Wireless New 27 13:335 349 DOI 1.17/s11276-6-7529-7 QoS Mechanisms for he MAC Proocol of IEEE 82.11 WLANs José R. Gallardo Paúl Medina Weihua Zhuang Published online: 3 July 26 C Science + Business Media, LLC 26 Absrac There are wo essenial ingrediens in order for any elecommunicaions sysem o be able o provide Qualiyof-Service QoS guaranees: connecion admission conrol CAC and service differeniaion. In wireless local area neworks WLANs, i is essenial o carry ou hese funcions a he MAC level. The original version of IEEE 82.11 medium access conrol MAC proocol for WLANs does no include eiher funcion. The IEEE 82.11e draf sandard includes new feaures o faciliae and promoe he provision of QoS guaranees, bu no specific mechanisms are defined in he proocol o avoid over sauraing he medium via CAC or o decide how o assign he available resources via service differeniaion hrough scheduling. This paper inroduces specific mechanisms for boh admission conrol and service differeniaion ino he IEEE 82.11 MAC proocol. The main conribuions of his work are a novel CAC algorihm for leaky-bucke consrained raffic sreams, an original frame scheduling mechanism referred o as DM-SCFQ, and a simulaion sudy of he performance of a WLAN including hese feaures. This work has been parly funded by he Mexican Science and Technology Council CONACYT hrough gran 38833-A. J.R. Gallardo. P. Medina Elecronics and Telecommunicaions Deparmen, CICESE Research Cener Km. 17 Carreera Tiuana-Ensenada, Ensenada, Baa California 2286, México e-mail: gallard@cicese.mx W. Zhuang Deparmen of Elecrical and Compuer Engineering, Universiy of Waerloo 2 Universiy Avenue Wes, Waerloo, Onario N2L 3G1, Canada Keywords Wireless local area neworks. Medium access conrol. Qualiy of Service. Connecion admission conrol. Packe scheduling. Leaky-bucke consrained raffic 1. Inroducion IEEE 82.11-based wireless local area neworks WLANs have been widely deployed in home and business environmens. The growing ineres in new applicaions over WLANs, such as voice, video and mulimedia sreaming, also brings unique and challenging qualiy-of-service QoS requiremens. The wo main ingrediens for any elecommunicaions sysem o be able o provide QoS guaranees are connecion admission conrol CAC o avoid over sauraing he medium, and service differeniaion o give each user wha i needs and when i needs i. So far, he popular approaches o provide mulimedia users wih QoS guaranees, proposed by he Inerne Engineering Task Force IETF, are InServ [4], and DiffServ [3], along wih all he underlying echnologies and proocols. These resource-reservaion and service-differeniaion mechanisms are implemened a he nework or Inerne Proocol IP layer. This means ha mobile erminals in a WLAN will no be able o receive he reamen hey deserve if here is no service differeniaion when hey compee for he wireless medium a he medium access conrol MAC level. The MAC proocol of he original IEEE 82.11 sandard conains an opional mechanism for providing realime applicaions wih a proper QoS service. This mechanism, known as Poin Coordinaion Funcion PCF, allows real-ime applicaions o periodically ransmi conenionfree frames during cerain inervals, called Conenion-Free Periods CFPs. The provided guaranee is ha a leas one frame of one of he regisered saions will be sen during
336 Wireless New 27 13:335 349 each CFP. This is no enough in mos siuaions since no all real-ime applicaions have he same QoS needs in pracice. There are some proposed exensions o his MAC proocol under discussion, documened in he IEEE 82.11e draf sandard. The new feaures allow for raffic sreams o indicae heir characerisics and needs hrough wha is called TSPEC, which in urn can be used by he resource allocaor for admission conrol or he assignmen of ransmission opporuniies TXOPs. I is also possible for he conenion-free periods o sar anywhere wihin he ransmission superframe, as needed o saisfy QoS crieria of delay-sensiive applicaions. However, he imporan issues of how o locally limi he amoun of raffic ha compees for he wireless medium and how o decide which backlogged real-ime connecion o service firs and which nex remain open. The main goal of his research work is o idenify specific admission conrol and service differeniaion scheduling mechanisms so ha IEEE 82.11 WLANs can effecively offer QoS guaranees o heir users. The remainder of his paper is organized as follows. Secion 2 briefly reviews he IEEE 82.11 and 82.11e MAC proocols. Secion 3 proposes he Defici-based Modified Self-Clocked Fair Queueing DM-SCFQ algorihm for packe scheduling. Deails of he proposed CAC algorihm are presened in Secion 4. Performance of he proposed QoS suppor mechanisms are evaluaed in Secion 5 via compuer simulaions, followed by concluding remarks in Secion 6. 2. IEEE 82.11 and 82.11e MAC proocols The IEEE 82.11 MAC proocol [5, 9] defines wo ransmission modes for daa packes: he CSMA/CA-based Disribued Coordinaion Funcion DCF, and he conenionfree Poin Coordinaion Funcion PCF, where here is a Poin Coordinaor ha conrols all ransmissions based on a polling mechanism. The poin coordinaor usually resides in he Access Poin AP. The DCF and PCF modes are ime muliplexed in a superframe, which is formed by a PCF conenion-free period followed by a DCF conenion period. The boundaries beween conenion-free periods and conenion periods are marked by beacons ransmied by he access poin. During a conenion period, all saions compee wih each oher o gain access o he medium. All saions are equally likely o ge o ransmi a frame since here is no mechanism in he proocol o give some saions more righs han ohers. During he conenion-free period, on he oher hand, he poin coordinaor in he access poin polls erminals in he polling lis and acceps heir response frames. Each erminal ha wans he PCF o schedule is ransmissions needs o submi a reques o he access poin during a conenion period. However, he mehod used o selec he saions o be polled does no ake ino accoun he saions needs, and here is no guaranee of a frame delivery wihin a cerain imeframe. So, he MAC proocol defined in he IEEE 82.11 sandard canno fulfill he QoS requiremens of mulimedia applicaions. Since he PCF has limied capabiliy o deliver QoS and has no been widely implemened, he IEEE 82.11 Task Group E has underaken he definiion of a new sandard, namely IEEE 82.11e [8, 1, 12], in which a single coordinaion funcion is inroduced, wih hree access schemes: Enhanced DCF, Conrolled Access, and Conrolled Conenion. The new coordinaion funcion is known as Hybrid Coordinaion Funcion HCF because here is no clear disincion now beween wha was previously known as conenion period and conenion-free period. The HCF inroduces a Hybrid Coordinaor, which exends he noion of poin coordinaor in he sense ha i can gain access o he channel during he conenion periods o creae addiional conenion-free periods as needed o saisfy QoS crieria. Enhanced DCF is sill a conenion scheme, bu i has now he capaciy o give some saions a higher prioriy han ohers by manipulaing boh he ime ha a saion senses he medium o decide if i is free called AIFS and he backoff ime. The AIFS period is shorer for higher-prioriy saions allowing hem o win access o he medium over he lower-prioriy ones. Likewise, even hough he backoff ime is seleced randomly, he se of possible values for a higherprioriy saion includes smaller values han hose allowed for a lower-prioriy one, esablishing again an advanage for he former. Conrolled access, on he oher hand, is a polling mechanism ha akes ino accoun he TSPECs of he differen raffic sreams when admiing a new session ino he polling lis and when deciding how o assign ransmission opporuniies. Finally, conrolled conenion is used when he hybrid coordinaor decides ha i has some exra ransmission opporuniies available for disribuion. I sends a message o a specific se of saions inviing hem o place new resource requess. Even hough IEEE 82.11e includes several new feaures ha faciliae and promoe he provision of QoS guaranees o he users, here are no specific mechanisms in he proocol o avoid sauraing he medium wih excessive raffic or for he hybrid coordinaor o decide how o assign he available resources. I is here ha our proposed schemes find heir usefulness. 3. DM-SCFQ scheduling mechanism Our proposed QoS suppor schemes ake advanage of he new feaures inroduced in he IEEE 82.11e MAC proocol.
Wireless New 27 13:335 349 337 The obecive is o provide QoS guaranees o hose sessions included in he polling lis i.e., regisered o receive conenion-free ransmission services, in he conrolled access scheme. The Generalized Processor Sharing GPS algorihm and is pracical realizaion Weighed Fair Queueing WFQ are he mos imporan packe scheduling disciplines nowadays due o heir unsurpassed fairness and predicabiliy [16,17]. This is he reason why several elecommunicaions sandards have adoped WFQ as par of heir QoS mechanisms. However, i is no pracical o implemen WFQ in a WLAN environmen due o he fac ha he packe queues and he medium conroller are no co-locaed. In oher words, he packes waiing o be serviced are queued in he mobile saions, while he hybrid coordinaor is locaed in he access poin. I is necessary for every mobile saion o somehow inform he hybrid coordinaor of how many packes i has in is queue, and of he arrival ime and size of each packe. Forunaely, here are alernaives o WFQ ha also work adequaely and ha do no need as much informaion. In his research, we propose Defici-based Modified SCFQ, which combines he ideas embedded in Self-Clocked Fair Queueing SCFQ [6], wih hose in Disribued Defici Round Robin DDRR [18]. DM-SCFQ is similar o DDRR in he sense ha i firs allows each raffic source o ransmi a packe and hen i les i wai long enough o pay for he ransmission opporuniy already given before allowing i o ransmi again; i.e., i is based on defici. DM-SCFQ is also similar o SCFQ in ha i relies on a self-generaed virual ime. In he DM-SCFQ scheduling, he same raffic source is polled repeaedly unil all he frames corresponding o a single packe are ransmied; he ransmission of all he frames corresponding o he same packe couns as a single ransmission opporuniy. To decide which raffic source will be allowed o ransmi nex, a service ag is compued for each of he sources a he end of a ransmission opporuniy. Since, as menioned above, we canno have informaion relaed o he packes waiing in disan queues a he mobile saions wihou incurring a high overhead cos, we compue he service ag for deciding he ransmission of he i-h packe coming from source as follows: F i = Ei r + F i 1 1 where r is he share of channel capaciy allocaed o session, and E i is equal o he lengh of he previously ransmied packe L i if he mos recen poll sen o saion resuled in a ransmission and he More Daa bi of he frame is se o 1, or n L i oherwise. The value of he consan n is a design parameer and is used o avoid wasing an excessive amoun of bandwidh by polling repeaedly a saion ha has nohing o ransmi for he momen. We used a value of n = 2 in our simulaions and he resuls are saisfacory. E i can be hough of as he effecive packe lengh. Equaion 1 is equivalen o assuming ha all raffic sources are consanly backlogged and we charge hem for all of he ransmission opporuniies ha hey are given, regardless of wheher hey use hem o acually send packes or no. Charging for each ransmission opporuniy is done a poseriori. DM-SCFQ has wo main advanages as compared o WFQ, which are inheried from he scheduling mechanisms i is derived from: I is compuaionally more efficien since i relies on a selfgeneraed virual ime, and I does no need o know he arrival ime or size of he packe a he head of each queue, since i makes is decisions based on previously ransmied packes. For comparison purposes, we also presen in his paper resuls relaed o an implemenaion of WFQ in a WLAN. As menioned before, in order o implemen WFQ he access poin needs o know abou he packes waiing in each mobile saion s queue. This is done by leing every saion inform he access poin of how many new packes have arrived o is queue, along wih a lis of heir sizes, every ime he saion has a chance o send a daa frame. To say updaed, he access poin polls all saions periodically every m superframes, which is again a design parameer ha in our case was se o 2. We call his a cyclic poll. The res of he ime, he access poin polls only he winning saions, according o he rules of he relevan scheduling mechanism. This mechanism has he added disadvanage ha i needs he modificaion of he MAC frame srucure o include a new field used o exchange he queue-occupancy updae informaion. 4. The proposed CAC algorihm One of he opions included in IEEE 82.11e o specify he raffic characerisics and needs is hrough leaky-bucke parameers plus an upper bound for he olerable delay. This opion assumes ha he raffic is shaped before i eners he WLAN. This opion is criical because of he relaively limied capaciy of he wireless medium. When raffic complies wih his resricion, i is possible o offer deerminisic bounds for he delay ha he raffic will experience when serviced using a fair scheduling policy, such as GPS [14, 16]. In oher words, hese resuls can be used o devise a CAC mechanism for delay-sensiive raffic as long as a fair scheduling algorihm is being used. The oupu of he CAC mechanism would be a decision as o wheher a se of raffic sreams can be serviced wihou violaing heir QoS guaranees, as well as he fracion of he available bandwidh ha has o be allocaed o each one of hese raffic sreams. Resource reservaion is implici in such a CAC mechanism.
338 Wireless New 27 13:335 349 Recenly, an expression has been found in [2, 14, 2], for he igh delay bounds ha characerize GPS sysems wih leaky-bucke consrained inpu raffic sources, and CAC algorihms have been proposed based on such bounds. Unforunaely, he previous algorihms do no ake full advanage of he available informaion. In he remainder of his secion we propose a CAC algorihm o maximize bandwidh uilizaion by allocaing each session only he porion of he available bandwidh ha is sricly needed so as no o exceed a given delay bound. In addiion, he proposed CAC algorihm does a sraighforward allocaion based on sysemaic calculaions, wihou he need of random search or successive approximaions, as is done in [2]. Our algorihm is designed for a single-node sysem bu, as explained in [15], i can be exended o muliple-node sysems in many siuaions of pracical ineress. For insance, when i is applied a he boleneck node, or when i is applied a an edge node and he same bandwidh is reserved for each connecion a all he nodes included in is end-o-end pah. 4.1. Terminology and previous resuls A raffic source is said o be leaky-bucke consrained if is arrivals in any given inerval 1, 2 ], denoed by A 1, 2, saisfy he following inequaliy: A 1, 2 σ + ρ 2 1 2 When he equaliy holds for 1 = and all values of 2, he raffic source is called greedy. A GPS scheduler serving N sessions is characerized by a se of posiive real numbers φ 1,φ 2,...,φ N, which denoes he relaive amoun of service given o each session in he sense ha, if session i is coninuously backlogged during he ime inerval 1, 2, hen: W i 1, 2 W 1, 2 φ i φ = 1, 2,...,N 3 where W i 1, 2 is he amoun of service given o session i during he ime inerval 1, 2. As menioned above, a igh bound is found in [2, 14], and [2], for he maximum delay ha a leaky-bucke consrained session experiences in a GPS server. The relevan bound is igher han ha found in [17], as i explicily akes ino accoun ha he bandwidh unused by some sessions will be redisribued among he backlogged sessions proporionally o heir respecive weighs φ i. I also akes ino accoun informaion relaed o he specific ime insans a which bandwidh will be released by each session and herefore used by he oher sessions ha sill need i. In his work, each session will be assumed o be greedy and he incoming raffic of session will be characerized by ρ and σ, which are he parameers of is respecive leakybucke. Le L = L i i = 1, 2,...,N denoe an ordered se of indices meaning ha he backlog of session Li is cleared i h in order, and le C be he server capaciy. The ime insan when he backlog of session Li is empied is denoed by L i. By definiion, le L = be he sar of a sysem busy period. If he buffer of session is iniially empy bu sars o build up because is iniial allocaed bandwidh is less han is raffic arrival rae, hen is backlog-clearing ime is no =, bu he ime when he buffer goes back o being empy. In addiion, le r denoe he bandwidh or service rae effecively allocaed o session a ime, including is iniial bandwidh r = φ C plus is share of bandwidh released by oher sessions. To simplify noaion, le r k be he session- service rae during he ime inerval [ Lk 1], Lk. I is shown in Eq. 5 of [14], ha, when N =1 r = C he server is iniially sauraed, r k saisfies he following equaion: r k = coef k 1 r 4 where C coef k = C k ρ Lm m=1 k m=1 r Lm. 5 Equaion 4 is valid as long as he buffer of session is no empy. Afer is own backlog has been cleared i.e., >, session will be serviced a rae ρ, which is he raffic arriving rae. Because of his, a necessary condiion for he sysem o be able o properly serve all he sessions afer all he backlogs have been cleared is ha N =1 ρ C. Le τ denoe he ime a which he iniial burs of session has been serviced, i.e., he ime when W,τ = σ. Le T denoe he ime a which A, T = W,, i.e., he ime a which he amoun of raffic received was equal o he amoun of raffic ha has been serviced by ime.ifwele D denoe he delay experienced by he raffic compleing service a ime, hen D = T. I is parly shown in [2], ha, in he wors case when he raffic source is greedy, he following equaliy holds: [ for τ ] D = 1 ρ r τ dτ σ for τ < 6 for > The maximum delay ha session can olerae is denoed by d. The following wo ses are also defined: B1 = σ / d ρ 7 B2 = σ / d <ρ 8
Wireless New 27 13:335 349 339 In he following, we sae he main heorem of [14, 2], and [2]. For a proof, see he references. Theorem 1. Le d denoe he maximum delay ha will be experienced by session. i If r τ ρ, hen d = τ ii Conversely, if r τ <ρ, hen: d = D Lk = Lk 1 ρ [ W, Lk σ ] r <ρ for all < Lk and where k 1, 2,...,N is such ha r ρ for all Lk. In addiion o his resul, a few lemmas are included in [14], ha are used o suppor heir CAC algorihm. Here we enunciae wo of hose lemmas ha are useful for our proposed algorihm. We use he same numbers uilized in he reference o idenify he lemmas. Lemma 1. If session B1 and is maximum-delay requiremen is saisfied d d, hen: i The service rae of session is no less han ρ when he las bi of he iniial burs is being served, i.e. r τ ρ. ii The maximum delay of session is experienced by he las bi of he iniial burs, or equivalenly d = D τ = τ. Lemma 3. If d > Li 1, hen he rae r needed in order for he las bi of he iniial burs o experience a delay equal o d is: σ r = i 1 Lk Lk 1 coef k 1 + d Li 1 coef i 1 k=1 1 where coef k is defined as in Eq. 5. Corollary 3.1. If session B1 hen he rae r shown in Eq. 1 is he opimum allocaion in he sense ha d = d. From Lemma 3 and Corollary 3.1, we can conclude ha, for sessions in B1, he maximum allowed delay will no be exceeded as long as τ is equal o d. Thus he allocaed bandwidh for he sessions always has o be compued using Lemma 3. This is already done in [14], and we do no change his par of he algorihm. 4.2. Our new resuls This secion presens some resuls ha suppor he main conribuion of his research, namely he new CAC algorihm 9 described in nex secion. All of our resuls are inended o suppor he new algorihm for allocaing bandwidh o sessions in B2. The CAC algorihm defines he minimum iniial r ha has o be allocaed o session so ha is maximum delay d does no exceed a value d, a QoS parameer esablished by he user. This ask is equivalen o finding he coefficiens φ 1,φ 2,...,φ N for he GPS server as r = φ C. To achieve he opimum allocaion, we need o define crieria o ensure ha we do no unnecessarily overprovision he sessions. Thus, we need o ake ino accoun he ime a which each session frees some bandwidh afer empying is buffer and o keep rack of he amoun of exra bandwidh ha is redisribued among he backlogged sessions. Our algorihm finds he opimum values r 1, r 2,...,r N one by one. Each sep, indexed by i in our algorihm, is marked by one more session empying is backlog or, equivalenly, reaching is. Lemma 4. D for all. Proof: Noice from Eq. 6 ha D = for all τ and ha D = < for all >. Also noice ha for τ < : r τ dτ = W, >σ 11 Using his resul we can conclude from Eq. 6 ha D < for τ < as well. Definiion. Le be he smalles ime for which he maximum delay d is reached for session, ha is = min D = d 12 The minimum is aken in Eq. 12 o avoid ambiguiies, since here can be more han one value of for which D = d. Theorem 2. A necessary condiion for d no o be smaller han d bandwidh no o be wased is ha d. Proof: Assume ha < d. From Lemma 4, we have D. By definiion, we know ha d = D. Puing all ogeher we have: d = D < d 13 This means ha, if < d, d will ineviably be smaller han d and as a resul bandwidh will be wased. Theorem 3. Assume ha r and r are wo differen values of he iniial rae allocaed o session. Assume also ha
34 Wireless New 27 13:335 349 D and D are he respecive values of he delay experienced a ime. Le τ and τ be he imes a which he iniial burs is serviced corresponding o raes r and r, respecively. Finally, le and be he corresponding backlog clearing imes. If r r hen i r r, min, ii W, W,, min, iii τ τ iv D D for all values of. Proof: Since he rae allocaed o session increases in a muliplicaive fashion every ime some bandwidh is freed by a user, as shown in Eq. 4, we have r r, min,. This in urn implies ha, for he same ime inerval, W, W,. From here we can easily conclude ha τ τ. Using hese resuls and Eq. 6, we have D D for τ [ ] 1 ρ W, σ > for τ < τ [ = 1 ρ W, W, ] for τ < D for < for > 14 which proves ha D D for all values of, as desired. Theorem 4. If / L k k = 1, 2,...,i 1, in order for r i o be equal o ρ, i is necessary ha r = ρ /coef i 1. Proof: Since he backlog of session has no been cleared by L i 1, we can use Eq. 4 o compue r i. Then we have r i = coef i 1 r = ρ 15 Solving for r we obain he desired resul. The following heorem is similar o Lemma 4 in [14], bu i is rephrased o make i more useful for our algorihm. Theorem 5. Assume ha / L k k = 1, 2,, i 1 and ha d Li 1.If r = i 1 k=1 ρ Li 1 d + σ 16 Lk Lk 1 coef k 1 hen τ Li 1 < and D Li 1 = d. Proof: Noice ha: i 1 W, Li 1 = Lk Lk 1 r k 17 k=1 From Eqs. 4, 16 and 17, we can conclude ha W, Li 1 = ρ Li 1 d + σ 18 Since d Li 1, we can conclude from Eq. 18 ha W, Li 1 σ, which in urn implies ha τ Li 1. On he oher hand, by definiion a ime will be idenified as when he iniial burs, plus whaever exra raffic ha has arrived since he beginning, has been served. Or equivalenly, whenw, = σ + ρ. However, again from Eq. 18, we have ha: W, Li 1 = [ σ + ρ Li 1 ] ρ d 19 which indicaes ha a L i 1 here is sill a backlog of size ρ d, and ha has no been reached ye. Tha is, Li 1 <. Now, knowing ha τ Li 1 <, we can subsiue Eq. 18 ino Eq. 6 and obain ha D Li 1 = d. Theorem 6. D is a monoonically increasing funcion for all values of such ha r <ρ. Proof: Firs noice ha he condiion r <ρ can only be saisfied for values of less han, because a he session has o be serviced a a rae higher han is arrival rae ρ in order o empy is backlog. If we assume ha r <ρ and ake he derivaive of Eq. 6, we can see ha dd d and ha dd d = 1 >, for τ 2 = 1 r ρ >, for τ < < 21 In boh cases he derivaive of D is posiive, which indicaes ha i is a monoonically increasing funcion of. The checkpoins for a session are defined as hose ime insans a which he maximum delay d can be achieved i.e., candidaes for [15]. According o Theorem 1, he checkpoins include d iself and hose L k greaer han d, k 1, 2,...,N. Theorem 7. If a bandwidh allocaion for session is such ha D = d and r + <ρ for a given checkpoin d, hen he bandwidh allocaion ha saisfies D = d for he subsequen checkpoin > is such ha r > r.
Wireless New 27 13:335 349 341 Proof: Wihou loss of generaliy, assume ha = Li, for some value of i 1, 2,...,N. From Lemma 3 if = d, or from Theorem 5 if > d,wehave W, = ρ d + σ 22 where W, = r [ i 1 Lk Lk 1 coef k 1 k=1 + Li 1 coef i 1 ] Under he assumpion ha r + <ρ,wehave W, + r + Li <ρ d 23 + σ + ρ Li. 24 Bu, r + = r coef i 1, which implies ha [ ] i r Lk Lk 1 coef k 1 k=1 <ρ Li d + σ 25 or equivalenly ha r < ρ Li d + σ = r i 26 Lk Lk 1 coef k 1 k=1 The las equaliy is obained using Theorem 5. In he following, we discuss how he preceding heorems can be applied o achieve he opimum bandwidh allocaion. The key of he CAC algorihm is o ry o achieve D = d a each checkpoin by allocaing he proper amoun of bandwidh as indicaed by Lemma 3 or by Theorem 5, respecively, depending on wheher is equal o d or no. A sep i, from Theorem 2, if d > Li 1, he only viable candidae for he locaion of is d. If we allocae enough bandwidh so ha τ = d and herefore D d = d and if he allocaed bandwidh is such ha r d + = r i ρ,we have found he opimum bandwidh allocaion for his session, unless here are sessions ha free bandwidh before d in which case we can allocae less bandwidh and sill saisfy he delay requiremen. On he oher hand, if he allocaed bandwidh is such ha r d + <ρ only possible for sessions in B2, according o Lemma 1, he delay will coninue o increase, as indicaed by Theorem 6, wih he maximum value being greaer han d. Therefore, we need o allocae more bandwidh o his session, bu we do no have sufficien informaion ye o know how much more. For session, if here exiss a checkpoin L i 1 greaer han d, i is because he allocaion calculaed a he previous checkpoin L i 2 eiher equal o or larger han d was such ha r + <ρ. From Theorem 7, we can see ha he new bandwidh calculaed a his sep is greaer han he previous one. Again we have o check if his new allocaion is such ha r + Li 1 ρ. If so, we have found he opimum allocaion for his session because here is no more bandwidh freed by any sessions before his checkpoin. On he oher hand, if he calculaed bandwidh is such ha r + Li 1 <ρ,we do no have he opimum bandwidh allocaion and should keep rying a he subsequen checkpoins, if here are any. If all he finie checkpoins have been ried and he opimaliy condiion is no me, using Theorem 4, session will be allocaed an amoun of bandwidh such ha r + =ρ, where is equal o he larges finie checkpoin. This bandwidh allocaion saisfies he delay requiremen since, according o Theorem 1, he maximum delay happens a and, from Theorem 3, his maximum delay is less han d because a smaller amoun of bandwidh would make D = d. There are wo reasons o pospone he decision on he opimum bandwidh allocaion for a given session: eiher we do no ye have enough informaion o decide wha he opimum allocaion is as explained above, or we know how much bandwidh should be allocaed bu here is no sufficien bandwidh available unless some sessions free a porion of heir own bandwidh. In he laer case, we pospone our decision, hoping for oher sessions o free enough bandwidh for use by he relevan session. If a decision canno be made for session a a specific sep i, for eiher of he wo reasons, he session will be moved emporarily ino anoher se referred o as N1 or N2 depending on wheher he original se was B1 or B2, indicaing ha his session is no a suiable candidae for Li. Once we find he opimum bandwidh allocaion for session, i is removed from is original se B1 orb2 and is pu ino anoher se denoed by P. A he end of sep i, we selec L i as he smalles among he poenial backlog-clearing ime values compued a his sep denoed by,i, corresponding o sessions in B1, B2 and P. As derived in [14], he poenial backlog-clearing imes are given by,i = r i Li 1 + σ W Li 1 r i ρ. 27 The seleced session, defined above as Li L, is hen removed from is curren se and placed ino a se denoed by H. If here is a draw a ie beween wo or more sessions because heir corresponding values,i coincide, hen all such sessions are seleced as he new elemens of L because heir allocaed bandwidh will no change afer his poin.
342 Wireless New 27 13:335 349 However, i is possible ha, afer going hrough B1, B2 and P, here is no candidae for Li because all he decisions relaed o sessions in B1 and B2 were posponed. When his siuaion is encounered, if N1 is no empy, i means ha here is no enough bandwidh o serve all he sessions; oherwise we are forced o allocae bandwidh o all he sessions in B2 such ha r + = ρ, where is equal o he larges finie checkpoin. As an ineresing remark, recall ha Eq. 4 is only valid when he server is iniially sauraed, i.e. when N =1 r = C. If we find a he end of he algorihm ha his condiion is no saisfied, i is proposed o repea he process wih a reduced value of C unil he relevan condiion is approximaely me [14]. This approach, alhough no necessary for finding a suiable se of weighs, can be useful o find ou roughly he minimum value of he channel capaciy C min ha would be enough o serve he relevan se of sessions. I is also imporan o poin ou ha a byproduc of his algorihm is he possibiliy o calculae he minimum buffer requiremens for each session, given by B = A W = σ + ρ r z dz. 28 This expression can be easily evaluaed afer knowing he values of r and. We describe our CAC algorihm in more deail in he following secion using pseudo-code, where BW sands for bandwidh. 4.3. Our CAC algorihm Parameers needed: ρ, σ, d for all sessions 1, 2,...,N and C Variables: i, Li, L i, r L i, r aux, C rem, se, forced If sum of all ρ s, 1,2,...,N, is greaer han C Display Impossible o serve all sessions ; Sop algorihm; 1 Separae sessions ino B1 and B2; Define N1 and N2 iniially as empy ses sessions for which decision is posponed for a subsequen sep; Define P iniially as an empy se sessions for which BW has been assigned bu has no been compued ye; Define H iniially as an empy se sessions for which BW has been assigned and has been compued; Define H iniially as an empy se sessions wih he minimum,i a he end of sep i; Make i = 1, L =, L =, C rem = C, forced = FALSE; 2 For all sessions in B1: If L i 1 < d r aux = Lemma 3i, ; // We wan o make τ = d If r aux C rem Calculae,i ; Else se = N1; // Hoping for oher sessions o free enough BW Else // If L i 1 d Display Impossible o serve all sessions ; Sop algorihm; // Session wen ino N1 hoping for BW o be freed by oher sessions, bu i did no happen 3 For all sessions in B2: If L i 1 < d r aux = Lemma3i, ; // We wan o make τ = d If r i ρ If r aux C rem Calculae,i ; Else If! forced se = N2; // Hoping for oher sessions o free enough bandwidh Else Display Impossible o serve all sessions ; Sop algorihm; // No session will free BW afer his poin Else // If L i 1 d r aux = Theorem5i, ; //We wan D L i 1 = d If r i ρ If r aux C rem se = P; // Found opimum allocaion Else Display Impossible o serve all sessions ; Sop algorihm; // BW freed afer his poin will no help If r i <ρ If! forced se = N2; //We have o wai o find opimum Else // If forced
Wireless New 27 13:335 349 343 r aux = Theorem4i, ; // We wan o make r i = ρ If r aux C rem,i = INFINITY; Else Display Impossible o serve all sessions ; Sop algorihm; // No session will free BW afer his poin 4 For all sessions in P: Calculae,i ; 5 If B1 B2 P is no empy Find min = min,i; // I could be INFINITY B1 B2 P Move all sessions wih,i = min ino H; Move all sessions in B1 B2 wih d min ino P; Else // If B1 B2 P is empy If N1 is no empy Display Impossible o serve all sessions ; Sop algorihm; Else // If N1 is empy forced = TRUE; Reurn sessions in N2 ino B2 Repea from sep 3. 6 For all sessions in H: C rem = C rem r aux ; If C rem Li = ; L i =,i; r L i = r aux ; se = H; i = i + 1; Else // If C rem < Display Impossible o serve all sessions ; Sop algorihm; 7 If i < N Reurn sessions in N1 ino B1 and sessions in N2 ino B2 forced = FALSE; Repea from sep 2. Else // If i ==N Display All sessions can be served simulaneously ; Display all he values r, 1, 2,...,N Sop algorihm; 4.4. Comparison of our algorihm o previously proposed ones If we compare our algorihm o hose presened in [2], he main difference is ha he previously proposed algorihms are no based on sysemaic calculaions, bu rely insead eiher on random search echniques or on successive approximaions o find a soluion, which can be very inefficien or even ineffecive, as explained in [15]. If, on he oher hand, we compare our algorihm o ha presened in [14], we can easily see he following differences. Our algorihm akes advanage of draws ies when i comes o finding he smalles backlog-clearing ime a each sep, in he sense ha we can find several opimum allocaion values in a single sep. The algorihm in [14], does no check for draws, which causes he unnecessary repeiion of many calculaions; For sessions in B2, when d Li 1 he algorihm in [14] uses a resul similar o our Theorem 5 o allocae a rae such ha D Li 1 = d. To ensure ha his is he maximum delay ha he session will experience, he algorihm verifies if ha allocaion is such ha r i 1 <ρ r i. If no, he algorihm proceeds o allocae a rae such ha eiher r i 1 = ρ or r i = ρ, depending on wheher i is he firs inequaliy wha is no saisfied or he second. The firs opion represens a poenial under-provision of bandwidh because he rae is reduced from he value corresponding o D Li 1 = d. The second opion represens a poenial over-provision of bandwidh because D Li 1 may have become he maximum delay experienced by session wih D Li 1 < d and r i = ρ. To summarize, he algorihm proposed in [14], can overprovide bandwidh o some sessions, resuling in inefficien use of he resources. More imporanly, i can also under-provide bandwidh, leading o QoS violaions of he sessions involved. Finally, he algorihm presened in [15] is equivalen o he algorihm presened here. I is imporan o menion, however, ha our algorihm was developed independenly as par of an M.Sc. hesis proec [13]. Even hough boh algorihms rely on compleely differen calculaions and comparisons, we can see heir equivalence by noicing ha he algorihm in [15] allocaes weighs derived from Proposiion 5, which in urn is based on Eqs. 9 and 1 o compue wha hey call
344 Wireless New 27 13:335 349 φ and φ +, where is one of he so-called checkpoins. Assigning φ = φ causes ha D = d equivalen o Lemma 3 in his work when is equal o d, or o Theorem 5 when is equal o Li 1 > d. In addiion, ha algorihm makes sure ha he allocaed weighs are such ha φ φ +, which implies ha r + ρ similar o using our Theorem 4. A deailed look a sep B4 of ha algorihm and a seps 2 and 3 of our algorihm reveals ha boh algorihms are based on he same principles. 5. Performance evaluaion In his secion, we analyze he applicabiliy of he specific QoS mechanisms ha we propose for an IEEE 82.11 WLAN. According o he IEEE 82.11e draf sandard, in order for a session o be esablished beween delay-sensiive saions, an ADDTS reques frame has o be received by he HC locaed in he AP. This frame has o include a TSPEC, which in our case consiss of he parameers ρ, σ, and d.ifi is an upsream only he ransmier is locaed in he relevan BSS or bidirecional communicaion boh he ransmier and receiver are locaed in he same BSS, he ADDTS reques frame has o be sen by he ransmiing saion. If i is a downsream communicaion only he receiver is locaed in he relevan BSS, i has o be he receiving saion which iniiaes he service reques. We assume ha here is a higherlayer mechanism for he receiving saion o learn abou he need o esablish he session and abou he corresponding TSPEC parameers. When an upsream or downsream session is requesed, he hybrid coordinaor will divide he maximum allowed delay d by wo, assuming ha here will be a comparable delay a he disan side of he nework. When a bidirecional session is o be esablished, which is no explicily included in he draf sandard, he ransmier will make a reques as in he case of an upsream session. The hybrid coordinaor will idenify bidirecional sessions whose receivers are locaed wihin he same BSS. The parameers σ and ρ are muliplied by wo by he hybrid coordinaor o ake ino accoun he fac ha he wireless medium will be used wice by each frame. The weighs produced by he CAC algorihm corresponding o bidirecional sessions are divided by wo, o indicae he packe scheduling mechanism ha each frame will use he channel wice equivalen o muliplying he lengh of each packe by wo. The variabiliy of he channel characerisics, including he effecive ransmission rae or channel capaciy C, is always a source of concern in a wireless environmen. Any CAC algorihm needs o know he channel capaciy in order o decide if i is possible o serve a se of cliens, and ours is no excepion. We esimaed he average channel capac- Table 1 Traffic characerizaion TRAFFIC TYPE σ bis ρ bps d sec r bps Voice 2,25 27,.2 46,484.82 Video-conference 33, 165,.4 269,135.14 Video Jurassic Park 247, 96,.4 1,235,. WWW 45, 64, 2. 14,391.81 iy empirically via simulaions assuming ha here are no ransmission errors and ha he only facor affecing i is he overhead inroduced by he packe-scheduling porion of he MAC proocol. We found ou ha he channel capaciy is reduced by abou 12% for WFQ and by abou 5% for DM- SCFQ. To ake ino accoun ha he channel effecive rae may be varying, i can be moniored in real ime using he mehod described in [11], and compared each ime wih he minimum capaciy needed C min, which can be esimaed using a repeiion of he CAC algorihm, as described a he end of Secion 4.B. If a any ime he channel capaciy goes below his hreshold, a warning can be issued o all he sessions in he polling lis. I is also possible ha a saion already included in he polling lis decides o decrease or increase is ransmission rae because of a degradaion or improvemen of he wireless medium. The hybrid coordinaor has o keep rack of hese changes and, when deeced, i runs again he CAC algorihm wih he TSPEC values modified accordingly for he saions ha inend o change heir ransmission raes. The parameers are changed o incorporae ha a saion wih a reduced ransmission rae occupies he channel for a longer period in order o ransmi he same informaion, and vice versa. If he oucome of he CAC algorihm indicaes ha i Fig. 1 Simulaion configuraion
Wireless New 27 13:335 349 345 is no longer possible o serve all sessions, a DELTS reques frame will be sen o he saions ha inend o change heir ransmission raes. Anoher imporan ask o be performed by he hybrid coordinaor is o idenify long inaciviy periods of specific sessions in order o drop hem from he polling lis, hus avoiding he wase of reserved resources. The lengh of he maximum inaciviy period can be negoiaed when he session esablishmen is requesed. 5.1. Simulaion model Our simulaion scenario is an infrasrucure nework wih a physical layer compaible wih he IEEE 82.11b sandard running a 5.5 Mbps. Three differen ypes of raffic models are used o simulae erminals using video, voice and WWW services. For video raffic, we use race-driven raffic generaors, one corresponding o he movie Jurassic Park encoded in MPEG1, and he oher corresponding o videoconference encoded in H.263 [19]. For voice raffic, we use a wo-sae Markov chain, as described in [7], in which he voice raffic has been compressed according o he ITU-T G.729 sandard. Finally, for WWW raffic, we use he model described in [1], in which each erminal generaes raffic equivalen o ha produced by 2 users browsing he web. All applicaions run on TCP/UDP-IP. Each of hese raffic sources is characerized wih is corresponding oken-bucke parameers ρ, σ, as described in [21]. Table 1 gives hese parameers, along wih he iniial rae obained from our CAC algorihm, which is he minimum necessary o saisfy he QoS requiremens of each raffic source. In all of he simulaions, we include 4 voice saions and 2 videoconference saions communicaing in pairs. We also include a video source Jurassic Park sending is raffic o a disan receiver hrough he AP. The number of WWW raffic sources varies, depending on he arge nework raffic load and on he scheduling mechanism which influences he Fig. 2 Maximum and average delay comparison when only sessions acceped by he CAC algorihm are acive.7.6 Maximum Delay sec.5.4.3.2 DM-SCFQ WFQ Original Max allowed.1 Voice_1 Voice_2 Voice_3 Voice_4 VideoH263_5 VideoH263_6 VideoJP_7 WWW_8 WWW_9 WWW_1 WWW_11 WWW_12.12 Average Delay sec.1.8.6.4 DM-SCFQ WFQ Original.2 Voice_1 Voice_2 Voice_3 Voice_4 eoh263_5 oh263_6 VideoJP_7 WWW_8 WWW_9 WWW_1 WWW_11 WWW_12
346 Wireless New 27 13:335 349 Fig. 3 Maximum delay comparison when more sessions are acive han hose acceped by he CAC algorihm sauraion 3 2.5 Maximum Delay sec 2 1.5 1 WFQ WFQ-S.5 Voice_1 Voice_2 Voice_3 Voice_4 VideoH263_5 VideoH263_6 VideoJP_7 WWW_8 WWW_9 WWW_1 WWW_11 2.5 Maximum Delay sec 2 1.5 1 DM-SCFQ DM-SCFQ-S.5 Voice_1 Voice_2 Voice_3 Voice_4 oh263_5 oh263_6 VideoJP_7 WWW_8 WWW_9 WWW_1 WWW_11 WWW_12 WWW_13 effecive ransmission rae. Figure 1 illusraes he simulaion configuraion. We use OPNET as our plaform o simulae and evaluae our proposed schemes, as well as o compare i o he widely known WFQ scheduling mechanism. The simulaion ime in all cases is 2 minues wih a warm-up period of 1 minue. We compare he performance of HCF-CA alone o ha of HCF-CA when i is combined wih he QoS mechanisms proposed in his paper. Our emphasis is on he delay experienced by packes from he momen hey are generaed o he momen hey arrive a heir final desinaion afer having used he wireless medium. 5.2. Numerical resuls The following hree experimens are carried ou: Experimen I only sessions acceped by he CAC algorihm are allowed o be acive; Experimen II one WWW exra saion is allowed o ransmi in addiion o hose acceped by he CAC algorihm. We denoe his case wih a leer S o indicae ha he sysem is sauraed; Experimen III one of he WWW saions ha was acceped by he CAC algorihm in he firs seup saion WWW 1 ransmis en imes as much raffic as i specified in is reques. We denoe his case wih a leer M o indicae ha a misbehaving saion is presen. Figure 2 shows a comparison of he maximum and average delay achieved in Experimen I. Wih he original MAC proocol, he maximum delay experienced by he raffic coming from he differen sources is raher random, as no differeniaion is exercised. The maximum accepable delay is exceeded for several sessions. Conrary o his, he maximum delay achieved for boh WFQ and DM-SCFQ is below he maximum allowed delay in all cases. I can also be seen ha he maximum delay achieved by he source ransmiing he
Wireless New 27 13:335 349 347 Fig. 4 Maximum delay comparison when one session is sending more raffic han i specified in is reques misbehavior 2.5 2 Maximum Delay sec 5.43 1.5 1 WFQ WFQ-M.5 Voice_1 Voice_2 Voice_3 Voice_4 VideoH263_5 VideoH263_6 VideoJP_7 WWW_8 WWW_9 WWW_1.7 Maximum Delay sec 11.77.6.5.4.3 DM-SCFQ DM-SCFQ-M.2.1 Jurassic Park movie source 7 is smaller for WFQ han i is for DM-SCFQ, while he maximum delay achieved by all he oher delay-sensiive applicaions is larger for WFQ han i is for DM-SCFQ. This is due o he fac ha he more opporuniies a session has o ransmi he more updaed he hybrid coordinaor remains regarding he condiion of is buffer. Since source 7 is he one ha ransmis wih he highes daa rae, i has a chance o inform more ofen he hybrid coordinaor of is newly arrived packes, which means ha i does no have o wai for he cyclic poll as ofen as he oher saions, hus keeping is delays smaller. Figure 3 shows a comparison of he maximum delay achieved in Experimen II, in which here is an exra acive session in addiion o hose acceped by he CAC algorihm. I is expeced ha he maximum allowed delay will be exceeded because he sysem is sauraed. The goal of hese simulaions, however, is o show how sensiive he scheduling mechanisms are o he exra load. I can be seen ha he added delay for DM-SCFQ is proporional o he maximum delay ha a session can olerae, while i is quie unpredicable for Voice_1 Voice_2 Voice_3 Voice_4 ideoh263_5 ideoh263_6 VideoJP_7 WWW_8 WWW_9 WWW_1 WWW_11 WWW_12 WFQ. This again is a consequence of he random delay inroduced in WFQ beween he generaion of a new burs of packes and he ime when he hybrid coordinaor is informed of is exisence. Figure 4 shows he resuls obained in Experimen III. I is clear ha WFQ is very sensiive o his ype of behavior due o he fac ha no only is he misbehaving source generaing exra raffic, bu i is also aking advanage of each ransmission o inform he hybrid coordinaor more ofen of is new packes. DM-SCFQ, on he oher hand, is very robus agains his ype of behavior and only he misbehaving saion WWW 1 ges an excessive delay. 6. Conclusions This work proposes and analyzes specific admission conrol and service differeniaion mechanisms ha can be used o enhance he IEEE 82.11 MAC proocol in order o effecively provide QoS guaranees. Our proposal includes a
348 Wireless New 27 13:335 349 novel raffic scheduling mechanism ha akes ino accoun he limiaions ypical of a wireless environmen, as well as a CAC algorihm based on recen resuls relaed o igh delay bounds for sysems wih leaky-bucke consrained inpu raffic serviced using a fair scheduling policy. The simulaion sudy indicaes ha our proposed mechanisms have accepable performance. In comparison, our proposed schemes have much beer performance in erms of fairness and robusness han he well-known WFQ scheduling algorihm. References 1. P. Barford and M.E. Crovella, Generaing represenaive web workloads for nework and server performance evaluaion, ACM SIGMETRICS, Madison WI, pp. 151 16 July 1998. 2. P. Bara, F. Némeh, R. Szabó and J. Bíró, Call admission conrol in generalized processor sharing schedulers wih igh deerminisic delay bounds, Compuer Communicaions, Vol. 26, No. 2 Feb. 23 pp. 65 78. 3. S. Blake, D. Black and M. Carlson, Inerne Engineering Task Force, RFC 2475, An Archiecure for Differeniaed Services December 1998. 4. R. Braden, D. Clark and S. Shenker, Inerne Engineering Task Force, RFC 1633, Inegraed Services in he Inerne Archiecure: An Overview June 1994. 5. M.S. Gas, 82.11 Wireless Neworks: The Definiive Guide, Sebasopol, CA: O Reilly, 22. 6. S.J. Golesani, A self-clocked fair queuing scheme for broadband applicaions, Proc. 1994 IEEE INFOCOM, Torono, Canada, pp. 636 646, 12 16 June 1994. 7. D.J. Goodman and S.X. Wei, Efficiency of Packe Reservaion Muliple Access, IEEE Transacions on Vehicular Technology, Vol. 4, No. 2 Feb. 1991 pp. 17 176. 8. A. Grilo and M. Nunes, Performance evaluaion of IEEE 82.11e, Proc. 13h Inernaional Symposium on Personal, Indoor and Mobile Radio Communicaions PIMRC 22, Lisbon, Porugal Sepember 15 18, 22. 9. IEEE, Wireless LAN Medium Access Conrol MAC and Physical Layer PHY Specificaions, ANSI/IEEE Sandard 82.11 1999 Ediion. 1. IEEE, Wireless medium access conrol MAC and physical layer PHY specificaions: Medium access conrol MAC enhancemens for qualiy of service QoS, Draf Supplemen o ANSI/IEEE Sd 82.11, 1999 Ediion, Sd 82.11e/D3.1 July 22. 11. M. Kazanzidis, M. Gerla and S.-J. Lee, Permissible hroughpu nework feedback for adapive mulimedia in AODV MANETs, Proc. 21 IEEE Inernaional Conference on Communicaions ICC 21, Helsinki, Finland, Vol. 5, No. 11 14, pp. 1352 1356, 11 14 June 21. 12. S. Mangold, S. Choi, P. May, O. Klein, G. Hierz and L. Sibor, IEEE 82.11e wireless LAN for qualiy of service, invied paper, European Wireless 22 EW 22, Florence, Ialy February 22. 13. P. Medina, Inroducion of qualiy of service mechanisms in he medium access proocol of IEEE 82.11 wireless local area neworks. M. Sc. Thesis, CICESE Research Cener, 24. 14. D. Nandia, J. Kuri and H.S. Jamadagni, Opimal Call Admission Conrol in Generalized Processor Sharing GPS Schedulers, IEEE INFOCOM 21, Anchorage, AK, April 22 26, 21. Also in Technical Repor TR--2, Cener for Elecronic Design and Technology, Indian Insiue of Science. 15. A. Panagakis., N. Dukkipai, I. Savrakakis and J. Kuri, Opimal call admission conrol on a single link wih a GPS scheduler, IEEE/ACM Transacions on Neworking, Vol. 12, No. 5, pp. 865 878, Oc. 24. 16. A.K. Parekh and R.G. Gallager, A generalized processor sharing approach o flow conrol in inegraed services neworks: he singlenode case, IEEE/ACM Transacions on Neworking, Vol. 1, No. 3 1993, pp. 344 357. 17. A.K. Parekh and R.G. Gallager, A generalized processor sharing approach o flow conrol in inegraed services neworks: he muliple-node case, IEEE/ACM Transacions on Neworking, Vol. 2, No. 2 1994, pp. 137 15. 18. R.S. Ranasinghe, L.L.H. Andrew and D. Everi, Disribued Conenion-Free Traffic Scheduling in IEEE 82.11 Mulimedia Neworks, 1h IEEE Workshop on Local and Meropolian Area Neworks. Seleced Papers, D. Skellern, A. Guha and F. Neri, Ediors. Piscaaway, NJ: IEEE, pp. 18 28 21. 19. O. Rose, Saisical properies of MPEG video raffic and heir impac on raffic modeling in ATM sysems, Proc. 2h Annual Conference on Local Compuer Neworks, Minneapolis, MN, pp 397 46 1995. 2. R. Szabó, P. Bara, F. Némeh and J. Bíró, Wors-case deerminisic delay bounds for arbirary weighed generalized processor sharing schedulers, Proc. Inl. Conference on Broadband Communicaions, High Performance Neworking, and Performance of Communicaion Neworks, IFIP-TC6/European Commission NETWORKING 2, Paris, France May 14 19, 2. 21. D.E. Wrege, E.W. Knighly, H. Zhang and J. Liebeherr, Deerminisic Delay Bounds for VBR Video in Packe-Swiching Neworks: Fundamenal Limis and Pracical Trade-Offs, IEEE/ACM Transacions on Neworking, Vol. 4, No. 4 1996, pp. 352 362. José R. Gallardo received he B.Sc. degree in Physics and Mahemaics from he Naional Polyechnic Insiue in Mexico Ciy, he M.Sc. degree in Elecrical Engineering from CICESE Research and Graduae Educaion Cener in Ensenada, Mexico, and he D.Sc. degree in Elecrical Engineering from he George Washingon Universiy, Washingon, DC. From 1997 o 2 he worked as a Research Associae a he Advanced Communicaions Engineering Cenre of he Universiy of Wesern Onario, London, Onario, Canada. From May o December 2, he worked as a Posdocoral Fellow a he Broadband Wireless and Inerneworking Research Laboraory of he Universiy of Oawa. Since December 2, Dr. Gallardo has been wih he Elecronics and Telecommunicaions Deparmen of CICESE Research Cener, where he is a full professor. His main areas of ineres are raffic modeling, raffic conrol, as well as simulaion and performance evaluaion of broadband communicaions neworks, wih recen emphasis on wireless local area neworks WLANs and wireless sensor neworks WSNs. Paúl Medina received he B.Eng. degree from he Sonora Insiue of Technology, Obregon, Mexico, and he M.Sc. degree from CICESE Research and Graduae Educaion Cener, Ensenada, Mexico, boh in Elecrical Engineering. From July o Sepember 25, he worked as a Research Associae a he Broadband Wireless and Inerneworking Research Laboraory of he Universiy of Oawa, Canada. Mr. Medina is currenly wih CENI 2 T, Ensenada, Mexico, working as a lead engineer in proecs relaed o rouing
Wireless New 27 13:335 349 349 and access conrol in wireless sensor neworks, as well as IP elephony over wireless LANs. Weihua Zhuang received he B.Eng. and M.Eng. degrees from Dalian Mariime Universiy, Liaoning, China, and he Ph.D. degree from he Universiy of New Brunswick, Canada, all in elecrical engineering. Since Ocober 1993, she has been wih he Deparmen of Elecrical and Compuer Engineering, Universiy of Waerloo, ON, Canada, where she is a full professor. She is a co-auhor of he exbook Wireless Communicaions and Neworking Prenice Hall, 23. Dr. Zhuang received he Ousanding Performance Award in 25 from he Universiy of Waerloo, and he Premier s Research Excellence Award in 21 from he Onario Governmen. She is an Edior/Associae Edior of IEEE Transacions on Wireless Communicaions, IEEE Transacions on Vehicular Technology, EURASIP Journal on Wireless Communicaions and Neworking, and Inernaional Journal of Sensor Neworks. Her curren research ineress include mulimedia wireless communicaions, wireless neworks, and radio posiioning.