Auction-based Resource Allocation in UMTS High Speed Downlink Packet Access (HSDPA)

Transcription

1 Aucton-based Resource Allocaton n UMTS Hgh Speed Downlnk Packet Access (HSDPA) Manos Dramtnos, George D. Stamouls and Costas Courcoubets Network Economcs and Servces Group (N.E.S.), Department of Informatcs, Athens Unversty of Economcs and Busness 76 Patson Str. Athens, GR 10434, Greece Emal: {mdramt, gstamoul, courcou}@aueb.gr Abstract Data, audo and vdeo servces over 3G networks possess certan requrements n QoS and thus n resources. Accommodatng such servce requests consttutes a challenge for UMTS networks. To ths end, 3GPP Release 5 ntroduces the Hgh Speed Downlnk Packet Access (HSDPA). In ths paper, we propose: ) an aucton-based HSDPA resource allocaton mechansm, whch conssts of a seres of Generalzed Vckrey Auctons that are conducted n the short tme-scale over whch the network resources are allocated and ) a set of nnovatve user utlty functons. These functons provde a quantfcaton of both the user s wllngness to pay for reservng resources n the short tme-scale auctons and hs attaned utlty from the pattern of resources allocated durng hs long tme-scale servce sesson. The utlty functons are addtve and are used as bddng functons n the seres of auctons the user partcpates. Each user selects one of these functons, whch s scaled by hs total wllngness to pay. Then, the network runs the auctons by bddng on behalf of each user. The effectveness of our mechansm has been assessed expermentally and t appears that most of the users ether are served very satsfactorly or are allocated very lmted quanttes of resources (f at all) at a low total charge. Index Terms Auctons, resource reservaton, UMTS, HSDPA. I. INTRODUCTION AUCTIONS have recently receved consderable attenton as an economc mechansm for resource reservaton and prce dscovery n networks. Ths s of partcular mportance for many practcal cases such as the provson of network servces wth long duraton and hgh demand for network resources. A promnent case s that of UMTS [1], [2]: except for voce, the duraton of servces that users generally request s sgnfcantly longer than a sngle tme slot 1 over whch resource unts can be reserved; e.g., large fle transfers or vdeo streamng of a certan bt rate. These servces requre Qualty of Servce (QoS), and thus are very demandng 1 For UMTS, the tme slot s taken to be the 10msec UTRAN frame. n resources. Accommodatng such servces conssts a challenge for 3G networks due to the scarcty of rado spectrum. To ths end, 3GPP Release 5 defnes the Hgh Speed Downlnk Packet Access (HSDPA) wth a maxmum bt rate of 10Mbps, a sgnfcant mprovement compared to the 2Mbps of the 3GPP Release 99 [3]. Snce the moble ndustry s extremely compettve and spectrum s a scarce resource, effcent explotaton of spectrum s crucal for commercal provders. In ths paper, we propose an aucton-based Resource Allocaton mechansm for HSDPA (HERA). Ths mechansm comprses: (a) a seres of repeated Generalzed Vckrey Auctons, one per slot, and (b) certan bddng functons for the user to choose from. HERA s an extenson of the ATHENA mechansm whch we had proposed n [4] for resource reservaton n 3GPP Release 99 UMTS networks. In partcular, here we extend and adapt the approach of [4] n order to be applcable wth HSDPA. We deal wth how to satsfy the demand for large tme scale servces of constant bt rate (CBR), by allocatng nearly consstently resources to those users that value them the most. We also provde an extenson for varable bt rate (VBR) servces. We assume that the populaton of users generally vares over tme, as n practcal cases, whch further complcates the problem. Our objectve s to allocate resources effcently. That s, award the network resources to the users who value them the most whle provdng proper ncentves for ratonal network usage. Note that demand fluctuatons make t mpossble for a provder to publsh fxed prces for the resources, whch would have been the way to attan effcency under steady demand. As already mentoned, we propose a seres of consecutve auctons as a means for attanng effcently consstent reservaton of resources. Each aucton spans a tme perod of 1 sec and ts outcome specfes the allocaton of 500 2msec HSDPA frames (see below). Each user ams n beng allocated a certan quantty of frames wthn the 1 sec perod so that hs target

