Network Sharing and its Energy Benefits: a Study of European Mobile Network Operators

Transcription

1 Network Shring nd its Energy Benefits: Study of Europen Mobile Network Opertors Mrco Ajmone Mrsn Electronics nd Telecommunictions Dept Politecnico di Torino, nd Institute IMDEA Networks, ABSTRACT In this pper we investigte the potentil energy sving inherent in the network shring pproch, whereby ll (or significnt prts) of the network infrstructures existing in country cn be shred by different network opertors. In our study we consider Europen mobile network opertors, nd we use simple nlyticl models to show tht in most Europen countries the mount of energy necessry to run mobile networks cn be reduced by 35 to 6% with respect to the cse in which ech opertor mnges seprte network infrstructure. 1. INTRODUCTION The telecommuniction mrket in Europe ws dominted by monopolistic public compnies for bout century (in, the first experimentl telephone clls dte bck to 1877), until competition strted with mobile digitl telephone services (dopting the GSM 2G cellulr technology) in the mid 9s. Since then, competition hs been fierce for over decde; in those yers, ny form of coopertion mong mobile network opertors (MNOs) ws out of the question: the networks of different MNOs were completely seprted, s if physicl wlls existed round ech network. The interctions between networks, necessry to llow customers of one MNO to rech customers of nother, required specific gtewys. Then, some initil, limited crcks in the wlls were opened, mostly becuse of the difficulty in identifying good new ntenn sites, nd the concept of "mst shring" [1] emerged, through which different MNOs cn plce their ntenns on the sme pole. More recently, MNOs on the one hnd re under strong pressure for cost reduction, nd on the other hnd they re fced with n explosive growth of smrtphone users (predicted to rech hlf billion in Europe by 214), with corresponding exponentil increse of mobile dt trffic, growing t nnul rtes close to 8% worldwide (68% in Western Europe) nd forecsted to rech lmost 11 exbytes per month by 216 (see [3]). In these conditions, MNOs re strting to shre bse sttions (BSs), through concept often clled "tower shring" [2], nd the ide of "network shring" [4, Michel Meo Electronics nd Telecommunictions Dept Politecnico di Torino, michel.meo@polito.it 5], whereby ll (or significnt prts) of network infrstructure cn be shred by different MNOs, is not considered tody s sinful s it ws just few yers go. For exmple, recent nnouncement of n greement between Ornge nd T-Mobile sys tht their customers will be llowed to use either network interchngebly [6]. Of course, mny difficulties still exist in the pth to network shring, tht relte to both opertionl problems nd commercil sensitivity of informtion; for exmple: the high initil cost incurred to llow the shring of networks, due to the complexity of the control of severl prllel networks operted s pool the need for extended roming nd billing procedures to llow the semless trnsfer of services from one network to nother, nd to shre revenues between the involved MNOs the need for the definition of cost shring pproches for the introduction of new technologies (e.g., LTE) the difference in the QoS levels dopted by MNOs, nd the fct tht ech opertor tries to use its QoS level (in terms of both performnce nd coverge) s service differentitor, thus being reluctnt to trnsfer its customers to different network the possibility for competitors to profile MNO s customers nd to ttrct the most profitble ones In this pper we investigte the energy sving potentil of the network shring concept in severl Europen countries. Our erly ppers on energy efficiency in coopertive cellulr networks [7, 8] investigted the sme issue in n bstrct setting, by defining simple mthemticl models tht llow the quntifiction of the energy sving tht cn be chieved when cellulr network infrstructures in country cn be collectively mnged, so s to minimize energy consumption. In tht period, the time for the network shring concept hd not yet come, nd our study ws dismissed by industry insiders s nice theory, but techniclly nd commercilly nive, see [9]. Tody, MNOs re beginning to see the dvntges behind the network shring concept, nd network shring is becoming relity in some Europen countries, lthough with different motivtions from energy efficiency (minly reduction of cpitl costs). It seems thus importnt to revisit the energy efficiency spects of network shring, with specil reference to Europen countries, so s to provide t lest first rough indiction of the potentil energy svings inherent in this pproch.

