Multicast in UMTS: Evaluation and recommendations

Transcription

1 WIRELESS COMMUNICATIONS AND MOBILE COMPUTING Wrel. Commun. Mob. Comput. 2008; 8: Publshed onlne 10 October 2006 n Wley InterScence ( Multcast n UMTS: Evaluaton and recommendatons Antonos Alexou and Chrstos Bouras *, Computer Engneerng and Informatcs Department, Unversty of Patras, Research Academc Computer Technology Insttute, Patras, Greece Summary It s known that multcastng s an effcent method of supportng group communcaton as t allows the transmsson of packets to multple destnatons usng fewer network resources. Along wth the wdespread deployment of the thrd generaton cellular networks, the fast-mprovng capabltes of the moble devces, content and servce provders are ncreasngly nterested n supportng multcast communcatons over wreless networks and n partcular over Unversal Moble Telecommuncatons System (UMTS). Multcastng s a more effcent method of supportng group communcaton than uncastng or broadcastng, as t allows transmsson and routng of packets to multple destnatons usng fewer network resources. In ths paper, the three above mentoned methods of supportng group communcaton are analyzed n terms of ther performance. The crtcal parameters of prmary nterest for the evaluaton of any method are the packet delvery cost and the scalablty of the method. Copyrght 2006 John Wley & Sons, Ltd. KEY WORDS: UMTS; multcast n UMTS; MBMS; multple uncast; broadcast 1. Introducton Unversal Moble Telecommuncatons System (UMTS) consttutes the thrd generaton of cellular wreless networks whch ams to provde hgh-speed data access along wth real tme voce calls. Wreless data s one of the major boosters of wreless communcatons and one of the man motvatons of the next generaton standards [1]. Multcast communcatons for wrelne users has been deployed n the Internet for at least the past 10 years. The multcast transmsson of real tme multmeda data s an mportant component of many current and future emergng Internet applcatons, such as vdeoconference, dstance learnng, and vdeo dstr- buton. The multcast mechansm offers effcent multdestnaton delvery, snce data s transmtted n an optmal manner wth mnmal packet duplcaton [2]. Although UMTS networks offer hgh capacty, the expected demand wll certanly overcome the avalable resources. Thus, the multcast transmsson over the UMTS networks consttutes a challenge and an area of research. Actually, the adopton of multcast routng over moble networks poses a dfferent set of challenges n comparson wth multcastng over the Internet. Frst of all, multcast recevers are non-statonary, and consequently, they may change ther access pont at any tme. Second, moble networks are generally based on a welldefned tree topology wth the non-statonary multcast recevers beng located at the leaves of the network tree. *Correspondence to: Chrstos Bouras, Computer Engneerng and Informatcs Department, Unversty of Patras, Research Academc Computer Technology Insttute, N. Kazantzak str, GR 26500, Patras, Greece. E-mal: bouras@ct.gr Copyrght 2006 John Wley & Sons, Ltd.

2 464 A. ALEXIOU AND C. BOURAS The constructon of a source-rooted shortest-path tree over such a topology s trval and may be acheved by transmttng only a sngle packet over the paths that are shared by several multcast recpents. However, as a result of user moblty, there are several cases where ths smplfed vew of the moble network s volated [3]. It s, therefore, not approprate to apply IP multcast routng mechansms n UMTS, snce they are not desgned to take nto account the need for moblty management that moble networks requre. In ths paper, we present an overvew of three dfferent one-to-many packet delvery schemes for UMTS. These schemes nclude the Broadcast, the Multple Uncast, and the Multcast scheme. We analytcally present these schemes and analyze ther performance n terms of the packet delvery cost and the scalablty of each scheme. Furthermore, for the evaluaton of the schemes, we consder dfferent transport channels for the transmsson of the data over the Iub and Uu nterfaces. Snce the performance of these schemes depends manly on the confguraton of the UMTS network that s under nvestgaton, we consder dfferent network topologes and user dstrbutons. Ths paper s structured as follows. In Secton 2, we provde an overvew of the UMTS n packet swtched doman. Secton 3 s dedcated to descrbng the related work, whle Secton 4 presents the Multmeda Broadcast/Multcast Servce framework of UMTS. In Secton 5, we present a number of alternatve one-tomany packet delvery schemes for UMTS. Followng ths, Secton 6 analyzes the dfferent delvery schemes n terms of telecommuncaton costs, whle Secton 7 presents some numercal results that characterze the above schemes. Fnally, some concludng remarks and planned next steps are brefly descrbed. 2. Overvew of the UMTS n the Packet Swtched Doman UMTS network s splt n two man domans: the User Equpment (UE) doman and the Publc Land Moble Network (PLMN) doman. The UE doman conssts of the equpment employed by the user to access the UMTS servces. The PLMN doman conssts of two land-based nfrastructures: the Core Network (CN) and the UMTS Terrestral Rado-Access Network (UTRAN) (Fgure 1). The CN s responsble for swtchng/routng voce and data connectons, whle the UTRAN handles all rado-related functonaltes. The CN s logcally dvded nto two servce domans: the Crcut-Swtched (CS) servce doman and the Packet-Swtched (PS) servce doman [1,4]. The PS porton of the CN n UMTS conssts of two knds of General Packet Rado Servce (GPRS) Support Nodes (GSNs), namely Gateway GSN (GGSN) and Servng GSN (SGSN) (Fgure 1). SGSN s the centerpece of the PS doman. It provdes routng functonalty, nteracts wth databases (lke Home Locaton Regster (HLR)) and manages many Rado Network Controllers (RNCs). SGSN s connected to GGSN va the Gn nterface and to RNCs va the Iu nterface. GGSN provdes the nterconnecton of UMTS network (through the Broadcast Multcast Servce Center) wth other Packet Data Networks (PDNs) lke the Internet [1]. UTRAN conssts of two knds of nodes: the frst s the RNC and the second s the Node B. Node B consttutes the base staton and provdes rado coverage to one or more cells (Fgure 1). Node B s connected to the UE va the Uu nterface (based on the Wdeband Code Dvson Multple Access, W-CDMA technology) and Fg. 1. UMTS and MBMS Archtecture. Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