2 Watch lve concert Read news UMTS Base Staton New Vdeo-Clp Request (vdeoname, Type2, 0.5, hgh qualty) Dsconnect Receve stock market data Download MMS clparts gallery Fg. 1. A user of a UMTS cell submts a vdeo servce request consstng of the vdeo name, hs selecton of a utlty functon and hs total wllngness to pay for that servce of a certan qualty (bt rate). bt rate s attaned. Snce constant resource allocaton among auctons may not be feasble for all users, t s very mportant to construct meanngful user utlty functons that actually express users preferences regardng the degradaton of the servce qualty. These utlty functons are addtve, although not memoryless. Hence, they can be used as bddng functons n our mechansm, thus provdng a quantfcaton of users preferences n cases of nconsstent resource allocaton patterns. Each user selects one of these functons, whch s scaled by hs maxmum total wllngness to pay, whch s to be gven by the user hmself (as part of hs servce request). Then, the network runs the varous auctons by bddng on behalf of each user, accordng to hs respectve selecton of bddng functon. Ths way, a user demandng a servce just declares once ths selecton as well as hs wllngness to pay for the servce desred, as shown n Fg. 1. The defnton of these functons and the fact that users have to choose one of them consttutes an ntegral part of our mechansm for nearly consstent resource reservaton. The effectveness of our approach has been assessed expermentally by analyzng the resultng resource allocaton patterns. It s seen that users wth compettve bds are allocated patterns that are n very good accordance to ther respectve preferences, whle users wth non-compettve bds are allocated very lmted quanttes of resources (f at all) at low charges. As already mentoned, HERA s an extenson of the ATHENA mechansm whch we had proposed n [4] for resource reservaton n 3GPP Release 99 UMTS networks. In ths paper, we extend and adapt ATHENA n order to be applcable wth HSDPA. The new features ntroduced and the modfcatons concern the tme-scale of the aucton, the capacty of the downlnk shared channel, the defnton of the resource unts auctoned, and the technologcal means for mappng servce requests to resource reservatons. Moreover, n HERA we also provde a new user utlty functon pertanng to varable bt rate (VBR) servces. Both the smulaton model and the software employed for the evaluaton of the mechansm have been enhanced accordngly. We have run numerous new experments nvolvng a varety of mxes of competng CBR and VBR users servce sessons under varous network congeston levels,.e. low, medum and hgh demand. An alternatve to performng auctons sequentally (one per slot) would be to run combnatoral auctons, thus allowng users to bd for resources spannng several consecutve slots. Such an approach should not be adopted due to the followng reasons: ) the computatonal overhead s excessve, ) guaranteeng resources for the entre duraton of the sessons currently served would render mpossble for the network to serve users wth hgher value that would arrve late, thus resultng n hgh connectvty delay, poor revenue and neffcency. The remander of ths paper s organzed as follows: In Secton II, we provde a bref background on auctons. In Secton III, we present UMTS HSDPA and ts man features. In Secton IV, we defne the proposed auctonbased mechansm for the HSDPA. In Secton V, we defne user utlty functons capable to express certan types of user preferences wth respect to patterns of nconsstent resource allocaton. mechansm. In Secton VI, we present certan extensons of our utlty functons. In Secton VII we provde an expermental assessment of our aucton. Fnally, n Secton VIII, we provde some concludng remarks. II. BACKGROUND ON AUCTIONS In ths secton, we present fundamental defntons, theorems and results from aucton theory; see [5], [6]. An aucton s a mechansm based on a par of rules, namely the allocaton rule that defnes whch good s allocated to whom and the payment rule that defnes the charge of the aucton wnner(s). A partcpant of an aucton s called bdder whle the entty conductng the aucton auctoneer. There are many ways to classfy the varous aucton mechansms. Thus, auctons are referred to as smple or sngle-unt f only one good s auctoned and mult-unt f multple unts of a good (e.g. ntegral unts of a lnk s bandwdth) are to be traded. Moreover, dependng on whether bds are made n publc or submtted as sealed envelopes, the aucton s referred to as open or sealed respectvely. Auctons that maxmze seller s revenue are referred to as optmal and those that maxmze

