Internatonal Dstrbuted Sensor etworks Volume 3, Artcle ID 3847, 7 pages http://dx.do.org/.55/3/3847 Research Artcle A Utlty-Based Rate Allocaton of MM Servce n Heterogeneous Wreless Envronments Yao Huang, Hu Tan, Je Zhang, Cheng Qn, and Zhbo Wang State Key Laboratory of etworkng and Swtchng Technology, Bejng Unversty of Posts and Telecommuncatons, P.O. Box 9, X Tu Cheng Road, Hadan Dstrct, Bejng 876, Chna The Communcatons Group, Department of Electronc and Electrcal Engneerng, Unversty of Sheffeld, Mappn Street, Sheffeld S 3JD, UK Correspondence should be addressed to Yao Huang; bupthuangyao@63.com Receved 4 June 3; Revsed 8 August 3; Accepted 4 August 3 Academc Edtor: Ln Ba Copyrght 3 Yao Huang et al. Ths s an open access artcle dstrbuted under the Creatve Commons Attrbuton Lcense, whch permts unrestrcted use, dstrbuton, and reproducton n any medum, provded the orgnal work s properly cted. Ths paper presents a dynamc adaptve machne-to-machne (MM) servce rate allocaton scheme for optmum traffc dstrbuton n heterogeneous wreless envronments (HWEs). Accordng to the MM servce characterstcs, t proposes the utlty functon, whch forms a convex optmzaton problem that maxmzes the utlty of MM servce and can be solved by the Lagrange multpler. The smulaton results show the proposed method convergence, and t can acheve load balance wth maxmzed throughput and mnmzed cost.. Introducton Machne-type communcaton (MTC) s regarded as one of the next fronters n wreless communcatons. Recently, there are growng researches about MTC due to the fact t can be wdely used n many aspects, such as smart buldng, smart grd, and envronment montorng. In 3GPP, t s proposed that each MTC devce attaches to the exstng cellular nfrastructure (e.g., LTE-Advanced) [, ], by whch hgher layers connectons between the MTC controller and MTC devces are provded. Thus, the subsequent problems le n the access management on the ar nterface [3]. Furthermore, there are also many systems archtecturedffcultesaboutmtc [4 6]. Reference [4] dentfes potental ssues on the ar nterface. Reference [5] explores systems archtecture problems that are assocated wth the evolvng home MM network. Reference [6] addresses the new requrement to use the MM equpment wth hgh confguraton and flexblty to open up exctng new use cases, servces, and applcatons. Although MTC has been developed for years, t stll faces many challenges due to the dfferences from the conventonal human-type communcatons, such as low moblty, tme controlled []. A man MTC characterstc s that there could be a large number of MM devces exstent n the network. It s possble that congeston wll happen when massve MM devces smultaneously connect to the network, as wreless network resources are lmted. In order to avod congestons, some mproved measures are proposed. Access class barng [7 9] s effectvely to solve plenty of MM devces access, but the cost s not nvolved n [7 9]. Meanwhle, because massve MM devces can produce a great deal of MM servce data, t s necessary to reduce the transmsson cost. The MM servce characterstcs are also dfferent for varous MTC applcatons []. For example, some applcatons requre hard tmng constrants. It wll be therefore dangerous when tmng constrants of securty montorng are volated. Some applcatons demand soft tmng constrants, such as meterng water and gas or electrcty. These MM servces can tolerate soft tme delay. Therefore, the characterstcs of MM servce need to be fully consdered n MM communcatons. In addton, t s well known that n the future multple rado access technologes, such as 3G/4G, WMax, and WLA, wll coexst and form heterogeneous wreless envronments (HWEs), where users have optons to access the best wreless network that fts ther servces requrements, and network operators can ncrease revenue wth a more