2 In this pper we first summrize the pproch developed in [7, 8] nd we then pply tht methodology to the mix of MNOs existing in severl Europen countries, showing tht the energy sving possible with the infrstructures nd the equipment vilble tody is of the order of t lest 4%. This vlue is comprble to the energy sving potentil of BS sleep modes within n individul network, s shown for exmple in [11, 12, 13, 14, 15, 16, 17, 18], but the two pproches re not incomptible, so tht combintion of n internetwork pproch with n intr-network pproch cn led to even higher mounts of energy sved. In ddition, it should be considered tht the chrcteristics of the intr-network nd inter-network pproches re different, so tht they might be pplicble in different portions of the service re. An intr-network pproch might be more suited to dense urbn re, where cells re mny, nd highly redundnt in terms of coverge. An inter-network pproch might be more suitble for rurl re, where cells of one MNO re few, nd overlps re scrce. The rest of this pper is orgnized s follows. In Section 2 we introduce our definition for the network shring scenrio nd we summrize the pproch developed in [7, 8]. In Section 3 we present numericl results for Europen countries with more thn 15 million mobile subscriptions, nd finlly in Section 4 we present our conclusions nd directions for future work. 2. NETWORK SHARING We consider n re served by n MNOs, which operte seprte networks. Ech one of the n networks is exctly dimensioned ccording to the pek trffic demnd of the opertor s customers, so s to provide full coverge of the service re, while meeting fixed QoS constrint t ll times. In other words, we ssume tht ll MNOs provide equivlent coverge nd QoS. Note tht, for strters, we ssume tht t pek trffic no excess cpcity or overprovisioning exists. This is quite conservtive ssumption for our study, s we shll see lter on. Due to end user behvior (i.e., the combintion of user ctivity nd mobility ptterns), trffic fluctutes significntly during dy. For exmple, plots in Fig. 1 show the trffic mesured on cells of n Itlin mobile network in opertion; solid lines refer to cell in business re; the empty mrkers identify the profile of week-dy, solid mrkers refer to week-end dy. Trffic vlues re obtined by verging the mesurements (t 15 minute intervls) collected during week, nd re then normlized to the pek verge vlue in the cell. The steep growth of trffic in the morning of week dys corresponds to people rriving t work; trffic decreses from mid fternoon to evening, when people go home. At night, nd during weekends, trffic is extremely low, s is usul nd expected in business neighborhoods. Quite different behvior cn be observed in the sme figure for the trffic profile of cell in consumer re, shown with dshed lines. Peks occur now in the evening, differences between weekdys nd weekends re less significnt, nd trnsitions from pek to off-pek periods re slower. Obtining such rel trffic dt from MNOs is extremely problemtic, since they re considered sensitive informtion. We could not obtin equivlent dt for ll the Europen countries we will consider lter on, but we cn confidently ssume tht the generl behviours shown by the trffic profiles in Fig. 1 re representtive of ny business nd consumer re, irrespective of the Europen Normlized trffic, f(t)/f mx Business Consumer Week-dy Week-end 4: 8: 12: 16: 2: : Time, t [h] Figure 1: Dily trffic profiles for cell in business re nd cell in consumer re, week-dy nd week-end profiles mesured in network in opertion. country. The trffic profiles clerly indicte tht network which is exctly dimensioned to meet given QoS constrint t the pek trffic lod, offers cpcity which is underutilized for long periods of time, during which trffic is lower (possibly much lower) thn the pek. In terms of consumed energy, most networking devices, including BSs of mobile networks, tend to consume bout the sme quntity of energy, regrdless the mount of crried trffic; i.e., the consumption of device tht crries no trffic is lmost s lrge s the consumption t full lod. Due to