3 MULTICAST IN UMTS 465 to the RNC va the Iub nterface. One RNC wth all the connected to t Node Bs s called Rado Network Subsystem (RNS). Before a UE can exchange data wth an external PDN, t must frstly establsh a vrtual connecton wth ths PDN. Once the UE s known to the network, packets are transferred between t and the network, based on the Packet Data Protocol (PDP), the network-layer protocol carred by UMTS. An nstance of a PDP type s called a PDP context and contans all the parameters descrbng the characterstcs of the connecton to an external network by means of end-pont addresses and Qualty of Servce (QoS). A PDP context s establshed for all applcaton traffc sourced from and destned for one IP address. A PDP context actvaton s a request reply procedure between a UE and the GGSN. A successful context actvaton leads to the creaton of two GPRS Tunnelng Protocol (GTP) sessons, specfc to the subscrber: between the GGSN and SGSN over the Gn nterface and between the SGSN and RNC over the Iu nterface. IP packets destned for an applcaton usng a partcular PDP context are augmented wth UE- and PDP-specfc felds and are tunneled usng GTP to the approprate SGSN. The SGSN recovers the IP packets, queres the approprate PDP context based on the UE- and PDP-specfc felds, and forwards the packets to the approprate RNC. The RNC mantans Rado-Access Bearer (RAB) contexts. Equvalently to PDP contexts, a RAB context allows the RNC to resolve the subscrber dentty assocated wth a GTPtunneled network packet data unt. The RNC recovers the GTP-tunneled packet and forwards the packet to the approprate Node B. Fnally, a Tunnel Endpont IDentfer s used across the Gn and Iu nterfaces to dentfy a tunnel endpont at the recevng network node [4]. In the UMTS PS doman, the cells are grouped nto Routng Areas (RAs), whle the cells n a RA are further grouped nto UTRAN Regstraton Areas (URAs). The Moblty-Management actvtes for a UE are characterzed by two fnte state machnes: the Moblty Management (MM) and the Rado Resource Control (RRC). The Packet MM (PMM) state machne for the UMTS PS doman s executed between the SGSN and the UE for CN-level trackng, whle the RRC state machne s executed between the UTRAN and the UE for UTRAN-level trackng. After the UE s attached to the PS servce doman, the PMM state machne s n one of the two states: PMM dle and PMM connected. In the RRC state machne, there are three states: RRC dle mode, RRC cell-connected mode, and RRC URA connected mode [5]. 3. Related Work Several multcast mechansms for UMTS have been proposed n the lterature. In Reference [6], the authors dscuss the use of commonly deployed IP multcast protocols n UMTS networks. Three potental Internet multcast archtectures are analyzed. The frst s the exstng multcast archtecture that s standardzed as an optonal feature n the UMTS networks. In ths archtecture, the IP multcast routng protocol s termnated n the gateway between the Internet and the UMTS network. Ths soluton requres few multcast aware UMTS nodes. However, ths archtecture does not provde any bandwdth savngs n the UMTS network. The two other desgns are Internet multcast archtectures where the multcast functonalty s pushed successvely further out towards the UMTS termnal. These two archtectures requre multcast awareness from an ncreased number of UMTS network nodes. Hgher complexty s ntroduced to acheve network resource savngs. The presented multcast mechansm employs the Internet Group Management Protocol (IGMP) for group management and reles on the standard herarchcal tunnelng of UMTS for dstrbutng multcast packets to the group. The herarchcal tunnelng mechansm of UMTS, however, does not lend tself to effcent multcast packet delvery, snce each tunnel may only be establshed for a sngle subscrber. Consderng a group of N multcast users, a sngle multcast packet must be duplcated and transmtted N tmes throughout the network n order to reach all the destnatons. Dependng on the dstrbuton of the multcast users wthn the coverage area, ths may lead to an neffcent usage of resources wthn the network. A soluton to the above descrbed problem s presented n Reference [3]. The authors, n order to overcome the one-to-one relatonshp between a sngle subscrber and a GPRS Tunnelng Protocol (GTP) tunnel that s nherent to the herarchcal routng n UMTS, mplement a Multcast-Packet Data Protocol (M-PDP) context for each multcast group n the GGSN and SGSN. In ths approach, the authors do not adopt the use of IP multcast protocols for multcast routng n UMTS and present an alternatve soluton. For multcast group management, the authors propose the ntroducton of a number of new tables n GGSN, SGSN, and RNC, whle for multcast packet forwardng some trval changes n the GTP are requred. In Reference [7], a multcast mechansm for CS GSM networks s outlned that only sends multcast messages to Locaton Areas (LAs) n whch multcast users resde. Ths mechansm uses the exsted Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

4 466 A. ALEXIOU AND C. BOURAS UMTS/GSM short message archtecture n order to perform multcast routng. In partcular, two new tables are consdered n the Home Locaton Regster (HLR) and n the Vstor Locaton Regster (VLR). The multcast table at the HLR records the Moble Swtchng Centers (MSCs) that serve multcast users, whle the VLR keeps track of the LAs that have multcast users. However, the multcast messages are delvered to all the cells of an LA, ndependently of whether or not multcast users are located n all cells. Ths s neffcent f an LA s large or only sparngly populated wth multcast users. Fnally, the Multmeda Broadcast/Multcast Servce (MBMS) framework of UMTS s currently beng standardzed by the thrd Generaton Partnershp Project (3GPP) [8]. MBMS reles on the defnton of servce areas for delverng multmeda traffc to subscrbers. The MBMS framework s presented n detal n the followng secton. 4. Multmeda Broadcast/Multcast Servce n UMTS 3GPP s currently standardzng the MBMS. Actually, the MBMS s an IP datacast type of servce, whch can be offered va exstng GSM and UMTS cellular networks. It conssts of an MBMS bearer servce and a MBMS user servce. The latter represents applcatons, whch offer for example multmeda content to the users, whle the MBMS bearer servce provdes methods for user authorzaton, chargng and QoS mprovement to prevent unauthorzed recepton [8]. The major modfcaton n the exstng GPRS platform s the addton of a new entty called Broadcast Multcast-Servce Center (BM-SC). Fgure 1 presents the archtecture of the MBMS. The BM-SC communcates wth the exstng UMTS-GSM networks and the external PDNs [8]. As the term MBMS ndcates, there are two types of servce mode: the broadcast and the multcast mode. In broadcast mode, data s delvered to a specfed area wthout knowng the recevers and whether there s any recever at all n ths area. However, n the multcast mode, the recevers have to sgnal ther nterest for the data recepton to the network and then the network decdes whether the user may receve the data or not. Snce the multcast mode s more complcated than the broadcast mode, t s more useful to present the operaton of the MBMS multcast mode and the way that the moble user receves the multcast data of a servce. Actually, the recepton of an MBMS multcast servce s enabled by certan procedures. These are: Subscrpton, Servce Announcement, Jonng, Sesson Start, MBMS Notfcaton, Data Transfer, Sesson Stop, and Leavng. The phases Subscrpton, Jonng, and Leavng are performed ndvdually per user. The other phases are performed for a servce, that s, for all users nterested n the related servce. The sequence of the phases may be repeated, for example, dependng on the need to transfer data. Also Subscrpton, Jonng, Leavng, Servce Announcement, as well as MBMS Notfcaton may run n parallel to other phases. 5. One-to-Many Packet Delvery Schemes for UMTS In ths secton, we present an overvew of three dfferent one-to-many packet delvery schemes for UMTS. These schemes nclude the multple uncast scheme, the broadcast scheme, and the multcast scheme. The above schemes are presented n detal n the followng paragraphs. Addtonally, we analyze the number of the GTP tunnels establshed n every edge of the network. Fgure 2 shows a subset of a UMTS network. In ths archtecture, there are 2 SGSNs connected to a GGSN, 4 RNCs, and 12 Node Bs. Furthermore, 11 members of a multcast group are located n 6 cells. The BM- SC acts as the nterface towards external sources of traffc [9]. In the presented analyss, we assume that a data stream that comes from an external PDN through BM-SC, must be delvered to the 11 UEs as llustrated n Fgure 2. The analyss presented n the followng paragraphs, covers the forwardng mechansm of the data packets between the BM-SC and the UEs (Fgure 2). Regardng the transmsson of the packets over the Iub and Uu nterfaces, t may be performed on common (ex. Broadcast Channel BCH, Forward Access Channel FACH), dedcated (Dedcated Channel DCH) or shared transport channels (ex. Hgh Speed Downlnk Shared Channel HS-DSCH). As presented n Reference [10], the transport channel that the 3GPP decded to use as the man transport channel for pont-to-multpont MBMS data transmsson s the FACH wth turbo codng and QPSK modulaton at a constant transmsson power. Multple servces can be confgured n a cell, ether tme multplexed on one FACH or transmtted on separate channels. DCH s a pont-to-pont channel and hence, t suffers from the neffcences of requrng multple DCHs to carry the data to a group of users. However, DCH can employ fast closed-loop power control and soft handover mechansms and generally s a hghly relable channel. Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