Internatonal Dstrbuted Sensor etworks probablty usage of rado resources [ 3]. Reference [] proposes a rate allocaton algorthm over dfferent nterfaces wth an objectve to optmze multuser performance; the target of ths algorthm s to acheve maxmum throughput. An explct adaptve traffc allocaton scheme []s based on the cooperaton of wreless wde area networks and wreless personal area network. Ths method can acheve nternetworkng load balance and mnmze the whole transmsson delay. For MM servce, t also can be transmtted n HWEs. Inthspaper,weproposeaneffectverateallocatonscheme for MM servce, whch depends on the utlty functon. The utlty functon takes both the throughput and the cost nto account. It ams to maxmze the throughput and mnmze the cost accordng to the characterstcs of MM servce. The rest of ths paper s organzed as follows. The MM communcatons scenaro s descrbed n Secton. Secton 3 provdes an optmal soluton for dynamc traffc allocaton and descrbes our smulaton results, and Secton 4 concludes the paper.. MM Communcatons MTC sever Internet RA Server.. MM Communcatons Scenaro. In a cell, there are massve MM devces, for example, real-tme vdeo survellance of ntellgent transportaton MM systems and vdeo replay transmsson of envronment montorng MM systems. It wll result n network congeston when massve MM devces smultaneously access the same network. If MM servce can be transmtted n multple networks, t s known then that network congeston wll be decreased. However, f an MM devce can access dfferent rado access networks (RAs), ts prcewllbencreased,andtwllconsumemoreenergyfor lstenng to more RAs to access. In fact, t s unnecessary for each MM devce to access dfferent RAs. Meanwhle, MM devces do not move, move nfrequently, or move only wthn a certan regon. For certan management purposes, manymmdevcescanbegroupedasclusters[]. Basedonthedscussonabove,tsassumedthatdfferent applcaton MM devces can be dvded nto dfferent groups, and all of the MM devces data are sent to the access node. Thus, the access node can operate groups, MM devces, and there wll be large MM servce data on the access node. Between MM devces and the access node communcatons, many knds of wreless technologes can be nvolved, such as, W-F, ZgBee, RFID, and Bluetooth. And many measures can be taken to guarantee the access of these dfferent MM devces groups. For example, n the back off scheme, dfferent MM devces groups can take dfferent back off wndows, whch can effectvely decrease congeston [4]. In ths paper, the communcatons from MM devces to the access node are not consdered. And t s assumed that the access node can access dfferent RAs and operate massve MM devces. Ths assumpton can reduce the MM devce prce and access collsons among MM devces. In addton, f an MM devce data are large, t also can be vewed as the access node. In Fgure, An MM communcaton scenaro s llustrated.odeastheaccessnode.itmeansthatothergroups MTC devce ode A Group Group Fgure : An MM communcaton scenaro n HWEs. of MTC devces transmt data to the node A. Therefore, the datanthenodaaarelarge.andthenodeacanaccess multple RAs. For dfferent RAs, some RAs can offer shorter data transmsson delay, and some RAs can provde lower servce expense. Meanwhle, dfferent MM servces have dfferent requrements, such as delay. Hence, how to allocate the rate n the dfferent RAs should be solved. For MM servce, two mportant attrbutes areobservednthspaper,thethroughputandthecost.t(λ) and C(λ) represent them, respectvely. When the remanng resources of RAs are known, the proposed scheme can dstrbutetheratenthedfferentrasbasedonthemm servce characterstcs. Frstly, some of varables are explaned as follows. R bpssthetotaldatarate. L p bt s the packet length. λ (λ = R/L p ) s the packet arrval rate and follows the Posson dstrbuton. λ s the allocaton rate n the th RA, where =,...,. s the number of RAs. R a bps s the avalable resources of RAs. μ (μ = R a /L p) s the servng rate and follows the Posson dstrbuton.
Internatonal Dstrbuted Sensor etworks 3 T(λ) s the total gan functon. C(λ) s the total cost functon. The data transmsson through RAs s modeled as M/ M/ queue. Accordng to [], the steady-state probabltes of m packets n th RA can be wrtten as P RA m =( λ m ) ( λ ). () μ μ Consequently, the average packets L transmtted by the th RA are deduced from the followng formula []: L = m( λ m ) ( λ λ )=. () m μ μ μ λ Hence, the average delay T(λ ) n the th RA can be calculated as T(λ )= L =. (3) λ μ λ To some extent, mnmzaton of the transmsson delay can proportonally maxmze the throughput; thus, the total gan functon T(λ) s gven as T (λ) =. (4) μ λ Addtonally, for the th RA, b s the transmsson cost, whch means that the MM servce wll cost b when a certan number of packets are transmtted by the th RA, and the number s assgned as the unt quantty. If the MM packet length and the MM servce expense are known, then b can be obtaned. For example, f the packet length s bts and the servce expense s. Yuan per bt, b canbecalculated as. =. Yuan per packet. In other words, b s the known nformaton n the RA. Meanwhle, L s the number of the average transmtted packets n the th RA. The cost n the thracanbewrttenas C(λ )=L b = λ b. (5) μ λ The objectve C(λ ) s the ncreasng functon of λ.then,the total cost functon C(λ) s gven as λ C (λ) = b. (6) μ λ Let U k represent the utlty functon of the MM servce k.itscalculatedas U k =β T (λ) +β C (λ), (7) where β, β are weghtng factors. If the MM servce requres hard tmng constrants, then β >β.whenβ =, β =, the throughout s only consdered n ths stuaton. If the MM servce requres non real-tme and low cost, then β <β,andevenβ =, β =.Inthsstuaton,thecosts only consdered. The values of both β and β depend on the MM servce characterstcs. Because T(λ) means the overall delay and C(λ) means the total cost, smaller values of both T(λ) and C(λ) wll be better obtaned n the actual system. Hence, mnmzng U k can get better utlty for MM servce k... Optmal Soluton. When the utlty functon U k s obtaned by (7), there are also some constrants n U k optmzaton.... Allocaton Constrant. The delay of MM servce k s determned by the maxmum delay D k,whchsthetme from the MTC devce to the MTC sever n all RAs. D s the other delay of uplnk traffc towards the destnaton n the th RA. It s defned as /(μ λ )<D k D.... Allocaton Constrant. The total rate of MM servce dstrbuted n RAs s calculated by λ= λ...3. Allocaton Constrant 3. In an M/M/ queue wth arrval rate of λ and servng rate of μ,tsknownthat λ < μ. The rate allocaton of MM servce k should mnmze (7) by satsfyng these constrants. Thus, the optmzaton problem can be modeled as an objectve wth constrants as follows: mn U k s.t. { λ= { { λ <μ λ μ λ <D k D, where,,...,. To show the concavty of the objectve functon U k,the followng general form s observed: U k =β T (λ) +β C (λ) =β λ +β μ λ b. μ λ The frst dervatves of T(λ) and C(λ) are T (λ) = λ (μ λ ), C (λ) b = μ λ (μ λ ). The second dervatves of D(λ) and C(λ) are T (λ) = λ (μ λ ) 3, C (λ) b = μ λ (μ λ ) 3. (8) (9) () () As long as μ > λ, T(λ)/ λ >, C(λ)/ λ >, T(λ)/ λ >, C(λ)/ λ >,then U k / λ >, U k / λ >. Therefore U k =β T(λ) + β C(λ) s a convex
4 Internatonal Dstrbuted Sensor etworks functon n λ. The Lagrange functon can be wrtten as follows: λ L(λ,θ,] )=β +β μ λ b θ( λ μ λ λ) ] (μ λ D k D ) w (μ λ ). = () Accordng to /(μ λ )<D k D, t can be calculated that λ <μ (/(D k D ))(/(D k D )>). Meanwhle, λ <μ. That s to say that, f λ <μ (/(D k D )),tsknownthat λ <μ. Hence, the Lagrange functon can be expressed as λ L(λ,θ,] )=β +β μ λ b θ( λ μ λ λ) ] (μ D k D λ ). = (3) In order to deal wth the Lagrange functon, the Karush- Kuhn-Tucker (KKT) condtons [4] are used. Then,(4), (5), (6), and (7) are obtaned as follows: β (μ λ ) + β b μ (μ λ ) θ+v =, (4) μ V (μ D k D λ >, (5) D k D λ )=, (6) = λ λ=, (7) where,...,.becauseμ /(D k D ) λ >,wecan get V =.Then(4) can be expressed as β (μ λ ) + β b μ (μ λ ) θ=, thus θ>. (8) The frst dervatve of L(λ,θ,] ) about λ s L λ = β (μ λ ) + The second dervatve of L(λ,θ,] ) about λ s L = λ β (μ λ ) 3 + β b μ (μ λ ) θ. (9) β b μ (μ λ ) 3. () Let f(λ k ) and f (λ k ) represent L/ λk and L/ (λ k ) respectvely, where k represents kth teraton. It has been proved that the objectve functon (9) sconcaveandthat f(λ k ) s ncreasng n λ wth fxed β, β, b, μ for f (λ k )>. () Collect the avalable resources nformaton from each RA. () If k=,then (3) Intalze λ =and θ = (4) Else (5) Calculate usng ewton s method. λ k+ =λ k f(λk ) f (λ k ) (6) Updated θ k+ accordng to () θ k+ =[θ k +ε( λ k λ)] (7) k = k +.Gotostep(5) untl the dfference between λ k+ and λ k values are less than a mnmum, such as.. (8) End f (9) After the algorthm converges the rate s allocated accordng to λ k+. Algorthm Meanwhle, f (λ k )>. Therefore, we can use the ewton teraton method [5] to get the converged optmal value of λ. Regardless of ntal values λ,thelastconvergencevalue λ k wll satsfy (8). Then, a better approxmaton λ k+ can be wrtten as λ k+ =λ k f(λk ) f (λ k ). () The updated θ k value for rate allocaton s calculated by θ k+ =[θ k +ε( n + λ k λ)] +, () where ε(ε>)s a constant step sze and [θ k ] + = max(θ k,). The teraton wll contnue untl the dfference between λ k+ and λ k s less than a mnmal value, whch can be predefned by the algorthm. Because λ s the convergence value, λ k wll more approach the actual value f the number of teratons s more. In other words, the mnmal value controls the accuracy. The smaller the mnmal value, the hgher the accuracy. Then, the allocaton rate λ among RAs s obtaned basedon() and(), and t s the global optmal results of the convex optmzaton problem for (8)..3. Algorthm Flow. In concluson, we use an teratve algorthm to fnd the soluton of optmzaton problem, and t s shown n Algorthm. 3. Smulaton Results In ths secton, the performances of the proposed traffc allocaton scheme are evaluated. In the numercal smulaton, two dfferent MM servces are nvestgated. One MM servce requres hard tmng constrants; β =.9 and
Internatonal Dstrbuted Sensor etworks 5.5 3 The date rate of each path (Mbps).5.5 The date rate of each path (Mbps).5.5.5 3 4 5 6 7 Iteraton (k) RA, soft tmng constrants RA, soft tmng constrants RA 3, soft tmng constrants RA, hard tmng constrants RA, hard tmng constrants RA 3, hard tmng constrants 5 5 5 Iteraton (k) RA, soft tmng constrants RA, soft tmng constrants RA 3, soft tmng constrants RA, hard tmng constrants RA, hard tmng constrants RA 3, hard tmng constrants Fgure : Two MM servces n HWEs wth b=[ ]. Fgure 3: Two MM servces n HWEs wth b=[ 4 8]. β =. are defned. The other MM servce requres soft tmng constrants and low cost; β =. and β =.9 are defned. The data rate R s6mbpsforbothmmservces. Meanwhle, there are three RAs. It s assumed that the avalable resources are 4 Mbps for each RA. For comparson, two dfferent cost combnatons are defned for these three RAs. The frst cost combnaton s b=[ ] for RAs,, and 3, whch means that the MM servce wll consume when M packets are, respectvely, transmtted by RAs,, and 3. The second combnaton s b=[ 4 8] for RAs,, and 3. In addton, the rado channel envronment s good to evaluate the performance of the proposed algorthm. 3.. Two MM Servces n HWEs. Fgures and 3 show two MM servces n the same HWEs, respectvely. Fgures 4 and 5 showthesamemmservcenthedfferenthwes, respectvely. In Fgure, the allocated rates are almost equal n all RAs because the cost of each RA s. Ths s also consstent wth the actual stuaton. As mentoned above, the hard tmng constrants MM servce pays more attenton to the delay, and the soft tmng constrants MM servce focuses on the cost. Hence, the allocated rates are dfferent when the cost of RAs s charged n Fgures 3, 4, and 5. Meanwhle, for the soft tmng constrantsmmservce,becausesthelowestcostofras, the allocated rate s the hghest n the three RAs. BasedonFgures, 3, 4,and5,becausethetwodfferent MM servces have dfferent characterstcs, the convergence values are dfferent n the three RAs by the ewton method. And these fgures ndcated that the proposed scheme s effectve n rate allocaton for MM servce. 3.. The Cost and Delay Comparson. In Fgure 6, themaxmum delay of soft tmng constrants MM servce s hgher The date rate of each path (Mbps) 3.5.5.5 5 5 5 3 RA, b=[ 4 8] RA, b=[ 4 8] RA 3, b=[ 4 8] Iteraton (k) RA, b=[ ] RA, b=[ ] RA 3, b=[ ] Fgure 4: Soft tmng constrants n HWEs wth b=[ ] and b=[ 4 8]. than that of hard tmng constrants MM servce. For hard tmng constrants MM servce, because t pays more attentontothedelay,thedelaysareequalnhweswththedfferent cost. In Fgure 7, the sum cost of soft tmng constrants MM servce s lower than that of hard tmng constrants MM servce. Fgures 6 and 7 ndcate that the hard tmng constrants MM servce requres lower delay, whle the soft tmng constrants MM servce focuses on lower cost. In addton, because b = [ ], the cost of hard tmng constrants
6 Internatonal Dstrbuted Sensor etworks 3 The date rate of each path (Mbps).5.5.5 Cost 8 6 4 3 4 5 RA, b=[ 4 8] RA, b=[ 4 8] RA 3, b=[ 4 8] Iteraton (k) RA, b=[ ] RA, b=[ ] RA 3, b=[ ] Fgure 5: Hard tmng constrants n HWEs wth b=[ ] and b=[ 4 8]. RA RA RA 3 Sum cost b=[ 4 8], soft tmng constrants b=[ ], soft tmng constrants b=[ 4 8], hard tmng constrants b=[ ], hard tmng constrants Fgure 7: The cost comparson..8 4 Delay.6.4. The utlty 8 6 4 RA RA RA 3 Max delay b=[ 4 8] soft b=[ ] soft b=[ 4 8] hard b=[ ] hard b=[ 4 8], soft tmng constrants b=[ ], soft tmng constrants b=[ 4 8], hard tmng constrants b=[ ], hard tmng constrants Fgure 6: The delay comparson. MMservcesalmostthesamewththecostofsofttmng constrants MM servce. 3.3. Algorthm Comparson. In order to valdate the proposed rate allocaton scheme, the commonly used load balancng scheme [] s brefly gven as comparson as shown n Fgure 8. Itshouldbenotedthatlowervaluemeansbetterutlty.As s shown from Fgure 8, the proposed scheme can get better performance than the compared scheme, when the MM servce requres longer delay and lower cost. Meanwhle, the valuesoftheutltyarealmostequalwhenb=[ ] and the MM servce requres hard tmng constrants, because the dstrbuted rates are the smlar values as depcted n Fgure. The proposed scheme The compared scheme Fgure 8: The utlty comparson between two schemes. 4. Conclusons In the future MM communcatons, MM servce may be transmtted n HWEs. The rate allocaton for MM servce among mult-ras s an mportant and challengng research. In ths paper, an adaptve rate allocaton scheme s proposed based on the utlty functon. Accordng to the characterstcs of MM servce, the utlty functon may be dfferent. And theratesdstrbutedbasedontheutltyfunctonnhwes to acheve hgher throughput and lower cost. The smulaton showsthattsaneffectveonecomparedwthtradtonal method. Acknowledgments Ths work was sponsored by the Projects 6 and 6975 supported by the atural Scence Foundaton
Internatonal Dstrbuted Sensor etworks 7 of Chna, the Project IRT49 supported by the Program for Changjang Scholars and Innovatve Research Team n Unversty, Chnese Hghway Insttuton Project ZX35-3, and Chna Scholarshp Councl. References [] Servce requrements for machne-type communcatons, 3GPP TS. 368 V. 4.,. [] System mprovement for machne-type communcatons, 3GPP TR 3. 888 V. 6.,. [3] S.-Y. Len and K.-C. Chen, Massve access management for QoS guarantees n 3GPP machne-to-machne communcatons, IEEE Communcatons Letters, vol.5,no.3,pp.3 33,. [4] S.-Y. Len, K.-C. Chen, and Y. Ln, Toward ubqutous massve accesses n 3GPP machne-to-machne communcatons, IEEE Communcatons Magazne,vol.49,no.4,pp.66 74,. [5] M. Starsnc, System archtecture challenges n the home MM network, n Proceedngs of the Applcatons and Technology Conference (LISAT ), pp. 7, IEEE press, Farmngdale, Y, USA, May. [6] I. Cha, Y. Shah, A. U. Schmdt, A. Lecher, and M. Meyersten, Addressng new securty threats, IEEE Vehcular Technology Magazne,vol.4,no.3,pp.69 75,9. [7] CATT, R-8: access control of MTC devces, 3GPP TSG RAWGMeetng68bs,. [8] ZTE, R-466: MTC smulaton results wth specfc solutons, 3GPP TSG RA WG Meetng 7,. [9] S.-Y. Len, T.-H. Lau, C.-Y. Kao, and K.-C. Chen, Cooperatve access class barrng for machne-to-machne communcatons, IEEE Transactons on Wreless Communcatons,vol.,no.,pp. 7 3,. [] Servce requrements for machne-type communcatons, 3GPP TS. 368 V..,. [] W. Fu and D. P. Agrawal, Mult-connecton and rate allocaton n heterogeneous wreless networks, n Proceedngs of the IEEE Global Communcatons Conference Workshops (GLOBECOM ),pp.57 6,Mam,Fla,USA,December. [] L. Sun, H. Tan, Q.-Y. Sun, D.-M. Shen, and P. Zhang, Traffc allocaton scheme wth cooperaton of WWA and WPA, IEEE Communcatons Letters,vol.4,no.6,pp.55 553,. [3] E. Z. Tragos, G. Tsropoulos, G. T. Karetsos, and S. A. Kyrazakos, Admsson control for QoS support n heterogeneous 4G wreless networks, IEEE etwork,vol.,no.3,pp.3 37,8. [4] B. Chen, Optmzaton Theory and Algorthms, Tsnghua Unversty Press, Bejng, Chna, nd edton, 5. [5]J.Stoer,R.Bulrsch,R.Bartels,W.Gautsh,andC.Wtzgall, Introducton to umercal Analyss, Sprnger, ew York, Y, USA, 3rd edton,.
Internatonal Rotatng Machnery Engneerng The Scentfc World Journal Internatonal Dstrbuted Sensor etworks Sensors Control Scence and Engneerng Advances n Cvl Engneerng Submt your manuscrpts at Electrcal and Computer Engneerng Robotcs VLSI Desgn Advances n OptoElectroncs Internatonal avgaton and Observaton Chemcal Engneerng Actve and Passve Electronc Components Antennas and Propagaton Aerospace Engneerng Internatonal Internatonal Internatonal Modellng & Smulaton n Engneerng Shock and Vbraton Advances n Acoustcs and Vbraton