this chrcteristic of networking devices, we cn sy tht networks consume power more for the deployed cpcity, thn for the used cpcity. The network shring concept cn therefore be used to reduce energy consumption. When n MNOs coexist in the sme service re, the underutiliztion of the ccess networks cpcity occurs for ll ccess networks roughly t the sme time, due to similr verge customer behviors. The network shring pproch llows MNOs to tke dvntge of this sitution nd sve energy, by modulting the ctive cpcity so s to follow the trffic demnd. The key ide underlying the energy efficiency of network shring is tht network cpcity supply cn be modulted by switching off some networks for the time periods in which trffic is low over the service re, so tht subset of the ccess networks is sufficient to provide the cpcity necessry to chieve the desired QoS. Of course, while the network of n MNO is off, its customers must be llowed to rom to the networks of the MNOs tht re ctive. Note tht the dely required to llow users to rom out of cell tht is bout to be switched off hs been nlysed, nd shown to be of the order of one minute [19], considering tht switch-offs occur in periods of low utiliztion, so tht the users tht must hndover out of the cell re few. Let N = {1, 2,, n} be the set of ccess networks of the n MNOs. Denote by S i the number of subscribers of opertor i, nd by f i(t), with t [, T ] spnning over T = 24 hours, the dily trffic profile of network i. We ssume tht the verge per-user trffic in ll ccess networks is the sme, so tht the overll trffic of ech network is proportionl to the respective number of users: f i(t) = α if(t) (1) with α i/α j = S i/s j. Assume the function f(t) to be continuous nd differentible, nd let f mx identify its mximum; α if mx is,

3 thus, the mximum trffic tht network i cn crry without violting the QoS constrint, under our ssumption of no overprovisioning. If some spre cpcity were vilble, sy tht there is n overprovisioning fctor (1 + x), then the mximum trffic tht network i could crry without violting the QoS constrint would become (1 + x)α if mx. With no loss in generlity, we consider α 1 > α 2 > α 3 >... > α n nd we tke α 1 = 1 so tht α i is the reltive number of subscribers tht opertor i hs with respect to the lrgest opertor, i.e., opertor 1. We ssume lso tht the function f(t) hs minimum, f min nd it is monotoniclly decresing from f mx to f min nd monotoniclly incresing from f min to f mx. This corresponds to n bstrction of wht hppens in relity, but is supported by the shpe of the mesured trffic profiles shown in Fig. 1. We ssume tht subset of the ccess networks cn be switched off when the totl trffic reduces to level such tht the networks tht remin on cn crry the entire trffic of ll networks without violting the QoS constrint. Users cn rom through ny network, nd when some networks re switched off, their customers rom to the networks tht remin on, with probbility proportionl to the network size. In [7], we clled Roming-to-All this roming scheme. Consider network switch-off configurtion in which the networks in the subset N N re powered on, while the remining networks re off. This configurtion is possible t time t, if: f(t) α i f mx α i (2) i N i N where the left side of the expression represents the totl trffic to be crried t time t, nd the right side is the mximum trffic tht the networks in N cn crry in totl, without violting the QoS constrint. Expression (2) defines the times during 24 hour period in which the configurtion is fesible. In prticulr, s indicted in Fig. 2, the extremes of the period in which the switch-off configurtion is fesible re T off f(t off ) i N T off, T on nd T on, given by: α i = f(t on ) α i = f mx α i (3) i N i N ( ) fmx = f 1 i N α i (4) i N αi with T off > T on. Obviously, T off cooresponds to negtive vlue of the derivtive of f(t), nd T on to positive derivtive vlue. Notice tht if the term in brckets, i.e., the rgument of f 1 ( ) tht is represented by the stright horizontl line in the figure, is smller thn f min, the minimum vlue of f(t), the network