5 MULTICAST IN UMTS 467 Addtonally, DCH conssts of an uplnk channel. BCH s a pont-to-multpont channel, whch has no uplnk channel. Thus, even wth a large number of multcast recevers, only one BCH s requred for a multcast servce n the cell. Furthermore, a new transport channel named HS-DSCH has been ntroduced as the prmary rado bearer n Release 5 of UMTS. The HS-DSCH resources can be shared between all users n a partcular sector. The prmary channel multplexng occurs n the tme doman, where each Transmsson Tme Interval (TTI) conssts of three slots (or 2 ms) [1]. It has to be mentoned that no decson has yet been made wthn 3GPP on how to optmze the MBMS data flow over the Iub. It has been proposed to avod the duplcaton of Iub data flows and hence, only one Iub data flow per MBMS servce should be used [11] Descrpton of the Broadcast Scheme (Bs) Wth broadcast, the network smply floods multcast packets to all the nodes wthn the network. In ths approach, the sngle packet s ntally transmtted from the BM-SC to the GGSN. Ths procedure mples that the frst GTP sesson s created between the BM-SC and the GGSN. Then, when the GGSN receves the packet, t forwards t to every SGSN that exsts, whch n turn forward the packet to the correspondng RNCs. As we descrbed above, ths leads to the creaton of two GTP sessons for every path between the GGSN and RNCs. The frst GTP sesson s created n the edge Fg. 2. Packet delvery n UMTS. between GGSN and each SGSN, whle the second one between the SGSN and each RNC. Then, each RNC forwards the packet to the correspondng Node Bs and the Node Bs ought to transmt the packet to the moble users n ther range. Havng analyzed the way that the packets are transmtted n the broadcast method, we can calculate the number of the GTP sessons created n the example that s shown n Fgure 2. Frst, we have one GTP sesson n the edge between BM-SC and GGSN. Second, gven the fact that we have two SGSN nodes, the correspondng GTP sessons that are created are two-one for each edge between the GGSN and SGSN. Furthermore, we observe four addtonal GTP sessons, because we assume that there are four RNCs and hence, four edges between the SGSNs and RNCs. The RNCs forward the packets to all Node Bs that have establshed the approprate rado bearers. Fgure 2 shows the number of the GTP sessons that are created n the edges of the network Descrpton of the Multple Uncast Scheme (MUs) Wth multple uncast, each packet s forwarded once to each member of the group separately. Ths means that when the BM-SC receves a packet, the packet s duplcated and each copy of the packet s transferred to a sngle moble user. Consequently, n the edge between the BM-SC and the GGSN, the number of the GTP sessons that are created s equal to the number Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

6 468 A. ALEXIOU AND C. BOURAS of the multcast moble users. Thus, n our example, the number of the GTP sessons n ths edge s 11. Addtonally, when the GGSN receves the packets, t forwards them to the correspondng SGSNs that serve at least one moble user of the multcast group. In our example, the number of the GTP sessons n the edge between the GGSN and the SGSN1 s sx, whle n the edge between the GGSN and the SGSN2 s fve. Then, each of the two SGSNs forwards the receved packets to the correspondng RNCs that serve moble users. Addtonally, the number of the GTP sessons created n the edges SGSN1 RNC1 and SGSN1 RNC2 are four and two, respectvely. Lkewse, the number of the GTP sessons for the edges SGSN2 RNC3 s fve. Then, the RNCs forward the packets to the Node Bs that serve multcast users and have already establshed the approprate rado bearers. The multcast users receve the packets on the approprate rado bearers by FACHs or DCHs or HS-DSCHs. The number of the GTP sessons establshed n the edges of the network s shown n Fgure 2. However, ths scheme, especally when the multcast group conssts of a great number of UEs, produces excessve redundances n the rado transmssons. These redundances add up as nterference to the network and lmt ts potental capacty Descrpton of the Multcast Scheme (Ms) Wth multcast, the packets are forwarded to those Node Bs that have multcast users. Therefore, n Fgure 2, the Node Bs 2, 3, 5, 7, 8, 9 receve the multcast packets ssued by the BM-SC. We brefly summarze the fve steps occurred for the delvery of the multcast packets. Frst, the BM-SC receves a multcast packet and forwards t to the GGSN that has regstered to receve the multcast traffc. Then, the GGSN receves the multcast packet and by queryng ts multcast routng tables, t determnes whch downstream SGSCs have multcast users resdng n ther respectve servce areas. The term downstream refers to the topologcal poston of one node wth respect to another and relatve to the dstrbuton of the multcast data flow. In Fgure 2, the GGSN duplcates the multcast packet and forwards t to the SGSN1 and the SGSN2. Then, both destnaton SGSNs receve the multcast packets and, havng quered ther multcast routng tables, determne whch RNCs are to receve the multcast packets. The destnaton RNCs receve the multcast packet and send t to the Node Bs that have establshed the approprate rado bearers for the multcast applcaton. In Fgure 2, these are Node B2, B3, B5, B7, B8, B9. The multcast users receve the multcast packets on the approprate rado bearers, ether by pont-to-pont channels transmtted to ndvdual users separately or by pont-to-multpont channels transmtted to all group members n the cell. In our analyss, we consder transport channels such as FACH, HS-DSCH, and DCH. In ths approach, every multcast packet s ntally transmtted from the BM-SC to the GGSN. Ths procedure mples that the frst GTP tunnel sesson s created between the BM-SC and the GGSN. The GGSN forwards exactly one copy of the multcast packet to each SGSN that serves multcast users. Ths leads to the creaton of one GTP tunnel sesson between the GGSN and the SGSN1 and one GTP tunnel sesson between the GGSN and SGSN2 (Fgure 2). Havng receved the multcast packets, the SGSN1 forwards exactly one copy of the multcast packet to the RNCs that serve multcast users, whch are the RNC1 and the RNC2. In parallel, the SGSN2 forwards the multcast packets to the RNC3, whch s the only RNC, covered by the SGSN2 that serves multcast users. If none of the SGSNs does not have vald routng nformaton for any multcast user, the pagng procedure s performed n order to determne the requred nformaton from the multcast users. Regardng the edges between the SGSNs and the RNCs n Fgure 2, the frst GTP tunnel s created between the SGSN1 and RNC1, the second between the SGSN1 and RNC2 sesson, and the thrd between the SGSN2 and RNC3. Fnally, the RNCs forward the multcast packets to those Node Bs that multcast users resde n. Addtonally, Fgure 2 shows the exact number of the GTP sessons that are establshed n edges of the network for the Multcast scheme. 6. Evaluaton of Dfferent One-to-Many Packet Delvery Schemes In ths secton, we present an evaluaton, n terms of the telecommuncaton costs, of dfferent one-to-many delvery schemes. We consder dfferent UMTS network topologes and dfferent transport channels for the transmsson of the multcast data General Assumptons We consder a subset of a UMTS network consstng of a sngle GGSN and N SGSN nodes connected to the GGSN. Furthermore, each SGSN manages a number of N ra RAs. Each RA conssts of a number of N rnc RNC Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

