1. INTRUDUCTION. 2. ACCESS METHODS IN IEEE b WLAN

Transcription

1 Esiaion of he Effecive Throughpu in 80.b WLAN Neworks Josip Lörincz, Dinko Begusic Universiy of Spli, FESB udera Boskovica b.b., 000 Spli, Croaia Phone: +385 ( , E-ail: Absrac: eliable odel of he aainable hroughpu in IEEE 80.b WLAN is needed o faciliae capaciy planning and nework deployen. In his paper an approxiae analyical forula for calculaing effecive hroughpu of 80.b devices, as a funcion of a nuber of copeing saions for differen sizes of ransied packes has been proposed. Using his forula WLAN nework adisraor can esiae he hroughpu capaciy of single access poin (AP shared beween uliple wireless saions wih sufficien accuracy. The proposed ehod has been evaluaed by experienal resuls.. INTUDUCTION In recen years, IEEE 80.b Wireless Local Area Neworks (WLAN has eerged as a prevailing echnology for he high bandwidh wireless access in liied geographical area for he obile/porable devices. While allowing for ransission speeds of up o Mb/s in he.4 GHz ISM band, i has becoe widely ipleened WLAN echnology by any anufacurers and cusoers. The hroughpu effecively perceived by obile hoss in WLAN becoes increasingly iporan as any new eerging applicaions such as: obile inforaion access, real-ie uliedia counicaions, neworked gaes, cooperaive work, videoconference ipose high bandwidh requireens. An approxiae odel of he aainable hroughpus in such WLANs is needed o faciliae capaciy planning and nework deployen [] [3] [] []. This paper presens analyical and experienal, characerizaion of he insananeous hroughpu of a saion in an IEEE 80.b WLAN as a funcion of he nuber of copeing saions sharing an access poin (AP. The ain conribuion of his paper is he proposal of approxiae forula for calculaing effecive hroughpu of 80.b devices, as a funcion of a nuber of copeing saions for differen sizes of ransied packes. Using his forula WLAN nework adisraor can esiae he hroughpu capaciy of single AP shared beween any wireless saions wih sufficien accuracy. In his paper we validae he analyical expression wih easureen resuls. The 80.b WLAN nework physical layer suppors uliple ransission raes. Physical layer rae o be used for ransission of paricular frae is solely deered by ransiing saion (STA. Perforance of he WLAN nework will be affeced by signal srengh and degradaion in signal qualiy due o such facors as STA obiliy, ie - varying inerference, and locaion-dependen errors. In order o axiize he syse hroughpu Adapive ae Selecion (AS ay be invoked. Curren 80.b producs degrade he bi rae in seps fro he noal Mb/s o 5.5Mb/s, Mb/s or Mb/s when repeaed unsuccessful frae ransissions or degradaion in received signal srengh are deeced [][5]. We sudy he perforance of an 80.b WLAN BSS (Basic Service Se in he infrasrucure ode, i.e., when a nuber of saions are associaed wih a single AP a noal bi raes of Mb/s or 5.5Mb/s. This paper is organized as follows: Secion. briefly explains he DCF ehod of IEEE 80.b MAC proocol wih basic assess echanis. In Secion 3., we propose analyical forula for approxiae calculaion of effecive hroughpu experienced by hos. The nework opology used for experiens describes Secion 4. Secion 5 validaes analyical and easureen resuls. The ipac of differen sizes of ransied files on overall hroughpu is analyzed in Secion 6.. ACCESS METHODS IN IEEE 80.b WLAN The IEEE 80.b sandard defines a fair access o he shared wireless ediu hrough wo differen access echaniss: a andaory conenion-based access proocol, called he Disribuion Coordinaion Funcion (DCF and an opional polling-based access proocol, called he Poin Coordinaion Funcion (PCF providing unconesed access via arbiraion by a Poin Coordinaor, which resides in he AP. The DCF ehod provides a bes effor ype of service whereas he PCF guaranees a ie-bounded service []. Alhough boh ehods ay coexis, he PCF ehod ay coexis only in parallel wih he DCF ehod and can no exis alone. PCF ehod would be especially well suied for real-ie raffic as i peris o allocae he radio channel according o applicaions requireens, bu he PCF ehod is very rarely ipleened in curren 80. producs. In his paper, we focus on he hroughpu analysis for an IEEE 80.b WLAN based on he DCF proocol which is widespread in he arke oday... Transiing echniques in DCF access proocol There are wo echniques used for packe ransiing in DCF. The defaul one is a wo-way handshaking echanis, also known as basic access echanis. A posiive MAC acknowledgeen (ACK is ransied by he desinaion saion o confir he successful packe ransission. The oher opional one is a four-way handshaking echanis,

2 DIFS Cnenion Window PLCP preable 8 byes MAC header 30 byes PLCP header 6 byes FAME PAYLOAD (0-3 MPDU FCS 4 byes PLCP preable 8 byes PLCP header 6 byes ACK 4byes LLC MAC Sublayer SOUCE DIFS DESTINATION SLOT (0µs con [0,CW] SLOT OTHE MOBILE STATIONS pr 9 µs PLPDU PLPDU v r SIFS ACK SIFS 0 µs pr 9 µs ACK NAV T(N ack PIFS DIFS DIFS 50µs PHYSICAL LAYE con Figure. IEEE 80.b DCF basic access echanis which uses reques-o-send/clear-o-send (TS/CTS echnique o reserve he channel before daa ransission. This echnique has been inroduced o reduce he perforance degradaion due o hidden erals. However, he drawback of using he TS/CTS echanis is increased overhead for shor daa fraes [3]... Perforance of DCF basic access ehod The frae fora of an IEEE 80.b frae is shown in Figure. When a frae or MAC Service Daa Uni (MSDU arrives a he MAC layer fro he higher layer, i is encapsulaed in a MAC Proocol Daa Uni (MPDU by adding 30 byes of he MAC header, and 4 byes of Frae Conrol Sequence (FCS field execuing Cyclic edundancy Check (CC principle. The MPDU is hen passed o he PHY layer, which will aach 8 byes long Physical Layer Convergence Proocol (PLCP preable used for synchronizaion of he receiver and 6 byes long PLCP header. The DCF access echanis is a disribued ediu access proocol based on Collision Sense Muliple Access wih Collision Avoidance (CSMA/CA. CSMA/CD is no used because a saion is unable o lisen o he channel for collisions while ransiing. Basically, he DCF works as follows: before a saion sars a frae ransission, i shall sense he wireless ediu o deere if i is busy. If he saion deecs ha he wireless ediu has been idle during ore han a ie inerval called Disribued Iner Frae Space (DIFS, he saion can ransi he frae iediaely (Figure.. If he ediu is sensed as busy, he saion wais unil he channel becoes idle, hen defers for an exra DIFS inerval. If he ediu reains idle, he MAC sars he backoff procedure by selecing a rando backoff coun. While he ediu says idle, he backoff couner is being decreened every slo ie, and when he couner reaches zero, he frae is ransied. To selec he rando backoff coun, each saion ainains a conenion window (CW value. The backoff coun is deered as a rando ineger drawn fro a unifor disribuion over he inerval [0; CW- ][][][3]. Prioriy access o he wireless ediu is conrolled by use of Iner Frae Space (IFS inervals, i.e., ie inervals beween he ransissions of consecuive fraes. The 80.b sandard defines four differen IFS inervals: Shor IFS (SIFS, PCF IFS (PIFS, DCF IFS (DIFS, and Exended IFS (EIFS [7]. A basic ediu access ehod is shown in Figure. Upon having received a packe correcly, he desinaion saion wais for a SIFS inerval iediaely following he recepion of he daa frae and ransis a posiive ACK back o he source saion, indicaing ha he daa packe has been received correcly (Fig.. In case he source saion does no receive an ACK, he daa frae is assued o be los and he source saion schedules he reransission wih he CW for back-off ie doubled. When he daa frae is ransied, all he oher saions hearing he daa frae adjus heir Nework Allocaion Vecor (NAV, which is used for virual CS a he MAC layer, based on he duraion field value in he daa frae received correcly, which includes he SIFS and he ACK frae ransission ie following he daa frae [3]. 3. THE THOUGHPUT ANALYSIS In his secion, our objecive is o calculae he axiu hroughpu achievable wih only one 80.b AP, aking ino consideraion he nuber of copeing saions (N and size of generaed MSDU packes (. All saions are placed in he F coverage area of sae AP cell wih equal peak heoreical raes of 5.5Mb/s or Mb/s, hus eliaing perforance anoaly described in [4]. We assued siuaion in wich uliple hoss working in he DCF ode wih disabled TS/CTS funcion aeps o ransi a he sae ie, and hos ha chooses salles backoff inerval firs sars ransiing packe wihou collision (Fig... The signal propagaion delay is very sall (around 0.µs in our siulaion and can be negleced [][4]. The overall ransission ie ay be calculaed as follows: T(n DIFS + CONT (n + pr + r + SIFS + ACK (

3 To siplify our analysis furherore, we ake he following assupions: ( here are no uliple successive collisions on he wireless ediu; ( he nework consiss of a finie nuber of conening saions n and every saion always has L-bye long daa frae (or MSDU available for ransission a an infinie rae; (3 collision probabiliy p(n of a ransied frae is consan and independen of he nuber of reransissions ha his frae has experienced in he pas []; (4 fraes are received wih error only when hey encouner collisions due o oher siulaneous ransissions and he effec of frae errors due o bi errors inroduced by channel noise will be ignored []. Wih hese and previous assupions, a ransission cycle is coposed of he following phases ha are repeaed over ie: ( IFS deferral phase coposed of DIFS and SIFS phases; Backoff/Conenion phase; (3 Daa (or PLPDU ransission phase and (4 ACK ransission phase. The overall ransission ie of he single packe can now be expressed as follows: T(n IFS + CONT (n + PLPDU + ACK ( where IFS ie is given by: IFS SIFS + DIFS (3 The IFS ie is consan which represens he su of DIFS ie ( DIFS 50µs and SIFS ie ( SIFS 0µs (boh defined in IEEE 80.b sandard. The overall frae ransission ie experienced by a single hos when copeing wih N- oher hoss will be influenced by inerval con ha accouns for he ie spen in conenion procedures. Since here are uliple hoss aeping o ransi, he channel will be sensed busy and hoss ener a collision avoidance phase: a hos execues he exponenial backoff algorih - i wais for a rando ie inerval disribued uniforly beween [0;CW-] T SLOT called backoff coun ie or conenion ie ( CONT before ransission of he packe sars. The conenion window (CW value varies beween CW 3 and CW ax 04 and he value of slo ie (T SLOT is 0µs (hese paraeers are for 80.b [4] [7] [] [3]. The CW size is iniially assigned as CW, and i is exponenially increased up o CW ax when a ransission fails. Once CW reaches he value of CW ax (afer 6 reransission aeps, reains a he value of CW ax (04 unil i is rese. The CW is rese o CW afer a successful ransission or afer reaching he axiu rery lii []. A sauraion, a ransiing saion will always have a queue of packes o send, so every ransission is preceded wih backoff algorih. Since he backoff coun is uniforly disribued over 0,,, CW - for he firs aep, he backoff ier is (CW -/ T SLOT, on average. Each ransission has probabiliy p(n of collisions, and a saion reransis a packe unil i receives an acknowledgen, so we can odel he nuber of reransissions per packe as geoerically disribued wih probabiliy of success -p(n. Furherore, afer any unsuccessful ransission aep, anoher calculaion of backoff coun is perfored wih a new CW value deered as follows: CW + p CW [ ( ] ( [ ( ] ( CW p n + p n p n ( n [ p( n ] ( CW + p + CW ( n ( +... backoff (4 where (06 is nuber of consecuive reransission aeps before CW reaches is axiu value (04 []. In he case of a single hos ha ries o ransi a susained raffic (he hos has always a packe o send, he carrier sense applies also o he hos s own ransissions, so ha i insers a rando inerval beween each ransission. This is andaory, because ransiing fraes coninuously would preven any oher hos fro accessing he channel [4]. In ha case he analyical forula for con as a funcion of copeing hoss is difficul o derive and we propose o use a siple approxiaion by no aking ino accoun uliple successive collisions. If we neglec uliple successive collisions he hroughpu can be prediced wihin error of he order of 3% [4] [4]. Considering ha he hoss always sense a busy channel when hey aep o ransi and ha he nuber of ransissions ha are subjec o uliple successive collisions is negligible (for 0 or in equaion (4 we will ake in he accoun only he firs wo ers in equaion (4.: CW [ ( ] ( [ ( ] ( CW CW p n + p n p n (5 backoff where p(n is he probabiliy of collisions experienced for each packe successfully acknowledged a he MAC level (0p(n<. A siple expression for p(n can be derived by considering ha a hos aeping o ransi a frae will evenually experience a collision if he value of he chosen backoff inerval corresponds o he residual backoff inerval of a leas one oher hos [4] [3]. Such an approxiaion holds if uliple successive collisions are negligible. So we have: p(n ( /CW n- (6 We will assue ha average conenion ie increases wih every new conening saion (n wishing o ransi. So he expression for average conenion ie ( CONT as produc of conenion window value, slo ie, consan K and nuber of copeing hoss (n is as follows: ( n CW T K n (7 CONT backoff SLOT where K or K corresponds o ransission speeds of Mb/s and 5.5 Mb/s respecively. Consan K is inroduced because we assue ha average ie spen in conenion for he sae nuber of saions is wo ies increased (K since heoreical peak hroughpu is wo ies decreased

4 (5.5Mb/s. Finally, average ie CONT (n spen in conenion for each saion, expressed as a funcion of nuber of copeing saions is given by: n ( n 3 3 CW CONT ( n 3 TSLOT K 3 3 n (8 Transission ie of PLPDU is coposed of PLCP preable and header ransission ie ( pr and MPDU frae ransission ie ( r []: ( 34+ L 8 PLPDU pr + r pr + (9 where L corresponds o he size of he ransied packe and corresponds o he axiu heoreical bandwidh (Mb/s or 5.5 Mb/s. pr varies according o he bi rae used by he hos. Since he PLCP header is alos always ransied a Mb/s, pr equals o 9 µs. An ACK frae is ransied a he rae equal o daa frae rae, and is 4 byes long, which is usually uch shorer han he daa frae. Transission ie of ACK frae ( ack hen equals o 0 µs for ransission daa rae of Mb/s, or 0 µs for ransission daa rae of 5.5 Mb/s. The overall MAC acknowledgen ransission ie is coposed of PLCP preable and header ransission ie ( pr and ransission ie of ACK frae ( ack []: ACK 4 8 pr + ack pr + (0 Finally, o copue he effecive hroughpu H we need o divide he overall ransission ie of ha packe wih he nuber of bis in he ransied packe []: 8 L H ( T ( n According o he las equaion, we propose wo analyical forulas for calculaing insananeous hroughpu of a saion in an 80.b WLAN as a funcion of packe size and nuber of copeing saions sharing an AP. For axiu heoreical bandwidh of Mb/s and 5.5 Mb/s insananeous hroughpus is given by nex wo analyical equaions: H H 5.5 ( n, ( n, K + K + 3 ( 34+ ( L n 8 L n ( n ( n K n K n ( (3 Figure. Nework opology used for easureens where consans K are equal o: K 454µs, K 30µs and K 3 464µs. Values of consans K are coposed of slo ie, CW values and iner frae space paraeers defined in IEEE 80.b sandard. 4. MEASUEMENT SETUP The nework opology used for experiens is shown in Figure. We sudy he perforance of an 80.b WLAN BSS (Basic Service Se working in he infrasrucure ode. Up o wireless saions were associaed o he sae AP on channel. The AP is conneced o a 00 Mb/s Eherne access ( swich. There were oher saions conneced o he pors of he swich also. All wireless saions are he raffic sources excep one wireless saion which receives all he daa sen. We used lapops for our experiens wih siilar processing speed ranging fro Peniu-3. GHz hrough Peniu-4.8 GHz wih AM eory ranging fro 8Mb o 5Mb. Nework perforance of he wireless saion is deered ore by he wireless card ipleenaion hen is processing capaciy []. The AP used was Sybol Technologies Specru4 series 4 access poin [0]. The cards used for sending raffic were Sybol Technologies Specru4 PCMCIA wireless LAN adaper. We disabled WEP on he cards and he AP in order o avoid any poenial overhead. Since we assued ha proble of hidden saion can be negleced, TS/CTS usage was also disabled on AP, seing TS hreshold o a axiu value (347. Our choice for perforance esing sofware was Chario fro NeIQ Corporaion [8]. Chario is an applicaion ha allows for several ypes of end-o-end perforance ess (response ie, hroughpu, jier, delay and los daa for sreag applicaions using up o 4 differen proocols (TCP, UDP, SPX or IPX. By exacly eulaing ransacion raffic fro real applicaions, Chario ess and roubleshoos every segen of WLAN nework and provides seup insrucions called scrips o endpoin saion. These endpoins execue he ess insrucions and reurn he resuls o he console o be viewed by adisraor. On all 0 sending obile saions prograed o send as fas as posible,

5 Throughpu (Mb/s 7,00 6,00 5,00 4,00 3,00,00 L500 Analyical L500 Measured L000 Analyical L000 Measured L500 Analyical L500 Measured Throughpu (Mb/s 4,00 3,50 3,00,50,00,50,00 L500 Analyical L500 Measured L000 Analyical L000 Measured L500 Analyical L500 Measured,00 0,50 0, NUMBE OF STATIONS (n 0, NUMBE OF HOSTS (N Figure 3.Coparaion of he hroughpu experienced by an 80.b hos when all hoss ransi a Mb/s Chario console sofware wih he sae scrips has been insalled. In hese seup insruncions or scrips we defined equal raffic paraears for hree differen sizes (L 500, 000, 500 byes of generaed TCP packes. Chario endpoin sofware and Ehereal sofware used o capure and onior received packes, were insalled on wirelass saion receiving raffic genraed fro all oher saions [9]. 5. ANALYSIS OF THOUGHPUT ESULTS Coparison of analyically calculaed and easureen - based resuls of hroughpu experienced by a 80.b hos when all oher hoss ransi a Mb/s or 5.5 Mb/s are shown a Figures 3. and 4. respecively. Graphs on Figure 3. and 4. show insananeous hroughpu of 80.b hos as a funcion of nuber of copeing hoss (N for hree differen sizes ( of ransied packes. We can observe ha he easured values correspond fairly well o he analyical expressions. The consequences of neglecing propagaion ie, uliple successive collisions and frae errors due o bi errors inroduced by channel noise are observed in sall discrepancy beween analyically obained and easureenbased resuls. Analyzing graphs on Figure 3. and 4. we noice ha he hroughpu obained by he 80.b WLAN is uch saller han he noal bi rae of Mb/s or 5.5 Mb/s. For exaple, if here are no collisions, one hos ay expec he axiu hroughpu of approxiaely half he value of he noal bi rae. Furherore, he proporion of he useful hroughpu srongly depends on he nuber of copeing hoss. The insananeous hroughpu decreases as he nuber of hoss rying o ransi increases. We also observe ha he overall hroughpu of one hos (for he sae nuber of hoss wishing o ransi siulaneously increases when he size of ransied packe increases. Wih he larger size of he packe, he larger aoun of inforaion is ransied a he sae ie resuling in he increased overall hroughpu. On he oher hand, he probabiliy of a packe geing corruped Figure 4. Coparaion of he hroughpu experienced by an 80.b hos when all hoss ransi a 5.5 Mb/s increases wih he packe size, due o he higher Bi Error ae of a radio link. In he case of packe corrupion, he saller he packe, he less overhead o ransi, hus aking saller packes suiable for real-ie applicaions and longer packes suiable for file ransfer and Inerne raffic. 6. FTP THOUGHPUT MEASUEMENT Using he sae nework opology shown in Figure., in our second experien we easure effecive hroughpu of 80.b hos as a funcion of nuber of copeing hoss (N while ransiing hree differen file sizes. We varied he file size (Mb, 5Mb, 50Mb, 50Mb and download wih every new easureen o see he effec on he overall hroughpu. In each of four easureens, he files of he sae size are ransferred using FTP proocol fro all obile saions o he saion acing as FTP server. The hroughpu has also been easured wih NeIQ Chario console and endpoin sofware applicaion which was insalled on pair of wireless saions. Wih hese easureens we waned o analyze he hroughpu behavior of 80.b devices in realisic circusances. Such as, he use of he FTP proocol for file ransission as a siple everyday office service. The noal bi rae of all our hroughpu experiens was 5.5 Mb/s, wih axiu speed of packe ransission. Figure 5 shows he insananeous hroughpu as a funcion of ie for firs six hoss saring o ransi file of he sae size (5Mb, one by one during he ie, wih rando ie period beween beginnings of succeeding ransission. As expeced, we can observe ha insananeous hroughpu decreases wih every new wireless saion saring o ransi. All easureen resuls shown in Figure 7., indicae alos equal overall hroughpu for he sae size of ransied files. We can conclude ha overall hroughpu does no depend on he size of ransiing files. I depends on nuber of ransiing saions and oens a which hey sar o ransi. In las easureen we ransferred files of

6 n n n3 n4 n5 n6 Throughpu Mb/s 3,5,5 MB 5MB 00MB 5MB & download 0,5 Figure 5. Insananeous hroughpu of 80.b hos as a funcion of ie during ransission of 5Mb file NUMBE OF HOSTS (N Figure 7. Measured hroughpu of 80.b hos for differen sizes of ransied files of he 80.b WLAN is uch saller han he noal bi rae. Furherore, he proporion of he useful hroughpu srongly depends on he nuber of copeing hoss. The insananeous hroughpu decreases as he nuber of hoss wishing o ransi increases. We also observe ha overall hroughpu of one hos (for he sae nuber of hoss wishing o ransi siulaneously increases when size of ransied packe increases. Overall hroughpu does no depend on he size of ransiing files. I depends on nuber of ransiing saions and oens a which hey sar o ransi. Figure 6. Insananeous hroughpu of 80.b hos during ransission of 5Mb file when all saions sar ransission siulaneously 5Mb in parallel wih downloading file of 5Mb fro disan server over he Inerne. We can observe considerable hroughpu degradaion caused by wo siulaneous ransissions perfored by every hos. We also easured insananeous hroughpu of 80.b hos when all 0 saions sar file ransission process a he soe ie, downloading idenical file of he 5Mb fro paricular wireless saion acing as FTP server (Figure 6.. A he beginning of FTP ransission, we can observe rapid hroughpu decrease, since all 0 saions placed in he sae AP signal coverage area (wih equal peak heoreical hroughpu of 5.5 Mb/s siulaneously sar download of he 5Mb file. When all saions begin FTP ransissions, average insananeous hroughpu was.5 Mb/s which is alos equal o he resul obained by using equaion (3. 7. CONCLUSION In his paper we have analyzed he hroughpu perforance of he IEEE 80.b WLAN neworks. The easured values correspond fairly well o he analyical expressions we derived for calculaing effecive hroughpus as a funcion of nuber of copeing saions for differen sizes of ransied packes. This analysis shows ha he hroughpu EFEENCES [] Javier del Predo Pavon AND Sunghyun Choi, "Link Adapaion Sraegy for IEEE 80. WLAN VIA eceived Signal Srengh Measureen", hp://wnl.snu.ac.kr/~schoi/publicaion/conferences/03- ICC-LA.pdf [] Arunchandar Vasan and A. Udaya Shankar, "An epirical Characerizaion of Insananeous Throughpu in 80. WLANs", hp:// [3] Haiao Wu, Yong Peng, Keping Long, Shiduan Cheng, Jian Ma, " Perforance of eliable Transpor Proocol over IEEE 80. wireless LAN Analysis and Enhanceen", in Proceedings of IEEE INFOCOM 00. [4] Marin Heusse, Frank osseau, Gilles Berger-Sabbael, Andrzej Duda, "Perforance Anoaly of 80.b", in Proceedings of IEEE INFOCOM 003., San Francisco, USA, [5] Theo Pagzis, Peer Kirsein and Seve Hailes, "Operaional and Fairness Issues wih Connecion-less Traffic over IEEE 80.b", hp: // [6] Pablo Brenner, " A Technical Tuorial on he IEEE 80.b Proocol", Breezeco 997. [7] ANSI/IEEE, " 80.: Wireless LAN Mediu Access Conrol (MAC and Physical Layer (PHY specificaion", 000. [8] hp:// [9] hp:// [0] hp:// [] Y.C. TAY and K.C. CHUA, " A Capaciy Analysis for he IEEE 80. MAC Proocol", Wireless Neworks 7, 59 7, 00, [] P. Chaziisios, V. Visas and A. C. Boucouvalas, " Throughpu and delay analysis of IEEE 80. proocol", hp://dec.bourneouh.ac.uk/saff/boucouvalas/iwna.pdf [3] Giuseppe Bianchi, " Perforance Analysis of he IEEE 80. Disribued Coordinaion Funcion, IEEE journal on seleced areas in counicaions", vol. 8, no. 3, arch [4] Marin Heusse, Frank osseau, Gilles Berger-Sabbael, Andrzej Duda, "Allocaion for Diffserv Based Qualiy of Service over 80.b" in Proceedings of IEEE INFOCOM 003., San Francisco, USA, 003.