3 socal welfare are referred to as effcent. Iftheaucton s conducted n rounds, then t s called progressve. Dependng on the payment rule, the aucton can be ether unform f the same amount of money s pad by all for each unt of the good awarded and dscrmnatory (payyour-bd) f each user pays hs bd for each object he wns. A. Smple auctons A bd n the context of smple auctons s the amount of money offered by a bdder for the tem auctoned. The best-known mechansm s by far the Englsh aucton, where the seller starts wth a mnmum prce that s actually ncremented untl there s only one person clamng the tem, whom the tem s awarded to. The Dutch aucton s exactly the opposte mechansm. The prce s ntally hgh and s gradually decremented untl a bdder clams the object. The tem s awarded to hm for a charge equal to the current prce. The sealed bd auctons (1 st prce and 2 nd prce or Vckrey) consst of two phases: a) the frst one, where bdders submt sealed envelopes wth ther bds and b) the second one, where these envelopes are opened. The tem s then awarded to the bdder who submtted the hghest bd. The wnner pays hs bd at the 1 st prce aucton and the hghest losng bd, -.e. the 2 nd hghest bd - at the Vckrey aucton. It has been proved that under the Vckrey aucton t s best for each bdder to honestly bd hs true value for the tem beng awarded. Ths property s referred to as ncentve compatblty. B. Mult-unt auctons A bd n the context of mult-unt auctons s defned to be the par (p, q) of the per unt expressed wllngness to pay p for a quantty q of unts. All smple auctons can be generalzed to mult-unt auctons. Incentve compatblty holds only for the generalzatons of the Vckrey aucton. The rules of the Generalzed Vckrey Aucton (GVA) prescrbe that: ) each user reports hs valuaton for a subset or for all ponts of hs demand functon for unts of the good auctoned, ) unts are allocated to the hghest bds untl demand exhausts supply, ) each user s charged accordng to the socal opportunty cost that hs presence entals. Hence, each user pays for the unts he s awarded the losng bds that would have become wnnng f all hs own bds were set to zero. Hence, wnners pay less than ther respectve bds. III. HSDPA HSDPA was fnalzed n the 3GPP Release 5 specfcaton and s consdered the most sgnfcant radorelated update snce the release of the frst 3GPP Release 99 specfcaton. It s based on a dstrbuted archtecture where processng s closer to the ar nterface at the base staton (Node B). The HSDPA operaton s performed on the Hgh Speed Downlnk Shared Channel (HS- DSCH). Ths s superor to the Downlnk Shared Channel (DSCH) of the orgnal 3GPP Release 1999 n terms of performance, effcent usage of rado spectrum and target bt rates. HSDPA extends bt rates up to 10Mbps, a sgnfcant upgrade compared to the 2Mbps of the prevous release 2. Ths results from a set of unque addtonal features of HS-DSCH, namely: ) the very short frame length (only 2msec) ) the fxed spreadng factor of 16 wth maxmum 15 codes and ) the support of 16QAM modulaton n addton to QPSK of DSCH. The number of HS-DSCH codes n each cell s confgured by the Rado Network Controller (RNC). The HSDPA system archtecture s depcted n Fg. 2. Fg. 2. The HSDPA system archtecture and ts nnovatve features compared to UMTS Release 99. Each user s allocated a bt rate of the HS-DSCH whch depends on the number of codes and frames that have been assgned to hm. The HS-DSCH frame can be confgured to contan ether 1 or 15 codes. Therefore, several packet users can be code- and tme-multplexed. However, t s possble that the sharng of the HS-DSCH among users s done manly n the tme doman by the HSDPA scheduler; ths s consdered n [8] as the most preferable use of the channel. Hence, due to the very short tme duraton of the HS-DSCH frame, t s desrable that one user s awarded exclusvely the HS- DSCH wthn the 2msec tme scale. 2 The capacty of the WCDMA sector can be typcally mproved va base staton beamformng that allows to reduce nterference levels by transmttng the sgnal va a narrow beam to the desred user [7].

4 IV. HSDPA AUCTION-BASED RESOURCE ALLOCATION The HSDPA scheduler s responsble for awardng access to the HS-DSCH. We propose that the assgnment of resources,.e. the 2-msec HS-DSCH frames, be performed by means of auctons. We conduct a sequence of mn-auctons, each concernng reservaton of resources wthn 1 sec; hence 500 2msec frames are auctoned. Each mn-aucton s a sealed-bd Generalzed Vckrey Aucton (GVA) [5], wth atomc bds (.e., bds that are ether fully satsfed or rejected) of the type (p, q), where q s the quantty of resource unts sought n the present aucton and p s the prce proposed for each such unt. Henceforth, unless otherwse specfed, we restrct attenton to servces requrng constant bt rates. For UMTS, snce resources are allocated n bts, f the servce nvolves traffc of a specfc bt rate m, then we have q = m t a,wheret a s the tme perod per mnaucton. (For each such servce sesson, one bd s placed per mn-aucton; see below.) In a GVA, users have the ncentve to bd truthfully ther wllngness to pay. Ths s a very attractve property, motvatng our selecton of runnng a GVA n each mn-aucton. The tme perod t a, was set to be 1 sec, so that t s both (a) small enough to capture market fluctuatons due to the entry or departure of users from the network, and (b) large enough to accommodate varous bt rate requests. The mnmum bt rate that can be awarded to a user wthn 1 sec - by awardng just one frame per second - s very small (approxmately 20 Kbps or 1,33 Kbps, f the HS-DSCH frame s confgured to contan 1 or 15 codes respectvely), thus allowng for a very fne granularty n the set of achevable bt rates. Ths way, the exclusve use of the Hgh Speed Downlnk Shared CHannel (suggested n [8]) s also possble, as long as the RNC confguraton defnes that the number of HS- DSCH codes per frame s set to 1. The mappng of users bds to actual network resource reservaton s carred out arbtrarly wthout havng to resort to elaborate technques for the allocaton of the varous frames wthn the 1 sec perod. For example, a user awarded a bt rate of 5Mbps wthn the auctoned 1 sec, actually uses the HS-DSCH exclusvely for 250 frames of 2 msec. The frames used by the same sesson wthn two consecutve perods may be at most 2 secs apart, whch may affect jtter adversely. Ths problem can be taken care of, f user termnals are equpped wth a playout buffer to smooth the data flow and bound jtter. The necessary buffer capacty s of the order of magntude of MBytes,.e. capable of storng data correspondng to a tme duraton of few seconds. Ths requrement s reasonable wth the present technology for devces. Another soluton would be to spread the frames awarded wthn the 1 sec perod unformly n tme so that jtter s mantaned low. The formal defnton of such an algorthm s left for future research. In a realstc case of a UMTS network, t would not be feasble for users to partcpate n all these mnauctons. Hence, we defne utlty functons, pertanng to the varous servces. These functons are provded by the network operator as bddng functons for the user to choose from; they are scaled by the user s total wllngness to pay U s, for the servce s, whch s to be gven by the user hmself (as part of hs servce request). Ths approach s smlar to that of 3GPP regardng predefned Qualty of Servce (QoS) profles. Then, the network runs the varous mn-auctons by bddng on behalf of each user, accordng to hs respectve selecton of bddng functon. V. USER UTILITY FUNCTIONS Below, we brefly present the utlty functons and ther propertes. We assume that the user s value u s, for obtanng the servce s s the sum of the margnal subutltes attaned due to each successful allocaton s, (.e. the allocated bt rate x equals the desred m ); thus, u s, (x (1),..., x (Ks,) ) = K s, t=1 v(t) s, (x(1) ). (Indces and s denote the user and servce respectvely; K s, denotes the number of mn-auctons for user and servce sesson s; superscrpt t ranges from 1 to K s,, ndcatng the current mn-aucton.) Next, we defne meanngful utlty functons, pertanng to the varous servces. These functons reflect the fact that, when there are gaps n the resource allocaton pattern, not only the amount of slots but also the way these are allocated -.e. the resource allocaton pattern - makes consderable dfference to the degree of user satsfacton. Thus, by selectng one of the predefned user utlty functons, each user declares hs preferred form of allocaton pattern for the cases where perfectly consstent resource reservaton s not possble. We have defned the followng utlty functons: Type 1: Indfferent to the allocaton pattern. Ths type pertans to volume-orented users, such as those downloadng news. The utlty attaned depends only on the quantty of resources allocated, as opposed to the allocaton pattern. Hence the user s total value U s, s equally apportoned among the varous slots. That s, s, (x(1) )=1(x (t) = m ) Us, K s,, where U s, s the user s declared total wllngnessto-pay and 1( ) s the ndcator functon. There-