switch-off configurtion is not fesible without violting the QoS constrint. Also note tht, in the cse of fctor (1 + x) of overprovisioning in ll networks, the intervl in which the switch-off configurtion is fesible would hve extremes ( ) (1 + x)fmx T off, T on = f 1 i N α i (5) nd would thus become longer. i N αi 2.1 Energy cost Denote by W i the energy cost to operte network i, expressed s either the needed power, or corresponding monetry cost. In generl, W i is given by the sum of two terms: one which is constnt f(t) f mx f min T off configurtion NOT fesible T on configurtion fesible Figure 2: Sketch of trffic profile with indiction on how to derive the switching times. with respect to S i (the number of subscribers of MNO i), nd one which depends on S i. Indeed, the energy cost of the bckbone infrstructure nd of the ccess network devices tht provide complete rdio coverge re roughly independent of the number of subscribers; on the contrry, the number of dditionl devices needed to provide the necessry cpcity in the ccess network hevily depends on the number of users. For simplicity, we will lwys consider the two extreme cses in which W i is either constnt (this cse will be termed constnt cost), or directly proportionl to S i (this cse will be termed vrible cost). Thus, in the constnt cost cse we set W i = C irrespective of the number of subscribers of network i, nd in the vrible cost cse we set W i = cs i. The vlues of the constnts C nd c re rbitrry, since they re only used when compring network energy svings in the next section, nd in this cse they cncel wy. For given switch-off configurtion, such s the one previously considered, in which the networks in N N re powered on, while the remining networks re off, the dily energy cost cn be computed s: ( ) E = T T on + T off W i + T W i (6) i N \N i N since the networks in N re on ll the time, while the others re on for the time indicted in brckets. So fr, we hve ssumed tht the energy cost W i is independent of the mount of trffic ctully crried by network i, s long s network i is switched on. The energy cost drops to very low vlue when network i is switched off. This ssumption comes quite close to the chrcteristics of the presently instlled networking 2G nd 3G equipment [1]. New genertions of equipment (LTE, for exmple) exhibit better proportionlity of energy consumption to trffic, with bout 6% of the pek power consumption being fixed energy cost to hve the equipment on, nd the other hlf being proportionl to trffic [2]. To extend our nlysis to the cse of devices with some degree of lod proportionlity, in some cses we will ssume tht frction L P of the consumption is lod proportionl, mening tht under lod ρ i, with ρ i 1, the MNO i consumes W i(ρ i) = (1 L P )C + L P Cρ i (7) The ssumption of very low power consumption during the periods in which network is switched off is justified by the fct T

4 tht fst network rectivtions re not necessry, since the network switch-on time cn be scheduled in dvnce, bsed on historicl trffic trces. 2.2 Roming trffic When network i switches off, its trffic must rom to the networks tht re still powered on. The switch-off of network i t time T off nd its switch-on t time T on genertes dily roming trffic R i equl to: R i = T on T off α if(t)dt (8) This roming trffic is directed to the ctive networks in N. Assuming tht users rom to the networks in N proportionlly to the destintion network size, the dily trffic roming from network i to network j N is, α j R i,j = R i (9) k N α k 2.3 Switch-off Ptterns We focus now on different switch-off ptterns, indicting with this term the sequence ccording to which the networks re switched off, together with the switch-off nd switch-on instnts. We ssume tht ll but one networks switch off, in ech 24h period. A switch-off pttern P is thus defined by, {x i, i = 1,, n 1} with x i {1, 2,, n} the sequence tht specifies the order in which networks switch off; e.g., x i = k mens tht the i th network to switch off is network k. Denote by x n the network tht does not pper in the sequence nd tht never switches off in pttern P. {Ti off, i = 1,, n 1} with Ti off [, T ] the sequence of switch-off instnts, i.e., network x i switches off t time Ti off. {Ti on, i = 1,, n 1} with Ti on [, T ] nd Ti on > Ti off the sequence of switch-on instnts, i.e., network x i switches on t time T on i. The energy cost of switch-off pttern P, E P, cn be computed s: E P = n 1 i=1 ( T T on i ) + Ti off W xi + T W xn (1) from which sving is derived by normlizing C P over the energy cost of the lwys-on scenrio nd tking the complement: G P = 1 E P T n i=1 Wi (11) Note tht switch-off nd switch-on instnts cn be rther ccurtely determined by the nlysis of historicl trffic trces, which exhibit remrkble periodicity, dding mrgins to ccount for both unpredictble locl trffic vritions, nd trnsient delys. The fct tht network switch-on events cn be scheduled in dvnce, bsed on trffic predictions, justifies the ssumption of very low power consumption during the periods in which network is off. The level of precision with which the vilble cpcity is dpted to trffic fluctutions depends on the number of switch-on nd switchoff instnts during one dy. In this pper we will ssume tht ech Country MNOs Mrket shre [%] Subscr. [M] U.K Tble 1: Chrcteristics of the considered countries: Number of MNOs offering both 2G nd 3G services, mrket shre for ech of the MNOs, totl number of subscribers. network is t most switched off nd then on gin once dy. In [18] we proved tht, in the intr-network cse, one switch-off per dy is sufficient to obtin most of the possible energy sving. 3. ENERGY BENEFITS IN EUROPE In this section, we ssess the effectiveness of network shring in terms of chievble energy sving by considering number of Europen countries. In prticulr, we focus on the 12 countries indicted in Tble 1, which re the countries whose totl number of subscribers is lrger thn 15 Millions, ccording to publicly vilble dt. For ech country we collect pproximte dt bout the number of subscribers for ech of the ctive MNOs nd the kind of provided services. We then ssume tht network shring is pplicble only mong the MNOs tht offer both 2G nd 3G services. Indeed, MNO offering ccess to 2G terminls only cnnot switch off its network nd mke the users rom to purely 3G network. In this cse the opertor would probbly switch off the 3G network leving the 2G ccess network on; however, since we only hve ccess to dt bout the totl number of subscribers nd not the brekdown with respect to technology, we mke the simplistic ssumption tht network shring is implemented only mong MNOs offering services to both 2G nd 3G users. Interestingly, the considered Europen countries present quite similr scenrios. As summrized in Tble 1, except for two cses, nmely nd, ll considered countries hve 3 MNOs offering both 2G nd 3G services with reltively fir shre of mrket. The smllest of the 3 MNOs hs shre tht is usully between 2 nd 3%, only in the cse of nd the smllest of the three opertors ccounts for s low s 15% of the subscribers. Conversely, the lrgest of the 3 MNOs exceeds 5% of the mrket shre only in, where it is bout 51%; otherwise, it is between 37% nd 48%. The cse of, with 4 MNOs, is interesting becuse it presents two dominnt opertors with bout the sme number of subscribers, 36 millions, corresponding to 31% of the mrket, nd other two smller MNOs tht shre the remining mrket. In, three opertors re bout the sme size, with lmost 3% of the shre ech, while the fourth opertor ccounts for 14% of the mrket only. This substntil similrity of the situtions is probbly due to historicl resons: in most of the Europen countries similr network evolutions occurred roughly t the sme

5 Reltive sving consumer, const consumer, vr business, const business, vr consumer- const consumer- vr business- const business- vr.1 Figure 4: Kivit digrm of the sving chievble with network shring in the Europen countries with more thn 15M subscribers; business nd consumer profiles, constnt nd vrible cost models. Figure 3: Sving chievble with network shring in the Europen countries with more thn 15M subscribers; business nd consumer profiles, constnt nd vrible cost models. time. We compute the energy sving chievble through network shring for ech of the selected countries, nd for both the consumer nd the business trffic profiles shown in Fig. 1 (implicitly ssuming tht the trffic profiles in Fig. 1 cn be representtive of trffic in ll considered countries). Given trffic profile nd country, we consider ll the possible switch-off ptterns, i.e., ll the possible orderings in which the MNOs of tht country might switch off. Svings re obtined s described in the previous section, by deriving switch-off nd switch-on instnts from (4), nd by computing sving from (11). Both the cses of vrible nd constnt cost models re evluted. The sving chievble during week-dys nd week-ends re properly weighted to get the verge weekly sving. Figs. 3 nd 4 report the mximum chievble energy sving, mong those obtined from different switch-off ptterns in given scenrio. Fig. 3 uses br representtion, while Fig. 4 reports the sme dt with Kivit digrm. The svings re relly significnt, typiclly lrger thn 4%: this confirms tht network shring, besides being vible pproch, lredy fesible with tody technology, is very promising in terms of energy consumption reduction. Observe lso from the figure tht the business trffic profile leds to the lrgest sving. This is due to the profile hving prticulrly steep trnsitions between pek nd off-pek, nd long periods of very low trffic. Clerly, in relity, lrge service res re chrcterized by mixture of neighborhoods, some minly with business-like behvior of the users nd others with consumer-like trffic profiles. A switch-off scheme should then be pplied by dpting, neighborhood by neighborhood, switching times to the specific profiles. For exmple, MNO tht is going to switch-off its ccess network, might probbly strt from portions of the network in business res, s soon s trffic drops below some threshold; some time lter, when trffic drops lso in the consumer res, other portions of the ccess network would be powered off. In terms of sving, this mens tht the chievble sving will be in be- tween wht cn be obtined from business re nd consumer re, with ctul vlues depending on the trffic profiles nd on the proportions of res with business-like or consumer-like behvior. In cse of some spre cpcity, deployed to bsorb medium term trffic growth, some dditionl sving cn be expected. With n overprovisioning fctor 1 + x = 1.2, for exmple, it is possible to rech svings between 5 nd 59% for the consumer profile nd between 53 nd 63% for the business profile under the constnt cost model. These vlues re even closer to the mximum theoreticl sving tht would be chieved when one network only hs enough cpcity to crry ll the trffic; the mximum theoreticl sving is equl to 66% for 3 MNOs, corresponsing to 1 network over three tht is crrying trffic, nd it is equl to 75% for 4 MNOs. A positive side-effect of network shring is tht ctive resources re more effectively used thn in trditionl scenrios without shring. Indeed, network shring ims t reducing energy wstge tht derives from dily periods of over-provisioning by mking the vilble cpcity more closely follow the trffic profile. To evlute this effect, we compute the dily verge utiliztion of the ccess network resources, by dividing the mount of generted trffic by the mount of vilble cpcity. The results re reported in Fig. 5, for both business nd consumer profiles nd distinguishing weekdys from week-ends. When no network shring is used (first group of brs in the figure), the utiliztion is bout.38 for week-dys under both trffic profiles nd it is.7 nd 3 for week-ends, respectively, in business nd consumer res. When network shring is implemented, the verge utiliztion increses to bout.6: it lmost doubles. Network shring turns out to be relly effective. Notice tht week-ends in business res still present reltively low resource utiliztion; this is due to the fct tht trffic is so little tht even one network lone serving ll the trffic of the 3 or 4 coexisting MNOs is sort of under-utilized. It is worth noting tht increses in network utiliztion re specilly welcome in periods of reduced opertionl mrgins, like the one we re living. MNOs perceive s one of the most criticl spects of network shring the fct tht, by hving the opportunity to serve roming users, competitor MNO might profile subscribers. We, thus, compute the mount of outgoing roming trffic tht MNO genertes once it powers off its ccess network. Focusing on the Itlin cse, Fig. 6

6 Utiliztion consumer, WD consumer, WE business, WD business, WE Roming trffic.5.3 From MNO 1 From MNO 2 From MNO 3 Sving.1 No shring Figure 5: Utiliztion chievble with network shring in the Europen countries with more thn 15M subscribers; business nd consumer profiles, week-dys nd week-ends. reports the mount of roming trffic in vrious possible switch-off ptterns. Ptterns re denoted by pir of vlues (i, j) tht indictes tht the ccess network of MNO i is the first one to be powered off nd it is followed by the network of MNO j. The lbel B or C ssocited to ptterns indictes the trffic profile (business or consumer). Roming trffic is normlized with respect to the totl trffic of the MNO. For completeness, the figure reports lso the sving chieved by ech switch-off pttern under the constnt cost model (see the dshed curve). The mount of roming trffic cn be pretty lrge, up to 4-5% of the trffic of MNO. However, due to the reltively similr size of the MNOs, different ptterns correspond to similr percentge of roming trffic nd chieve more or less the sme sving. This mens tht, to reciprocte the inconvenience of roming trffic to competitor, MNOs might estblish schemes in which switch-off ptterns lternte periodiclly. So fr, we ssumed tht the energy consumption of the network is independent of the crried trffic. This ssumption is justified by the fct tht tody most of the instlled network devices, both the BSs t the ccess network, nd the switches nd routers in metro nd core networks, consume t full lod bout the sme mount of power tht is consumed when they re ctive, but crry no trffic. Newer equipment (for exmple, LTE BSs) shows better proportionlity between power consumption nd lod. Clerly, the energy sving chieved with network shring reduces, when network devices exhibit n incresing lod proportionlity (nd would completely vnish in the cse of perfect lod proportionlity). In order to ssess this energy sving reduction, we look now t the cse in which the network power consumption is lod proportionl for frction of power consumption expressed by the prmeter L P, s in (7). The energy svings chievble with network shring for different vlues of L P re reported in Figs. 7 nd 8 for the Europen countries considered in this pper. Also in this cse, Fig. 7 uses br representtion, while Fig. 8 reports the sme dt with Kivit digrm. While the energy svings decrese with incresing vlues of L P, 1,2 - C 2,1 - C 1,3 - C 3,1 - C 2,3 - C Figure 6: Roming trffic out of MNO with vrious switching ptterns nd for both business nd consumer trffic; Itlin scenrio. we cn observe tht the bsolute vlues of the svings remin lrge, even for L P =. Notice tht, s we lredy sid, in rel networks the ctul vlue of L P is very low tody, since most of the network devices re not lod proportionl, nd only very recent equipoment (like LTE BSs) cn chieve, individully, vlue of L P round. 4. CONCLUSIONS In this pper we hve quntified the energy svings which could be chieved by mobile network opertors offering service in the lrgest Europen countries, s result of widespred doption of the network shring pproch. Our results indicte tht bout hlf of the energy cost presently incurred by opertors could be voided by cleverly exploiting the fct tht most Europen countries re tody covered by severl (3-4) overlpping cellulr network infrstructures. These svings re ctully chievble thnks to the presence of prllel cellulr networks, which thus constitute significnt sset for the identifiction of energy-efficient solutions. If only one network were vilble, with cpcity corresponding to the sum of the cpcities of ll networks of tody, nd competition would rely on virtul opertors exploiting the sme infrstructure, the pproch discussed in this pper would not be fesible; it should, more effectively, be replced by energy-efficient mngement pproches within the only vilble infrstructure. In other words, the fct tht the totl vilble ccess network cpcity is frctioned in severl prllel infrstructures, llows simple pproches for the improvement of the proportionlity between energy consumption nd overll trffic lod. Acknowledgement The reserch leding to these results hs received funding from the Europen Union Seventh Frmework Progrmme (FP7/27-213) under grnt greement n (Network of Excellence TREND - Towrds Rel Energy-efficient Network Design). 5. REFERENCES 3,2 - C 1,2 - B 2,1 - B 1,3 - B 3,1 - B 2,3 - B 3,2 - B

7 Reltive sving LP= LP=.3 LP= LP= LP= LP=.3 LP= LP=.1.1 Figure 8: Kivit digrm of the sving chievble with network shring in the Europen countries with more thn 15M subscribers when frction L P of energy consumption is lod proportionl; consumer trffic profile. Figure 7: Sving chievble with network shring in the Europen countries with more thn 15M subscribers when frction L P of energy consumption is lod proportionl; consumer trffic profile. [1] [2] [3] /ns525/ns537/ns75/ns827/white_pper_c pdf. [4] M.F. Hossin, K.S. Munsinghe, A. Jmlipour, Two level coopertion for energy efficiency in multi-ran cellulr network environment, IEEE Wireless Communictions nd Networking Conference, WCNC, Pris,, April 212. [5] T. Li, L. Bi, Model of Wireless Telecommunictions Network Infrstructure Shring & Benefit-Cost Anlysis, Interntionl Conference on Informtion Mngement, Innovtion Mngement nd Industril Engineering, Shenzhen, Chin, November 211. [6] [7] M. Ajmone Mrsn, M. Meo, Energy efficient wireless Internet ccess with coopertive cellulr networks, Computer Networks, Volume 55, Issue 2, 1 Februry 211, Pges 386âĂŞ398. [8] M. Ajmone Mrsn, M. Meo, Energy efficient mngement of two cellulr ccess networks, SIGMETRICS Performnce Evlution Review, Volume 37 Issue 4, Mrch 21, Pges [9] P. Mrks, Coopertive cellphone networks could cut energy wste, New Scientist, No. 2796, Jnury 211. [1] J. Lorincz, T. Grm, G. Petrovic, Mesurements nd Modelling of Bse Sttion Power Consumption under Rel Trffic Lods, Sensors 212, Volume 12, Issue 4, Pges [11] M. Ajmone Mrsn, L. Chirviglio, D. Ciullo, M. Meo, "Energy-Efficient Mngement of UMTS Access Networks", ITC 21-21st Interntionl Teletrffic Congress, Pris,, September 29. [12] M. Ajmone Mrsn, L. Chirviglio, D. Ciullo, M. Meo, "Optiml Energy Svings in Cellulr Access Networks", GreenComm 9-1st Interntionl Workshop on Green Communictions, Dresden,, June 29. [13] L.i Sker, S. E. Elyoubi, "Sleep mode implementtion issues in green bse sttions", PIMRC 21, Istnbul, Turkey, September 21. [14] Z. Niu, Y. Wu, J. Gong, Z. Yng, "Cell zooming for cost-efficient green cellulr networks", IEEE Communiction Mgzine, Volume 48, Issue 11, Pges 74-79, November 21. [15] E. Oh, B. Krishnmchri, X. Liu, Z. Niu, "Towrd dynmic energy efficient opertion of cellulr network infrstructure", IEEE Communiction mgzine, Volume 49, Issue 6, Pges 56-61, June 211. [16] C. Peng, S. B. Lee, S. Lu, H. Luo, H. Li, "Trffic-Driven Power Sving in Opertionl 3G Cellulr Networks", ACM MobiCom 211, Ls Vegs, Nevd, USA, September 211. [17] K. Son, H. Kim, Y. Yi, B. Krishnmchri, "Bse Sttion Opertion nd User Assocition Mechnisms for Energy-Dely Trdeoffs in Green Cellulr Networks", IEEE Journl on Selected Are in Communictions: Specil Issue on Energy-Efficient Wireless Communictions, Volume 29, Issue 8, Pges , September 211. [18] M. Ajmone Mrsn, L. Chirviglio, D. Ciullo, M. Meo, "Multiple Dily Bse Sttion Switch-Offs in Cellulr Networks", Fourth Interntionl Conference on Communictions nd Electronics (ICCE 12), Hue, Vietnm, August 212. [19] M. Ajmone Mrsn, L. Chirviglio, D. Ciullo, M. Meo, "Switch-Off Trnsients in Cellulr Access Networks with Sleep Modes", GreenComm 4, Kyoto, Jpn, June 211. [2] G. Auer et l., "How much Energy is needed to run Wireless Network?", IEEE Wireless Communictions Mgzine, Volume 18, Issue 5, Pges 4-49, October 211.