7 nodes, whle each RNC node manages a number of N ura URAs. Fnally, each URA conssts of N nodeb cells. The total number of RAs, RNCs, URAs, and cells are: N RA = N SGSN N ra (1) N RNC = N SGSN N ra N rnc (2) N URA = N SGSN N ra N rnc N ura (3) N NODEB = N SGSN N ra N rnc N ura N nodeb (4) The total transmsson cost for packet delveres ncludng pagng s consdered as the performance metrc. Furthermore, the cost for pagng s dfferentated from the cost for packet delveres. We make a further dstncton between the processng costs at nodes and the transmsson costs on lnks, both for pagng and packet delveres. As presented n Reference [12] and analyzed n Reference [3], we assume that there s a cost assocated wth each lnk and each node of the network, both for pagng and packet delveres. For the analyss, we apply the followng notatons: D gs D sr D rb D BCH D DCH D HS DSCH D FACH S sr S rb S a p g p gm p s p sm p r p rm p b a s a r a b MULTICAST IN UMTS 469 Transmsson cost of packet delvery between GGSN and SGSG Transmsson cost of packet delvery between SGSN and RNC Transmsson cost of packet delvery between RNC and Node B Transmsson cost of packet delvery over Iub and Uu wth BCHs Transmsson cost of packet delvery over Iub and Uu wth DCHs Transmsson cost of packet delvery over Iub and Uu wth HS-DSCHs Transmsson cost of packet delvery over Iub and Uu wth FACHs Transmsson cost of pagng between SGSN and RNC Transmsson cost of pagng between RNC and Node B Transmsson cost of pagng over the ar Processng cost of packet delvery at GGSN Processng cost of multcast packet delvery at GGSN Processng cost of packet delvery at SGSN Processng cost of multcast packet delvery at SGSN Processng cost of packet delvery at RNC Processng cost of multcast packet delvery at RNC Processng cost of packet delvery at Node B Processng cost of pagng at SGSN Processng cost of pagng at RNC Processng cost of pagng at Node B The total number of the multcast UEs n the network s denoted by N UE. For the cost analyss, we defne the total packets per multcast sesson as N p. Snce network operators wll typcally deploy an IP backbone network between the GGSN, SGSN, and RNC, the lnks between these nodes wll consst of more than one hop. Addtonally, the dstance between the RNC and Node B conssts of a sngle hop (l rb = 1). In the presented analyss we assume that the dstance between GGSN and SGSN s l gs hops, whle the dstance between the SGSN and RNC s l sr hops. Furthermore, we assume that the probablty that a UE s n PMM detached state s P DET, the probablty that a UE s n PMM dle/rrc dle state s P RA,the probablty that a UE s n PMM connected/rrc URA connected state s P URA, and fnally the probablty that a UE s n PMM connected/rrc cell-connected state s P cell. Addtonally, n the multcast scheme, we consder dfferent values for the processng costs (P gm, P sm, P rm ) at the nodes of the UMTS network than the correspondng values (P g, P s, P r ) n the other two schemes snce some overhead s needed n the UMTS nodes n order to mantan the routng tables requred for the packet forwardng n the multcast scheme. In the remander of ths secton, we descrbe a method that models the multcast user dstrbuton n the network. In partcular, we present a probablstc method that calculates the number of multcast users n the network (n UE ), the number of SGSNs that serve multcast users (n SGSN ), the number of RNCs that serve multcast users (n RNC ), and fnally the number of Node Bs that serve multcast members (n NODEB ). As ntroduced n Reference [7] and analyzed n Reference [3], we classfy the RAs nto L RA categores. For 1 L RA there are N (RA) RAs of class. Therefore, the total number of RAs wthn the network s N RA = L RA =1 N(RA). Suppose that the dstrbuton of the multcast users among the classes of RAs follows the Posson dstrbuton wth λ = θ (RA) where 1 L RA. In general, the probablty that k exactly multcast users resde n the RAs of class s calculated from the followng equaton: ( ) k e θ(ra) ( ) p k, θ (RA) = Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8: k! θ (RA) (5)

8 470 A. ALEXIOU AND C. BOURAS Thus, the probablty none of the RAs of class serves multcast users s p(0,θ (RA) ) = e θ(ra), whch n turn means that the probablty at least one multcast user s served by the RAs of class s p = 1 p(0,θ (RA) ) = 1 e θ(ra). Snce every class conssts of N (RA) RAs, the total number of the RAs n the class, that serve multcast users s N (RA) (1 e θ(ra) ). Thus, the total number of the RAs of every class that serve multcast users s: L RA n RA = =1 N (RA) ) (1 e θ(ra) (6) where θ (RA) represents the number of multcast users for the N (RA) RAs of class. If there are n RA RAs that are servng multcast users, the probablty that an SGSN does not have any such RA s: N RA N ra n RA p SGSN = N, f n RA N RA N ra RA n RA 0, othewse (7) Based on Equaton (7), the total number of SGSNs that are servng multcast users can be calculated as follows: n SGSN = N SGSN (1 p SGSN ). The total number of multcast users n the network s: L RA N UE = N (RA) θ (8) =1 where θ s the number of multcast users n a RA of class. As n Reference [3], we assume that all RNCs wthn a servce area of class have the same multcast populaton dstrbuton densty as n the RA case. Based on a unform densty dstrbuton wthn a sngle RA, the multcast populaton of an RNC wthn the servce area of a class RA s θ (RNC) = θ (RA) number of RNCs of class s N (RNC) /N rnc. The total = N (RA) N rnc. Assumng that the number of RA categores s equal to the number of RNC categores (L RNC = L RA ), the total number of RNCs that serve multcast users s: n RNC = L RNC =1 N (RNC) ) (1 e θ(rnc) (9) The same are appled to the cells wthn the servce area of an RNC. The average number of multcast users for a sngle cell of class s θ (B) = θ (RNC) /(N ura N nodeb ). The number of Node Bs belongng to class s N (B) = N (RNC) N ura N nodeb. Assumng that the number of the RNC categores s equal to the number of the Node B categores (L RNC = L NODEB ), the total number of Node Bs that serve multcast users s: n NODEB = L NODEB =1 N (B) ) (1 e θ(b) (10) 6.2. Broadcast Scheme (Bs) In ths scheme, the packets are broadcasted to all the nodes of the network and no pagng procedure s requred. Regardng over the ar transmsson, the transport channel that s used n ths scheme s the BCH. It s obvous that the total cost of the packet delvery s ndependent from the number of the multcast users (N UE ). The total cost of the packet delvery to the multcast users s computed as follows: Bs = [ ( ) p g + N SGSN Dgs + p s + NRNC (D sr + p r ) + N NODEB (D rb + p b + D BCH ) ] N p (11) 6.3. Multple Uncast Scheme (MUs) Wth multple uncast, each packet s forwarded once to each member of the group separately. In Fgure 2, RNC1, for nstance, would receve four duplcate copes of the same multcast packet from the SGSN1. It s obvous that the cost of a sngle packet delvery to a multcast user depends on ts MM and RRC state. If the multcast member s n PMM connected/rrc cell-connected state, then there s no need for any pagng procedure nether from the SGSN nor from the servng RNC. In ths case, the packet delvery cost s derved from Eqnuaton (12). It has to be mentoned that ths quantty does not nclude the cost for the transmsson of the packets over the Iub and Uu nterfaces snce ths cost depends frst on the number of multcast users and second on the transport channel used for data transmsson. C cell = p g + D gs + p s + D sr + p r (12) If the multcast member s n PMM connected/rrc URA connected state, then the RNC must frst page all the cells wthn the URA n whch moble users resde and then proceeds to the data transfer. After the Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

9 MULTICAST IN UMTS 471 subscrber receves the pagng message from the RNC, t returns to the RNC ts cell ID. The cost for pagng such a multcast member s: C URA =N nodeb (S rb + a b + S a ) + S a + a b + S rb + a r (13) If the multcast member s n PMM dle/rrc dle state, the SGSN only stores the dentty of the RA n whch the user s located. Therefore, all cells n the RA must be paged. The cost for pagng such a multcast member s: C RA = N rnc (S sr + a r ) + (N rnc N ura N nodeb ) (S rb + a b + S a ) + S a + a b + S rb + a r + S sr + a s (14) Takng nto consderaton that the pagng procedure s performed on the frst packet of a data sesson, the total cost of the Multple Uncast scheme s derved from the followng equatons for the three dfferent transport channels (DCH, FACH, HS-DSCH), where n NODEB represent the number of Node Bs that serve multcast users. [ ( ) Pcell C cell N p + P URA CURA + C cell N p ( ) +P RA CRA + C cell N p + (Pcell + P URA ] +P RA ) (D DCH + D rb + p b ) N p NUE (15) [ ( ) Pcell C cell N p + P URA CURA + C cell N p ( )] MUs= +P RA CRA + C cell N p NUE +n NODEB (D FACH + D rb + p b ) N p (16) [ Pcell C cell N p + P URA ( CURA + C cell N p ) ( ) +P RA CRA + C cell N p + (Pcell + P URA ] +P RA ) (D HS-DSCH + D rb + p b ) N p NUE (17) The last term n each of the above three equatons represent the cost of the packet transmsson n the Iub and Uu nterfaces. In general, n case we use the FACH as transport channel, each multcast packet send once over the Iub nterface and then the packet s transmtted to the UEs that served by the correspondng Node B. In case we use DCHs for the transmsson of the multcast packets, each packet s replcated over the Iub as many tmes as the number of multcast users that the correspondng Node B serves. Fnally, wth HS-DSCH, a separate tmeslot must be used to transport the multcast data to each multcast recever. However, one could envson that all multcast recevers could receve the same tmeslot that contans the multcast data, but n ts current form the HS-DSCH has not been modfed to allow ths. Thus, the number of tme slots requred for the transmsson of the multcast data to the multcast users s equal to the number of multcast users resde n the correspondng cell Multcast Scheme (Ms) In the multcast scheme, the multcast group management s performed at the GGSN, SGSN, and RNC and multcast tunnels are establshed over the Gn and Iu nterfaces. All multcast users that are n PMM dle/rrc dle or PMM connected/rrc URA connected state must be paged. After the pagng procedure, the RNC stores the locaton of any UE at a cell level. The cost for that pagng procedure s gven by Equatons (13) and (14), respectvely. In multcast, the SGSN and the RNC forward a sngle copy of each multcast packet to those RNCs or Node Bs respectvely that are servng multcast users. After the correct multcast packet recepton at the Node Bs that serve multcast users, the Node Bs transmt the multcast packets to the multcast users va common, dedcated, or hgh speed shared transport channels. The total cost for the Multcast scheme s derved from the followng equaton where n SGSN, n RNC, n NODEB represent the number of SGSNs, RNCs, Node Bs, respectvely that serve multcast users. Ms = [ ( ) p gm +n SGSN Dgs +p sm +nrnc (D sr +p rm ) +Y] N p + (P RA C RA + P URA C URA ) N UE = D packet delvery + D pagng (18) where n NODEB (D FACH + D rb + p b ) f channel = FACH N UE (D DCH + D rb + p b ) Y = f channel = DCH N UE (D HS DSCH + D rb + p b ) f channel = HS DSCH D packet delvery = [ p gm + n SGSN ( Dgs + p sm ) + n RNC (D sr + p rm ) + Y ] N p D pagng = (P RA C RA + P URA C URA ) N UE The parameter Y represents the multcast cost for the transmsson of the multcast data over the Iub and Uu Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

10 472 A. ALEXIOU AND C. BOURAS Table I. Chosen values for the calculaton of transmsson costs n the lnks. Lnk Lnk capacty factor (k) Number of hops (l) Transmsson cost (D) GGSN SGSN k gs = 0.5 l gs = 6 D gs = 12 SGSN RNC k sr = 0.5 l sr = 3 D sr = 6 RNC Node B k rb = 0.2 l rb = 1 D rb = 5 nterfaces. Ths cost of the multcast scheme depends manly on the dstrbuton of the multcast group wthn the UMTS network and secondly on the transport channel that s used. In case we use the FACH as transport channel, each multcast packet send once over the Iub nterface and then the packet s transmtted to the UEs that served by the correspondng Node B. However, n case we use DCHs or HS-DSCHs for the transmsson of the multcast packets over the Iub each packet s replcated over the Iub as many tmes as the number of multcast users that the correspondng Node B serves. 7. Results Havng analyzed the costs of the above presented oneto-many packet delvery methods, we try to evaluate the cost of each scheme assumng a general network topology. In general, we assume a network confguraton, wth N SGSN = 10, N ra = 10, N rnc = 10, N ura = 5 and N nodeb = 5. The packet transmsson cost (D xx ) n any segment of the UMTS network depends on two parameters: the number of hops between the edge nodes of ths network segment and the capacty of the lnk of the network segment. Ths means that D gs = l gs /k gs, D sr = l sr /k sr and D rb = l rb /k rb. Parameter k xx represents the profle of the correspondng lnk between two UMTS network nodes. More specfcally, n the hgh capacty lnks at the CN, the values of k xx are greater than the correspondng values n the low capacty lnks at UTRAN. For the cost analyss and wthout loss of generalty, we assume that the dstance between the GGSN and SGSN s 6 hops, the dstance between Table II. Chosen parameters values. SGSN and RNC s 3 hops, and the dstance between RNC and Node B s 1 hop. The above parameters as well as the values of the k xx are presented n detal n Table I. Regardng the transmsson cost of pagng (S xx ) n the segments of the UMTS network, t s calculated n a smlar way as the packet transmsson cost (D xx ). More specfcally, S xx s a fracton of the calculated transmsson cost (D xx ) and n our case, we assume that t s three tmes smaller than D xx. As we can observe from the equatons n the prevous secton, the costs of the schemes depend on a number of other parameters. Thus, we have to estmate the value of these parameters. Ths procedure mples that we choose the values approprately, takng nto consderaton the relatons between them. The chosen values of the parameters are presented n Table II. As t s shown n Table II, the values for the transmsson costs of the packet delvery over the ar wth each of the four transport channels are dfferent. More specfcally, the transmsson cost over the ar wth BCH (D BCH )orfach(d FACH ) s bgger than the D DCH, whch n turn s bgger than the transmsson cost of the packet delvery over the ar wth HS-DSCH (D HS DSCH ). Ths occurs because BCH and FACH as common channels requre hgh transmsson power n order to reach all the users wthn the coverage area even f they are not members of the multcast group. Addtonally, regardng the relaton of the costs of the DCH and the HS-DSCH, the DCH s a power controlled channel whle the HS-DSCH s rate controlled. In Reference [1], t s shown that the HS-DSCH throughput s by far bgger than the DCH throughput for a gver fracton of Node B power allocated for these channels. For example, f we consder a macro cell envronment D gs D sr D rb S sr S rb S a p g p s p r p gm p sm p rm p b a s a r a b p RA p URA p cell /3 4/ D BCH D DCH D FACH D HS DSCH Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