5 fore, the lack of nformaton transfer due to the gaps n the resource allocaton pattern, results n proportonal loss of user satsfacton. As already mentoned, an approprate example of users belongng to ths type s those accessng news wth a certan target bt rate. Also, users downloadng nformaton by means of FTP can be consdered as users of ths type, provded that the FTP sesson s capable of resumng download when tme-outs occur. Ths utlty functon s sutable for the UMTS Background Class servces wth m pertanng to the Maxmum Bt-rate parameter of ths class [9]. Note that snce users are volume-orented, a rate less than m would also be useful to them. That s, f a user has the last wnnng bd and he can be awarded just a fracton of the target bt rate, t s both meanngful and effcent to do so. Ths can be attaned f the assumpton of atomc bds s relaxed. On the contrary, t s obvously neffcent to assgn to the hgher value users a bt rate less than ther maxmum. Type 2: Senstve to the servce contnuty. Ths type pertans (among other cases) to users that prefer watchng consstently half of a football match rather than watchng multple shorter perods. Thus, they prefer the allocaton pattern of Fg. 3(a) to that of Fg. 3(b). A sub-utlty expressng ths preference s s, (x(1) )=1(x (t) = m ) Us, K s, α d (t), where α (0, 1) s a dscount factor, d (t) s the dstance between the current and the prevous slots durng whch user acheved reservatons. Hstory of prevous allocatons nfluences s, through the value of d, whch s kept track of. Ths utlty functon s sutable for the UMTS Streamng Class servces [9]. Type 3: Senstve to the regularty of the resource allocaton pattern. Such users prefer the allocaton pattern of Fg. 3(b) to that of Fg. 3(a). The correspondng sub-utlty s s, (x(1) )=1(x (t) = m ) Us, K s, α max{0, d}, where d s the dfference of d (t) (as defned above) and the length of the prevous gap. Note that the factor α max{0, d} equals 1 (resp. s less than 1) f d s non-postve,.e. when the receved qualty of servce mproves (resp. deterorates). For all three types, f the user s constantly allocated resources (and thus the best possble qualty s acheved), then the utlty obtaned s U s,. The aforementoned utlty functons are not the only possble ones reflectng the user satsfacton w.r.t. the Fg. 3. Inconsstent resource allocaton patterns. allocaton pattern attaned. What s mportant, s that the form of these utltes reflects rghtly the preferences of each type of user. VI. EXTENSIONS OF THE USER UTILITY FUNCTIONS We have extended the defntons of these utlty functons so as to cover more nterestng cases. For nstance, user may be wllng to watch just good qualty vdeo - of a rate r good - whenever watchng the vdeo wth the preferred hgh qualty r hgh s not feasble. Watchng the vdeo wth consstently ether good qualty or hgh qualty results n dfferent degrees of user satsfacton; hence, the total wllngness to pay respectvely equals U good, and U hgh, = U good, + U where U expresses the user s extra satsfacton for vdeo of hgh qualty. Due to the possble fluctuatons of the rate attaned, a proper utlty functon for ths type of users s s, (x(1) )=1(x (t) r good ) U good, K s, + 1(x (t) α d (t),1 = r hgh ) U α d (t),2, K s, where d (t),1 (respectvely d(t),2 ) s the dstance between the current and the prevous slots durng whch user acheved a reservaton of at least r good (resp. exactly r hgh ). Fnally, note that the number of atomc bds to be gven on behalf of each user at each mn-aucton equals the number of alternatve qualty levels. Thus, for the aforementoned case of two such levels, two summable bds should be gven per user: one expressng hs wllngness to pay for the basc rate r good and the other expressng hs extra wllngness to pay for the extra rate r hgh r good. Of course, n the smple case of a sngle qualty level that was dscussed n the prevous paragraphs, only one bd s to be submtted per user n each mn-aucton. Ths generalzaton s very useful n the context of 3G networks where herarchcal encodng of streamng vdeo s common [10]. Under ths approach, streamng vdeo s coded n herarchcal streams of predefned qualty. The bottom stream s the vdeo s base layer,

6 whch contans all the basc vdeo frames; when ths layer s receved at the user termnals, t enables users to watch the vdeo at a very low qualty. The addtonal streams provde the enhancement layers that mprove the vdeo s qualty. Ths way, termnals wth dfferent QoS capabltes can be supported wthout wastng bandwdth, snce only the requred layers are transmtted over the network. Moreover, when the Rado Access Network (RAN) s congested, some enhancement layers can be dropped but the users contnue recevng vdeo servce (though the qualty s much poorer). Ths s reflected n the aforementoned utlty functon that defnes a dfferent user wllngness to pay for recevng servce under varous QoS levels. Next, we ntroduce a new utlty functon for Varable Bt Rate servces, n cases where a predefned traffc pattern must be obtaned by partcpatng n the aucton. Ths functon expresses the utlty attaned by computng the devaton of the allocated pattern from the desred: s, (x(1) ) = x(t) m Us, K s, α mgap,/ m, where: m (t) s the target bt-rate at the varous tme slots; m = 1 Ks, K s, t=1 m(t) s the mean rate of user s target resource allocaton pattern; and m gap, s the sum of the user s target bt-rates at the tme-slots of the ongong gap, that s the consecutve zero-allocaton slots snce the last slot τ<twhere due to an allocaton, or to the fact m (τ) =0,wehavex (τ) = m (τ). We have argued n [4] that a user whose preferences are accurately expressed by one of the predefned functons, has the ncentve to truthfully declare both ths functon and hs maxmum wllngness-to-pay U s,.furthermore, the charge reflects the actual market demand and proper ncentves for ratonal network usage are provded to the users: Recevng servce s expensve n perods of hgh network load, whle t s nexpensve when the network s under-utlzed. VII. EXPERIMENTAL ASSESSMENT We have extended the smulaton software of ATHENA [4] so that the HERA aucton s smulated. We have run numerous smulaton experments accordng to a detaled smulaton model, specfyng: ) the dstrbutons of user arrvals, departures, and servce requests, and ) the mx of users n terms of the number of users per type and the dstrbuton of ther total wllngness to pay. The total quantty of resource unts avalable at each mn-aucton also fluctuates n the smulaton model thus mtatng the varyng allocaton of resources to telephone calls and SMS/MMS n a real network. Fnally, n each experment, the value of the dscount factor α n user utlty functons was n the range [0.95, 0.995]. The quanttes demanded by users are n accordance wth the demand of servces as specfed n IST project B-BONE [10]. Hence hgh qualty audo streamng sessons demand Kbps, vdeo streamng sessons demand 384 Kbps, telemetry montorng requres 28 Kbps and non-rt VBR sessons are smulated as perodc on-off sources havng a peak rate of 100Kbps. A sample resource allocaton pattern of an on-off source s provded n Fg. 4. Fg. 4. Screenshot depcts the aggregate demand for resources (topmost bars), the total avalable resource unts (lne), and the allocaton pattern (lowest bars) of an on-off source under low competton n the aucton. A sample resource allocaton pattern of a Type 2 user (watchng vdeo streamng) s depcted n Fg. 5, whle hs bds n the auctons he partcpated are provded n Fg. 6. A sample resource allocaton pattern of a Type 3 user s depcted n Fg. 7, whle hs bds n the auctons he partcpated are provded n Fg. 8. It s seen that the usage of the aforementoned utlty functons as bddng functons n the aucton, results n meanngful resource allocaton patterns for the users. Users wth compettve bds are served very satsfactorly whle they are allocated patterns that are n very good accordance to ther respectve preferences. On the other hand, users wth non-compettve bds are allocated very lmted quanttes of resources (f at all) at a low total charge. Intermedate allocaton patterns n general arse rarely. (Such patterns are those that are not nearly consstent and yet comprse a non-neglgble percentage

7 Fg. 5. Screenshot depcts the aggregate demand for resources (topmost bars), the total avalable resource unts (lne), and the allocaton pattern (lowest bars) of a Type 2 sesson under hgh competton n the aucton. Fg. 7. Screenshot depcts the aggregate demand for resources (topmost bars), the total avalable resource unts (lne), and the allocaton pattern (lowest bars) of a Type 3 sesson under hgh competton n the aucton. Fg. 6. Screenshot depcts the user s bds (topmost lne) and the aucton s cutoff prce (fluctuatng lne) under hgh competton n the aucton. of the total quantty of resources that would arse for the same user n the case of consstent allocaton.) A typcal dstrbuton of allocaton patterns under hgh demand s depcted n Fg. 9. The total percentage of users beng allocated ntermedate patterns s very low, due to the effectveness of the aforementoned utlty functons. Indeed, ntermedate patterns mostly arse n cases of low market demand, and ther assocated charge s typcally very low. Also, the hgher the level of market demand, the larger the percentage of users allocated almost no resources, as depcted by the leftmost bar of Fg. 9. Ths s also depcted n Fg. 10 where the classfcaton of the user resource allocaton patterns s provded n cases of very hgh competton n the aucton. Hence, our mechansm serves as soft Call Admsson Control mechansm, whch takes nto account both the needs and the economc means of the varous contendng Fg. 8. Screenshot depcts the user s bds (topmost lne) and the aucton s cutoff prce (fluctuatng lne) under hgh competton n the aucton.

8 users. Overall, our approach can serve as the resource allocaton logc of the HSDPA scheduler n UMTS networks. bddng for many short tme scale network resources. We have assessed expermentally the effectveness of our mechansm and t appears that most of the users ether are served very satsfactorly or essentally are not served at all. IX. ACKNOWLEDGEMENTS Ths work was partly funded by EU project FP6-IST BBONE [10]. The authors wsh to thank BBONE and EURONGI partners for useful dscussons on the subject of ths paper. REFERENCES Fg. 9. Dstrbuton of resource allocaton patterns wth respect to the percentage of the total targeted resources that was attaned under hgh demand. Fg. 10. Dstrbuton of resource allocaton patterns wth respect to the percentage of the total targeted resources that was attaned under very hgh demand. VIII. CONCLUSIONS In ths paper we presented HERA, an effectve aucton-based mechansm for the UMTS Hgh Speed Downlnk Packet Access (HSDPA). The mechansm can deal successfully wth long tme-scale data, audo and vdeo servces n practcal cases of networks wth large numbers of competng users. It s based on a seres of Generalzed Vckrey Auctons and a set of user utlty functons expressng the user s preferences wth respect to the resource allocaton patterns. The addtve form of these functons enable ther usage as bddng functons n the seres of the auctons each user partcpates. Hence, we have developed sutable bddng functons that enable users to construct meanngful resource allocaton patterns on the long tme scale of user servce by [1] H. Holma and A. Toskala, WCDMA for UMTS: Rado Access for Thrd Generaton Moble Communcatons, John Wley, ISBN [2] 3GPP, URL: [3] H. Holma and A. Toskala, Thrd Generaton WCDMA rado evoluton, Wreless Communcatons and Moble Computng, 3, , [4] M. Dramtnos, G. D. Stamouls, C. Courcoubets, Auctonbased Resource Reservaton n 2.5/3G Networks, Kluwer/ACM Moble Networks and Applcatons [MONE] specal ssue, accepted 2003, to appear, Vol. 10, No. 2, Aprl Also avalable at [5] C. Courcoubets and R. Weber, Prcng Communcaton Networks : Economcs, Technology and Modellng, John Wley & Sons, 2003, ISBN [6] V. Krshna, Aucton Theory, Academc Press, Aprl 2002, ISBN X. [7] J. Ramro-Moreno, K.I. Pedersen, P.E. Mogensen, Capacty gan of beamformng technques n a WCDMA system under channelzaton code constrants, IEEE Transactons on Wreless Communcatons, 3(4), , July [8] M. Frodgh, S. Parkvall, C. Roobol, P. Johansson and P. Larsson, Future-Generaton Wreless Networks, IEEE Personal Communcatons, 8(5), 10-17, October [9] 3GPP, Techncal Specfcaton Group Servces and System Aspects, QoS Concept (3G TR verson 5.6.0), URL: [10] FP6-IST Project B-BONE, Broadcastng and multcastng over enhanced UMTS moble broadband networks, URL: