011.

[c73] S.Perrin(2010),HeavyReading,whitepaper,“TheNeedforNext-GenerationROADMNetworks”,Available:http://downloads.lightreading.com/wplib/heavyreading/NG_ROADM_WP_Final.pdf?p_redirone=yes

[c74] IEEEP802.3azEnergyEfficientEthernet,Available:http://www.ieee802.org/3/az/index.html[c75] Galan-Jimenez,J.,Gazo-Cervero,A.,“ELEE:EnergyLevels-EnergyEfficiencyTradeoffinWired

CommunicationNetworks”,IEEECommunicationsLetters,Volume:17,Issue:1,January2013,DOI:10.1109/LCOMM.2012.120312.122176

[c76] Gunaratne,C.,Christensen,K.,“EthernetAdaptiveLinkRate:SystemDesignandPerformanceEvaluation”,Proceedings200631stIEEEConferenceonLocalComputerNetworks,Nov.2006Page(s):28–35

[c77] BIANZINO,Arunaetal.:"ASurveyofGreenNetworkingResearch",IEEECommunicationsSurveys&Tutorials,vol.14,no.2,pp.3-20,FirstQuarter2012.

[c78] Puype,Bartetal.“PowerReductionTechniquesinMultilayerTrafficEngineering.”ICTON2009:11thInternationalConferenceonTransparentOpticalNetworks,Vols1and2.NewYork,NY,USA:IEEE,2009.41–44.

[c79] AlfredoParedes,AlbertoEGarcía:“EnergyefficiencyinnationalInternetnetworks”,MasterThesisinBusinessandInformationTechnologies.UniversityofCantabria,2012.URL:http://hdl.handle.net/10902/3804

[c80] SteveNRoy:“EnergyLogic:ARoadMaptoReducingEnergyConsumptioninTelecommunicationsNetworks”,IEEE30thInternationalTelecommunicationsEnergyConference,2008.INTELEC2008.

[c81] AlimianandB.NordmanandD.Kharitonov,“NetworkandTelecomEquipmentEnergyandPerformanceAssessment-TestProceduresandMeasurementMethodology,”,ECRInitiative,2008.

[c82] Anne-CécileOrgerie,LaurentLefèvre,“ERIDIS:Energy-EfficientReservationInfrastructureforLarge-ScaleDistributedSystems”,ParallelProcessingLetters,21:133-154,June2011.

[c83] Anne-CécileOrgerie,LaurentLefèvre,andIsabelleGuérin-Lassous."OntheEnergyEfficiencyofCentralizedandDecentralizedManagementforReservation-BasedNetworks",IEEEGlobalCommunicationsConference(GLOBECOM2011),Houston,USA,December2011

[c84] Anne-CécileOrgerieandLaurentLefèvre."Energy-EfficientOverlayforDataTransfersinPrivateNetworks",IEEEInternationalConferenceonNetworks(ICON2011),Singapore,December2011

[c85] L.Chiaraviglio,M.Mellia,andF.Neri,"Energy-awarenetworks:Reducingpowerconsumptionbyswitchingoffnetworkelements,"GTTI’08,2008.

[c86] R.Bolla,R.Bruschi,K.Christensen,F.Cucchietti,F.Davoli,andS.Singh,"ThePotentialImpactofGreenTechnologiesinNext-GenerationWirelineNetworks–IsThereRoomforEnergySavingOptimization?,"IEEECommunicationMagazine(COMMAG),2011.

[c87] F.Idzikowski,S.Orlowski,C.Raack,H.Woesner,andA.Wolisz,"SavingenergyinIP-over-WDMnetworksbyswitchingofflinecardsinlow-demandscenarios,"ConferenceonOpticalNetworkDesignandModeling,2010,pp.1-6.

[c88] Y.Zhang,M.Tornatore,P.Chowdhury,andB.Mukherjee,"Time-AwareEnergyConservationinIP-over-WDMNetworks,"PhotonicsinSwitching’10,2010.

[c89] W.Fisher,M.Suchara,andJ.Rexford,"Greeningbackbonenetworks:reducingenergyconsumptionbyshuttingoffcablesinbundledlinks,"GreenNetworking'10,2010,pp.29-34.

[c90] T.KellyandM.Adolph,"ITU-Tinitiativesonclimatechange,"CommunicationsMagazine,IEEE,vol.46,pp.108-114,2008.

[c91] “Aware,Inc.PatentStatementsandLicensingDeclarations,”,Availavle:http://www.atis.org/legal/Docs/PATENTS/ATIS/ATIS-0600007_LBE041_AWARE.pdf.

[c92] J.Baliga,R.Ayre,K.Hinton,W.V.Sorin,andR.S.Tucker,“EnergyconsumptioninopticalIPnetworks,”IEEE/OSAJLT,vol.27,no.13,2009.

[c93] L.Valcarenghi,D.PhamVan,P.G.Raponi,P.Castoldi,D.R.Campelo,S.-W.Wong,S.-H.Yen,L.Kazovsky,S.Yamashita,“EnergyEfficiencyinPassiveOpticalNetworks:Where,When,andHow?”,IEEENetwork;N.26;2012.

[c94] “10-Gigabit-capablepassiveopticalnetworks(XG-PON):Transmissionconvergence(TC)specifications,”ITU-TG.987.3Recommendation,2010.

[c95] R.Kubo,J.-I.Kani,Y.Fujimoto,N.Yoshimoto,andK.Kumozaki,“AdaptivePowerSavingMechanismfor10GigabitClassPONSystems,”IEICETransactionsonCommunications,vol.2,no.E93.B,pp.280–288,2010.

[c96] M.Marsan,A.Anta,V.Mancuso,B.Rengarajan,P.Vasallo,andG.Rizzo,"AsimpleanalyticalmodelforenergyefficientEthernet,"CommunicationsLetters,IEEE,vol.15,no.7,pp.773-775,July2011.

[c97] M.Fiammengo,A.Lindstrom,P.Monti,L.Wosinska,andB.Skubic,"Experimentalevaluationofcyclicsleepwithadapt-ablesleepperiodlengthforPON,"inOpticalCommunication(ECOC),201131thEuropeanConferenceandExhibitionon,sept.2011,pp.1-3.

[c98] L.Valcarenghi,M.Chincoli,P.Monti,L.Wosinska,andP.Castoldi,“EnergyEfficientPONswithServiceDelayGuarantees,”SustainableInternetandICTforSustainability(SustainIT),2012.

[c99] “NetworkperformanceobjectivesforIP-basedservices,”ITU-TRecommendationY.1541,Dec.2011.

[c100] Y.Chen,T.FarleyandN.Ye,“QoSRequirementsofNetworkApplicationsontheInternet,”IOSPress,4,pp.55-76,2004

[c101] D.He,L.Liang,andH.Cruickshank,“InternetQoSandtrafficmodelling,”inIEEProceedings-Software,Sept.2004.

[c102] “Yahoomessengerprotocolv9,”availableatlibyahoo2.sourceforge.net/ymsg-9.txt.[c103] P.Seeling,M.Reisslein,andB.Kulapala,“Networkperformanceevaluationusingframesize

andqualitytracesofsingle-layerandtwo-layervideo:Atutorial,”Commun.SurveysTuts.,vol.6,no.3,pp.58–78,Jul.2004.[Online].Available:http://dx.doi.org/10.1109/COMST.2004.5342293

[c104] “VoiceoverIP-percallbandwidthconsumption,”http://www.cisco.com/en/US/tech/tk652/tk698/technologies_tech_note09186a0080094ae2.shtml#topic1.

[c105] J.Baliga,R.Ayre,K.Hinton,W.V.Sorin,andR.S.Tucker,"EnergyconsumptioninopticalIPnetworks,"JournalofLightwaveTechnology,vol.27,pp.2391-2403,2009.

[c106] D.C.Dowden,R.D.Gitlin,andR.L.Martin,"Nextgenerationnetworks,"BellLabstechnicaljournal,vol.3,pp.3-14,1998.

[c107] G.L.RagsdaleandR.D.Lamm,"AdvancementsinPhotonicNetworkArchitectureMigration:TheEvolutionandDeploymentofMultiprotocolLabelSwitchingMPLS,Generalized

MultiprotocolLabelSwitchingGMPLS,andAdvancedOpticalSwitching,"ed:SouthwestResearchInstitute,OfficeoftheManagerNationalCommunicationsSystem,CommunicationTechnologies,Inc.USA,2002.

[c108] ArunaBianzinoetal,“GRiDA:aGreenDistributedAlgorithmforBackboneNetworks”,IEEEOnlineGreenCommunicationsConference,September2011.

[c109] J.Wangetal,“Green-AwareRoutinginGMPLSNetworks”,InternationalConferenceonComputing,NetworkingandCommunications(ICNC),page(s):227–231,2012.

[c110] SarahRuepp,AnnaManolovaFagertun,“EnergyEfficiencyEvaluationofRSVP-TEExtensionsforSurvivableTranslucentWSONNetworks”,DTUFotonik,TechnicalUniversityofDenmark,2012.

[c111] ChrisPhillipsetal,“Pro-activeenergymanagementforWideAreaNetworks”,IETInternationalConferenceonCommunicationTechnologyandApplications(ICCTA),page(s):317–322,2011.

[c112] SergioRicciardi,FrancescoPalmieri,UgoFiore,DavideCareglio,GermánSantos-Boada,JosepSolé-Pareta,"Anenergy-awaredynamicRWAframeworkfornext-generationwavelength-routednetworks",ComputerNetworks(Elsevier),Volume56,Issue10,5July2012,Pages:2420-2442,ISSN1389-1286,DOI:10.1016/j.comnet.2012.03.016.

[c113] SergioRicciardi,JiayuanWang,FrancescoPalmieri,DavideCareglio,AnnaManolova,GermánSantos-Boada,"Eco-sustainableRoutinginOpticalNetworks",PhotonicNetworkCommunications(Springer),2013,DOI10.1007/s11107-013-0416-0.

[c114] JiayuanWang,SergioRicciardi,AnnaV.Manolova,SarahRuepp,DavideCareglio,LarsDittmann,"OSPF-TEextensionsforgreenroutinginopticalnetworks,"Opto-ElectronicsandCommunicationsConference(OECC),201217th,vol.,no.,pp.411,412,2-6July2012.doi:10.1109/OECC.2012.6276497

[c115] JiayuanWang,SergioRicciardi,AnnaManolovaFagertun,SarahRuepp,DavideCareglio,LarsDittmann,"Energy-awareroutingoptimizationindynamicGMPLScontrolledopticalnetworks",14thInternationalConferenceonTransparentOpticalNetworks(ICTON),2012,pp.1-4,2-5July2012.doi:10.1109/ICTON.2012.6253843

[c116] JiayuanWang,SarahRuepp,AnnaVasilevaManolova,LarsDittmann,SergioRicciardi,DavideCareglio,"Green-awareroutinginGMPLSnetworks",InternationalConferenceonComputing,NetworkingandCommunications(ICNC),2012,pp.227-231,Jan.302012-Feb.22012.doi:10.1109/ICCNC.2012.6167416

[c117] Bozakov,Z.Anopenroutervirtualizationframeworkusingaprogrammableforwardingplane.SIGCOMMComput.Commun.Rev.,ACM,2010,40,439-440

[c118] Sherwood,R.;Gibb,G.;Yap,K.-K.;Appenzeller,G.;Casado,M.;McKeown,N.&Parulkar,G.FlowVisor:ANetworkVirtualizationLayer.OpenFlowSwitchConsortium,2009

[c119] Cisco(2008),whitepaper,“TechnicalOverviewofVirtualDeviceContexts,”Available:http://www.cisco.com/en/US/prod/collateral/switches/ps9441/ps9402/ps9512/White_Paper_Tech_Overview_Virtual_Device_Contexts.pdf

[c120] M.Kolon,"Intelligentlogicalrouterservice,"WhitePaper(JuniperNetworks,Inc.),2004.[c121] Clark,C.;Fraser,K.;Hand,S.;Hansen,J.G.;Jul,E.;Limpach,C.;Pratt,I.;Warfield,A.Live

migrationofvirtualmachines.InProceedingsofthe2ndconferenceonSymposiumonNetworkedSystemsDesign&Implementation-Volume2(NSDI'05),Vol.2.USENIXAssociation,Berkeley,CA,USA,273-286,2005.

[c123] Voorsluys,W.;Broberg,J.;Venugopal,S.&Buyya,R.CostofVirtualMachineLiveMigrationinClouds:APerformanceEvaluation.Proceedingsofthe1stInternationalConferenceonCloudComputing,Springer-Verlag,2009,254-265

[c122] Wang,Y.;Keller,E.;Biskeborn,B.;vanderMerwe,J.&Rexford,J.Virtualroutersonthemove:liveroutermigrationasanetwork-managementprimitive.SIGCOMMComput.Commun.Rev.,ACM,2008,38,231-242

[c124] Ramakrishnan,K.K.;Shenoy,P.&VanderMerwe,J.LivedatacentermigrationacrossWANs:arobustcooperativecontextawareapproach.Proceedingsofthe2007SIGCOMMworkshoponInternetnetworkmanagement,ACM,2007,262-267

[c125] Bradford,R.;Kotsovinos,E.;Feldmann,A.&Schiöberg,H.Livewide-areamigrationofvirtualmachinesincludinglocalpersistentstate.VEE'07:Proceedingsofthe3rdinternationalconferenceonVirtualexecutionenvironments,ACM,2007,169-179

[c126] A.Fischer,A.Fessi,G.CarleandH.DeMeer,Wide-AreaVirtualMachineMigrationasResilienceMechanism.Proc.oftheInt'lWorkshoponNetworkResilience:FromResearchtoPractice(WNR2011),page72—77,2011.

[c127] A.Fischer,J.F.Botero,M.T.Beck,H.DeMeerandX.Hesselbach,VirtualNetworkEmbedding:ASurvey.IEEECommunicationsSurveysandTutorials,2013.

[c128] XinChen,ChrisPhillips,“VirtualRouterMigrationandInfrastructureSleepingforEnergyManagementofIPoverWDMNetworks”,IEEEInternationalConferenceonTelecommunicationsandMultimedia(TEMU),pp:31-36,2012.

[c129] R.Bolla,R.Bruschi,andM.Listanti,"Enablingbackbonenetworkstosleep,"Network,IEEE,vol.25,pp.26-31,2011.

[c130] GreenTouchinitiative:http:/:greentouch.org[c131] Jean-PatrickGelas,LaurentLefevre,TeferiAssefaandMulugetaLibsie."VirtualizingHome

GatewaysforLargeScaleEnergyReductioninWirelineNetworks",ElectronicGoesGreen2012(EGG),Berlin,Germany,September2012

green wired networks - inria

Documents