11 MULTICAST IN UMTS 473 confguraton wth a ITU Pedestran-A delay profle and 7W allocated to HSDPA from the avalable Node B power, the average cell throughput on the HS-DSCH s 1.4 Mbps whle the value of the throughput wth only non HSDPA termnals actve (64 kbps R99 DCH users only) n the cell s 1 Mbps. The average throughput that the HSDPA users experence depends on the number of smultaneous users that are sharng the HS-DSCH, as well as ther relatve experenced sgnal qualty (E b /N 0 ). On the contrary, the total downlnk transmsson power allocated for DCH s varable and ncreasng exponentally whle the UE dstance from the node B s ncreasng. Also, the more the UEs n the cell thus the hgher the nterference, the more exponental the ncrease n the total power requred. Furthermore, the power requred for a reference DCH user depends on the experenced sgnal qualty (E b /N 0 ). The followng equaton calculates the total requred transmsson power at the Node B for a reference DCH user [13]. P T = L p, P N + χ + p P T L p, W (E b /N 0 ) R b, + p (19) where P T s the base staton transmtted power, P T s the power devoted to the th user, L p, s the path loss, R b, the th user transmsson rate, W the bandwdth, P N the background nose, p s the orthogonalty factor, and χ s the ntercell nterference observed by the th user. Regardng the parameters P gm, P sm, and P rm n the multcast scheme and ther relaton wth the parameters P g, P s, and P r of the other two schemes, the latter parameters have lower values that the values of the former parameters snce some overhead s needed n the UMTS nodes n order to mantan the routng tables requred for the multcast packet forwardng n the multcast scheme. At ths pont, we have to menton that snce the nodes that are responsble for the establshment of the rado bearers are the RNCs we assume that there s no need for mantanng any routng tables n the Node Bs and thus there s no any addtonal cost for the processng at these nodes. Furthermore, we have chosen approprately the probabltes P RA, P URA, and P cell More specfcally, the probablty that a UE s n PMM dle/rrc dle state s P RA = 0.6. The probablty that a UE s n PMM-connected/RRC URA-connected state s P URA = 0.2 and the probablty that a UE s n PMM connected/rrc cell-connected state s P cell = 0.1. Addtonally, there s a probablty that the UE s not reachable by the network and we consder t to be 0.1. It s true that the performance of the three schemes depends manly, on the confguraton of the UMTS network that s under nvestgaton. Therefore, we consder a general network confguraton, wth N SGSN = 10, N ra = 10, N rnc = 10, N ura = 5 and N nodeb = 5. In our analyss, we assume that we have two classes of RAs. A class = 1 RA has multcast user populaton of θ 1 = 1/δ and a class = 2 RA has a multcast user populaton of θ 2 = δ.ifδ 1, the class = 1RAhas a small multcast user populaton and the class = 2 RA has a large multcast user populaton. Let α be the proporton of the class = 1 RAs and (1 α) bethe proporton of the class = 2 RAs [7]. Thus, the number of class = 1 RAs s N (RA) 1 = αn RA and the number of class = 2 RAs sn (RA) 2 = (1 α)n RA. Each RA of class {1, 2} s n turn subdvded nto N rnc RNCs of the same class and smlarly, each RNC of class {1, 2} s subdvded nto N ura N nodeb Node Bs of the same class. Take nto consderaton the above mentoned parameters, Equaton (8) can be rewrtten as follows: N UE = 2 =1 N (RA) = N RA ( α δ + δ αδ ) θ = N (RA) 1 θ 1 + N (RA) 2 θ 2 (20) It s obvous from Equaton (20) that as α decreases and δ ncreases the number of multcast users ncreases rapdly. In Fgures 3 and 4, we plot the cost of the three schemes n functon of α, for dfferent values of δ and N p respectvely. Snce n our model we consder three dfferent transport channels over the ar for the MUs and the Ms, n the followng fgures only the channel wth the lowest cost of each scheme s presented. Addtonally, we provde the comparson of the costs of every scheme usng dfferent transport channels n separate fgures. Generally, we observe from Fgure 3 and Fgure 4, that the cost of the broadcast scheme s constant, whle the costs of the other two schemes decrease as α ncreases. Wth δ 1, t has to be mentoned that there are no multcast users n a RA of class = 1 and there are many multcast members n a RA of class = 2. Furthermore, as α ncreases, the number of class = 1 RAs wth no multcast users ncreases and hence, the costs of the MUs and the Ms decrease as t s shown n Fgures 3 and 4. More specfcally, n Fgure 3, we present the costs of the three schemes n functon of α for δ = 100 (Fgure 3a) and δ = 1000 (Fgure 3b). For δ = 100, the Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

12 474 A. ALEXIOU AND C. BOURAS Fg. 3. Total cost for the three schemes n functon of α wth N p = (a) δ = 100, (b) δ = MUs has the lowest cost whle the Bs has the hgher cost ndependently of the value of α. Ths occurs because n small values of α, there are many RAs wth large multcast users populaton (class = 2 RAs). However, the value of δ = 100 results to a small number of multcast users wthn the network and hence, the costs of MUs and Ms are kept n lower values than the value of the cost of the Bs. In addton, f δ = 1000 (Fgure 3b), the number of multcast users wthn the network s ncreased and ths results to an ncreased cost for the MUs. The costs of the other two schemes behave as follows: for α<0.06 the Bs has the lowest cost and for α>0.06 the Ms has the lowest cost. The latter occurs because for small values of α and ncreased number of δ, the number of multcast users wthn the network s ncreased and furthermore, there are many class = 2 RAs n the network wth large multcast users populaton. Ths means that the multcast users wthn the network are spreaded to many RAs and hence, the cost of the pagng whch s requred n the Ms s ncreased, makng the Ms neffcent for ths network topology. On the other hand, when the value of α s ncreased, the number of class = 1 RAs wth no multcast users s ncreased and all the multcast users are located n a small number of class = 2 RAs. Thus, the Ms s more effcent than the Bs as t s shown n Fgure 3b. Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

13 MULTICAST IN UMTS 475 Fg. 4. Total cost for the three schemes n functon of α wth δ = (a) N p = 50, (b) N p = In Fgure 4, we present the costs of the three schemes n functon of α for N p = 50 (Fgure 4a) and N p = 3000 (Fgure 4b). The value of N p = 50 corresponds to a multcast sesson wth few transmtted packets whle the value of N p = 3000 corresponds to a multcast sesson consstng of a large number of packets. For N p = 50 (Fgure 4a), f α<0.70 the Bs outperforms both the Ms and the MUs. The later has the hghest value for these values of α. Ths occurs because a small value of α results to a topology wth large number of hgh populated RAs (class = 2 RAs). As the parameter α ncreases the number of RAs wth no multcast users ncreases and the Ms becomes more favorable than the other two schemes. Obvously, for α>0.70 the Ms has the lowest cost compared to Bs and MUs. Addtonally, for α>0.90, the MUs outperforms the Bs. Ths occurs because a value of α close to 1 n conjuncton wth a bg value of δ (δ = 1000) results to small number of multcast users accordng to Equaton (21). The same observaton regardng the MUs and the Bs occurs also n Fgure 4b. Furthermore, for N p = 3000 (Fgure 4b), the multcast sesson s conssted of a large number of packets and hence, the Ms has the lowest cost ndependently of the parameter α. Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

14 476 A. ALEXIOU AND C. BOURAS Fg. 5. Total cost for the three schemes n functon of δ wth N p = (a) α = 0.1, (b) α = 0.9. Fgure 5 presents the costs of the three schemes n functon of δ for α = 0.1 (Fgure 5a) and α = 0.9 (Fgure 5b). Our frst observaton s that the cost of the Bs s constant and t s not dependng on the parameter δ whle the cost of the MUs ncreases as δ ncreases. As t has already been mentoned, a class = 1 RA has 1/δ multcast users whle a class = 2 RA has δ multcast users. Furthermore, f α converges to 1 (Fgure 5b), the number of class = 2 RAs n the network s small. Thus, the multcast users are located n small number of RAs and as a result, the cost of the Ms s kept n lower values (ndependently of the parameter δ) than the cost of the MUs and the Bs (Fgure 5b). The later, obvously, has the hgher cost as t s shown n Fgure 5b. Addtonally, f δ s small, the MUs outperforms Ms snce the number of multcast users s lmted. On the other hand, Fgure 5a shows that the Ms performs effcently f δ<900, but t s outperformed by the Bs for δ>900. Ths occurs because the value of α s small and the number of class = 2 RAs wth large multcast users populaton s ncreased and hence, as the number of multcast users n a class = 2RAncreases, there are many users among many cells n the network makng the Bs more approprate for the multcast packet delvery. Addtonally, f δ converges to zero the MUs outperforms Ms snce the number of multcast users n the network s lmted. Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

15 MULTICAST IN UMTS 477 Fg. 6. Total cost for the three schemes n functon of N p wth δ = (a) α = 0.1, (b) α = 0.9. The costs of the three schemes n functon of the number of packets per multcast sesson are presented n Fgure 6 for two dfferent values of α. Our frst observaton s that as N p ncreases, the total cost of each scheme ncreases. When α s small (Fgure 6a), the number of RAs n the class = 2 s ncreased and hence, the multcast users are spread to many RAs and cells wthn the network. Thus, the Ms and the Bs outperform the MUs for the gven value of δ = Regardng the Bs and the Ms they have smlar costs n functon of N p. When α = 0.9 (Fgure 6b), the number of RAs n the class = 2 s small and hence, the multcast users are resdng n small number of RAs and cells wthn the network. In that case the Ms outperforms both MUs and Bs. The latter has obvously the hghest cost for ths network topology. In Fgure 7, the total costs for the Ms usng dfferent transport channels over the ar n functon of α, are presented. As we can observe, the cost decreases as α ncreases, ndependently of the type of the transport channel used for the transmsson of the multcast data over the ar. Ths occurs because the ncrement of α entals that the number of RAs that have small populaton ncreases. Thus, the total number of the multcast users decreases and the total cost for the transmsson of the data decreases. Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

16 478 A. ALEXIOU AND C. BOURAS Fg. 7. Total costs for the Ms usng dfferent transport channels over the ar n functon of α wth N p = (a) δ = 100, (b) δ = More specfcally, n Fgure 7a, we observe that the cost of the Ms usng HS-DSCH s smaller than the cost of the Ms f we use DCHs or FACH. Ths occurs because the small value of δ, results to a reduced number of multcast users n the network and hence, the HS-DSCH (or DCH) s the more effcent transport channel n terms of the cost. On the other hand, the ncreased cost of the FACH (D FACH ) n conjuncton wth the small number of multcast users n the network makes FACH neffcent for ths user dstrbuton scheme. The opposte occurs n Fgure 7b where the value of δ s ncreased, whch means that the number of multcast users n the network s also ncreased. The ncreased number of multcast users n the network makes the use of DCHs and the HS-DSCHs neffcent for the transmsson of the data over the Iub and Uu nterfaces and the FACH the most approprate transport channel as t s shown n Fgure 7b. The same observatons occur n the case of the MUs. In Fgure 8, the total costs for the Ms and MUs usng dfferent transport channels over the ar n functon of δ, are presented. In the case of the Ms, we chose a small value for the parameter α because the Ms becomes effcent when there s an ncreased densty of multcast Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

17 MULTICAST IN UMTS 479 Fg. 8. Total costs for the Ms and MUs usng dfferent transport channels over the ar n functon of δ wth N p = (a) α = 0.1, (b) α = 0.5. users n the network. On the other hand, the value of α n the MUs must be bg for the approprate use of ths scheme. Therefore, for the Ms (Fgure 8a), we choose a value of α = 0.1 whch means that there are many RAs n the network wth a great number of multcast users n these. Regardng the MUs (Fgure 8b), we consder the same amount of RAs havng large multcast users populaton and RAs havng small multcast users populaton (α = 0.5). More specfcally, n Fgure 8a, we observe that for small values of δ, the cost of the Ms usng DCHs or HS-DSCHs s small because there s a small number of multcast users n the network. Thus, the use of DCHs or HS-DSCHs for the data transmsson over the ar whch reduces the cost of the scheme. On the other hand, bgger values of the parameter δ mply bgger number of multcast users n the network and hence, the use of FACH s more approprate. The latter has obvous advantage n topologes wth many hgh populated RAs snce the multcast packets are send once over Iub and Uu nterfaces. In Fgure 8b, we present the cost of the MUs usng the three transport channels. More specfcally, as n the Ms, for small values of δ, the number of the multcast users s small and hence, the use of DCHs or HS-DSCHs for the data transmsson over Uu s the most effcent. On the other hand, as Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

18 480 A. ALEXIOU AND C. BOURAS gven tme. In ths way, nterestng ssues must be taken nto account such as the user moblty, the max number of users that a channel can servce as well as power ssues. 9. Conclusons Fg. 9. Summary of the analyss. δ ncreases, the use of FACH makes the scheme more effcent. As t s shown n Fgures 7 and 8, the cost of usng HS-DSCHs for the transmsson of the multcast data over the Iub and Uu nterfaces n both the MUs and the Ms s always lower than the cost of usng DCHs. Our model handles dentcal the two types of channels and ts only dfference s the lower cost of the HS- DSCH (D HS DSCH ) than the cost of DCH (D DCH ). Our decson of usng lower cost for the HS-DSCH than the DCH s explaned on the begnnng of the results secton. In general, the selecton of the bearer type s operator-depended, typcally based on the downlnk rado resource stuaton so that the effcency of the resource usage can be maxmzed. A summary of our analyss s presented n Fgure Future work The step that follows ths work s to carry out experments usng the NS-2 smulator. Ths means that we have to mplement the above presented one-to-many packet delvery schemes and confrm the relaton of the costs through the experments. Further work could focus on evaluatng the performance of MBMS, once the complete specfcatons are avalable, as well as studyng the management of the user moblty. Furthermore, an MBMS handover mechansm n RNC should be realzed n order to optmze the transmsson of the multcast data n the Iub nterface. Addtonally, an nnovatve algorthm could be developed whch would choose the most effcent transport channel for the transmsson of the multcast packets dependng on the dstrbuton of the users and the network topology at any In ths paper, we presented an overvew of three dfferent packet delvery schemes for UMTS. These schemes nclude the Broadcast, the Multple Uncast, and the Multcast scheme. We analytcally presented these schemes and analyzed ther performance n terms of the packet delvery cost and the scalablty of each scheme. More specfcally, we provded a formula to dfferentate the number of the multcast users and ther dstrbuton. Ths formula s a functon of the α and δ, where as α decreases and δ ncreases the number of multcast users ncreases rapdly and vce-versa. Thus, n the multcast scheme, whch s useful when there are many multcast users gathered n some cells of the network, we chose a small value of α andabgvalueofδ. On the other hand, the multple uncast s an effcent transmsson scheme when there s a small number of multcast users and hence, we appled a bg value of α and a small value of δ. References 1. Holma H, Toskala A. WCDMA for UMTS: Rado Access for Thrd Generaton Moble Communcatons. John Wley & Sons: Chchester, ISBN Gossan H, Cordero C, Argawal D. Multcast: wred to wreless. IEEE Communcatons Magazne 2002; 40(6): Rummler R, Chung Y, Aghvam H. Modelng and Analyss of an effcent multcast mechansm for UMTS. IEEE Transactons on Vehcular Technology 2005; 54(1): GPP TS V Techncal Specfcaton Group Servces and System Aspects; General Packet Rado Servce (GPRS); Servce descrpton; Stage 2 (Release 7) Yang S, Ln Y. Performance evaluaton of locaton management n UMTS. IEEE Transactons on Vehcular Technology 2003; 52(6): Hauge M, Kure O. Multcast n 3G networks: employment of exstng IP multcast protocols n UMTS. 5th ACM Internatonal Workshop on Wreless Moble Multmeda 2002, Ln Y. A multcast mechansm for moble networks. IEEE Communcaton Letters 2001; 5(11): GPP TS V Techncal Specfcaton Group Servces and System Aspects; Multmeda Broadcast/Multcast Servce; Stage 1 (Release 7) GPP TS V Techncal Specfcaton Group Servces and System Aspects; Multmeda Broadcast/Multcast Servce (MBMS); Archtecture and functonal descrpton (Release 6) GPP, TR v Techncal Specfcaton Group Servces and System Aspects; Multmeda Broadcast/Multcast Servce; Archtecture and functonal descrpton (Release 6). Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8:

19 MULTICAST IN UMTS Bon A, Launay E, Menvlle T, Stuckmann P. Multmeda broadcast multcast servce technology overvew and servce aspects. Ffth IEEE Internatonal Conference on 3G Moble Communcaton Technologes (3G 2004) Ho J, Akyldz I. Local anchor scheme for reducng sgnalng costs n personal communcatons networks. IEEE/ACM Transactons on Networkng 1996; 4(5): Romero J, Sallent O, Agust R, Daz-Guerra M. Rado Resource Management Strateges n UMTS. John Wley & Sons: Chchester, ISBN Authors Bographes Antonos Alexou obtaned hs Dploma from the Department of Electrcal and Computer Engneerng of the Arstotle Unversty of Thessalonk (Greece) and hs Master Degree from the Computer Engneerng and Informatcs Department of Patras Unversty. He s currently a PhD Canddate of the Department of Computer Engneer and Informatcs of Patras Unversty. Furthermore, he s workng as R&D Computer Engneer at the Research Unt 6 of the Research Academc Computer Technology Insttute n Patras (Greece). Hs research nterests nclude Moble Telecommuncatons Networks, Multcast Routng, User Moblty n Cellular Networks, Congeston Control and Qualty of Servce, Moble and Wreless Ad-hoc Networks. He has publshed three papers n journals and 16 papers n well-known refereed conferences. Chrstos Bouras obtaned hs Dploma and PhD from the Computer Scence and Engneerng Department of Patras Unversty (Greece). He s currently an Assocate Professor n the above department. Also he s a scentfc advsor of Research Unt 6 n Research Academc Computer Technology Insttute (CTI), Patras, Greece. Hs research nterests nclude Analyss of Performance of Networkng and Computer Systems, Computer Networks and Protocols, Telematcs and New Servces, QoS and Prcng for Networks and Servces, e-learnng, Networked Vrtual Envronments and WWW Issues. He has extended professonal experence n Desgn and Analyss of Networks, Protocols, Telematcs and New Servces. He has publshed 200 papers n varous well-known refereed conferences and journals. He s a co-author of 7 books n Greek. He has been a PC member and referee n varous nternatonal journals and conferences. He has partcpated n R&D projects such as RACE, ESPRIT, TELEMATICS, EDUCATIONAL MULTIMEDIA, ISPO, EMPLOYMENT, ADAPT, STRIDE, EUROFORM, IST, GROWTH and others. Also he s member of, experts n the Greek Research and Technology Network (GRNET), Advsory Commttee Member to the World Wde Web Consortum (W3C), IEEE Learnng Technology Task Force, IEEE Techncal Communty for Servces Computng WG 3.3 Research on Educaton Applcatons of Informaton Technologes and W 6.4 Internet Applcatons Engneerng of IFIP, Task Force for Broadband Access n Greece, ACM, IEEE, EDEN, AACE and New York Academy of Scences. Copyrght 2006 John Wley & Sons, Ltd. Wrel. Commun. Mob. Comput. 2008; 8: