Performance Improvement for Based Wireless Local Area Networks

Size: px

Start display at page:

Download "Performance Improvement for Based Wireless Local Area Networks"

Donna Hancock
6 years ago
Views:

1 Performance Improvement for 82. Based Wreless Local Area Networs Lang Zhang and Yanta Shu Department of Computer Scence Tanjn Unversty Tanjn 372, Chna Olver W.W. Yang School of Informaton Technology and Engneerng Unversty of Ottawa Ottawa, Ontaro, Canada, KN 6N5 Astract - In typcal nstallaton of an 82. ased wreless local area networ (WLAN), mole hosts would access the networ through Access Ponts, even when two mole statons n the same WLAN communcate. That s, all the pacets n a WLAN are reured to forward through the AP. Snce the AP has the same prorty as the other mole statons to access the channel, accordng to the Medum Access Control Layer (MAC) protocol, the AP usually ecomes a ottlenec n WLANs and the networ performance degrades sgnfcantly. In ths paper, we propose a new MAC layer protocol for WLANs n order to mprove the throughput performance. Theoretcal analyss and smulaton results show that our new protocol wors much etter n WLAN than the standard DCF. Keywords - Wreless Local Area Networ, Access Pont, DCF I. INTRODUCTION In recent years, wreless networs ased on the IEEE 82. standard are ecomng ncreasngly prevalent. IEEE 82. Wreless Local Area Networ (WLAN) [] n partcular has ganed a domnant share n the maret among varous emergng technologes for roadand wreless access. The focus s now turnng to deployng these networs over hot spots such as arports, hotels, cafes, and other areas from whch people can have pulc access to the Internet. The IEEE 82. standard [] provdes a detaled MAC (Medum Access Control) and PHY (PHYscal) layer specfcaton for WLANs. The IEEE 82. MAC provdes an access to the shared wreless medum through two dfferent access mechansms: a mandatory random access protocol called DCF (the Dstruted Coordnaton Functon), and an optonal pollng-ased protocol called the Pont Coordnaton Functon (PCF). DCF s the fundamental mechansm to access the shared medum. It s a random access method ased on the carrer sense multple access wth collson avodance (CSMA/CA) wth a nary slotted exponental acoff mechansm. We focus the DCF n ths paper. Wreless networs can e used ether to replace wred networs, or as an extenson of the wred networ nfrastructure. There are two asc wreless networ topologes: nfrastructure WLANs and ad hoc networ. In ad hoc networ, mole statons communcate drectly, f they can hear each other. Ths type of networ s often formed on a temporary ass, and s commonly referred to as an Independent Basc Servce Set (IBSS). On the contrary, the asc uldng loc of an IEEE 82. WLAN deployment s a BSS, whch s composed of an access pont (AP) and multple statons assocated wth the AP. A Basc Servce Set (BSS) conssts of two or more wreless nodes, or statons that have recognzed and estalshed communcatons wth each other. The man functon of an AP s to form a rdge etween wreless and wred LANs. The AP s analogous to a ase staton used n cellular phone networs. Statons do not communcate drectly ecause all communcatons etween statons, or etween a staton and a wred networ node, have to go through the AP. The AP s not mole, and form part of the wred networ nfrastructure. In recent years, there are many nterestng nvestgatons on the performance evaluaton of IEEE 82. and smlar protocols [2-6]. Most of these studes are largely confned to Ad hoc networ. In ths paper we shall focus on WLANs where every pacet needs the AP to forward even when the sender and recever are all mole statons n the same WLAN and close enough to hear wth each other. Snce the AP just has the same prorty wth other mole statons to access the channel, t ecomes a ottlenec and degrades the networ throughput performance sgnfcantly. Furthermore, compared wth a wred networ, the lmted andwdth resource n a WLAN s already a g prolem, and the AP forwardng smply maes t worse. In ths paper, we propose an extenson to the standard DCF protocol n WLANs y gvng the AP a hgher prorty to access the channel. We expect our new protocol can mprove the throughput performance and the channel utlzaton greatly, and we shall demonstrate ths y our performance evaluaton va theoretcal analyss and smulaton. The rest of ths paper s organzed as follows. Secton II presents the overvew of the DCF of IEEE 82. MAC protocols. Secton III descres our new protocol n detal. Secton IV presents the performance evaluaton. The smulaton results and a concluson are presented n Sectons V and VI respectvely. For the remander of the paper, the followng symols and notatons are as follow: n: The numer of current transmttng statons. CW mn, CW max : The mnmum and maxmum contenton wndow sze as defned n IEEE 82. standard. W : The contenton wndow sze of th acoff stage, t euals to 2 *W, where W = ( CW mn +). m: The maxmum acoff stage.

2 m': The maxmum count of the exponental ncrease of contenton wndow from CW mn + to CW max +. s(t): The stochastc process representng the acoff stage of the staton at tme t. τ : The proalty that a staton transmts n a randomly chosen slot tme. : The proalty that other statons transmt the data. P tr : The proalty that there s at least one transmsson n the consdered slot tme. P s : The proalty that a transmsson s successful, gven the proalty P tr. T s, T c : The average tme the channel s sensed usy ecause of a successful transmsson or a collson respectvely. S: System throughput. G: Goodput of WLANs. σ: The duraton of an empty slot tme. δ : The propagaton delay. II. DISTRIBUTED COORDINATION FUNCTION We shall summarze the 82. DCF to ntroduce concepts and termnologes. More detals can e found from the 82. standard []. DCF s the fundamental access method of the IEEE 82. MAC protocol, whch s ased on carrer sense multple access wth collson avodance (CSMA/CA). The DCF provdes asynchronous access for est-effort data transmsson. For a staton to transmt, t shall nvoe the carrer sense mechansm to determne the usy/dle state of the medum. If the medum s usy, the staton shall defer untl the medum s determned to e dle wthout nterrupton for a perod of tme eual to DCF Inter-Frame Space (DIFS) when the last frame detected on the medum was receved correctly. After ths DIFS medum dle tme, the staton shall then generate a random acoff perod for an addtonal deferral tme efore transmttng, unless the acoff tmer already contans a nonzero value, n whch case the selecton of a random numer s not needed and not performed. Ths process mnmzes collsons durng contenton etween multple statons that have een deferrng to the same event. The random acoff nterval s unformly chosen n the range (, w-). The value (w-) s called Contenton Wndow (CW), whch s an nteger wth the range CW mn and CW max determned y the physcal layer characterstcs. The acoff tme counter s decremented whle the medum s sensed dle, frozen when a transmsson s detected on the channel, and reactvated when the channel s sensed dle agan for more than a DIFS. The staton transmts when the acoff tme reaches zero. After each unsuccessful transmsson, w s douled, up to a maxmum value 2 m' W = (CW max +), where W euals to (CW mn +). When the destnaton receves a data frame correctly, t wll transmt an ACK to ndcate the successful recepton after the Short Inter-Frame Space (SIFS) nterval. If the source staton does not receve the ACK wthn a specfed ACK_Tmeout, the data s assumed to e lost and the source staton schedules the retransmsson. III. DCF WITH AP HIGH PRIORITY In WLANs, the man functon of an AP s to form a rdge etween wreless and wred LANs. Statons do not communcate drectly ecause all communcatons etween statons, or etween a staton and a wred networ node, have to go through the AP. Therefore, every pacet needs the AP to forward n WLANs even when the sender and recever are all mole statons n the same WLAN and close enough to hear wth each other. However, snce the AP just has the same prorty wth other mole statons to access the channel, t has now ecome a ottlenec and would degrade the networ throughput performance sgnfcantly. For example, we assume there are n pars of mole statons n the WLAN. Each par conssts of one mole recever (MR) and one mole sender (MS). That s to say, there are (n+) actve statons (n MS plus the AP) n the WLAN where an actve staton s a staton that has data to send. Assumng all the actve statons are n saturaton mode,.e., they always have data to send, the AP can acheve at most /(n+) of the channel rate due to ts same prorty as other statons. Also each MR can otan on the average /n of the AP s throughput and /n(n+) of the channel rate. So the andwdth acheved y every MR s n nverse proporton to n 2 approxmately. That s why the throughput performance wll e degraded serously, when the numer of actve statons ncreases. In our opnon, as every pacet needs the AP to forward n WLANs, the AP should tae at least half of the total andwdth. In order to acheve ths, we propose a new MAC layer protocol for 82. ased WLANs, whch can gve the AP hgher prorty to access the channel so as to mprove the throughput performance and the channel utlzaton. In our new protocol, we can mae a smple change to the decreasng rule of the acoff tme counter n order to gve the AP a hgher prorty. We now that when a mole staton gets the channel, t must e transmttng pacet to the AP frst. So when the AP receves a data pacet correctly, t wll transmt an ACK to ndcate the successful recepton after a SIFS nterval. Both the mole statons and the AP then tae the followng actons. Acton for mole statons After a DIFS medum dle tme, the mole staton shall generate a random acoff perod for an addtonal deferral tme efore transmttng. The acoff tme counter s decremented whle the medum s sensed dle, and frozen when a transmsson s detected on the channel. If the source s the AP, t should reactvate the acoff tme counter when the channel s sensed dle agan for more than a DIFS. If the source s one of the mole statons, t wll ncrease the acoff tme counter y one (except when the acoff tme counter s already at ts maxmum) and then reactvate the decreasng process. The staton transmts when the acoff tme reaches zero.

3 Acton for the AP After a DIFS medum dle tme, the AP also generates a random acoff perod for an addtonal deferral tme efore transmttng. The acoff tme counter s decremented whle the medum s sensed dle. If the AP s acoff counter reaches zero, t should send the data. After a successful transmsson, the AP would egn the next contenton process as already defned n the standard. If others otan the channel durng the AP s acoff tme, the AP should e the recever, snce t must e nvolved n any communcaton. When the transmsson fnshed, the AP wll set ts acoff tme counter e zero and reactvate the acoff tme counter when the channel s sensed dle agan for more than a DIFS. At ths tme, snce ts acoff tme counter s zero, the AP should otan the channel mmedately and send the data. As all other mole statons has ncreased ther acoff tme counter y one after the communcaton, there should not e any collson occur. As a result, the AP can now send one pacet, after any mole statons sendng a pacet. In ths way, the AP wth hgher prorty can otan at least half of the andwdth n the WLAN and wll not e the ottlenec anymore. IV. PERFORMANCE ANALYSIS In ths secton, we propose a two-dmensonal Marov chan model to evaluate the performance of our scheme. Frst, we wll descre our model, whch we use to get the statonary proalty τ that the staton transmt a pacet. In addton, we assume a fxed numer n of mole statons, and each always havng a pacet avalale for transmsson. A. Marov Chan Model Let (t) e the stochastc process representng the acoff-tme counter for a gven staton at slot tme t. At the egnnng of each slot tme, f the medum s determned to e usy, the acoff tme counter of each staton s frozen; otherwse, the counter s decremented. Let s(t) e the stochastc process representng the acoff stage of the staton at tme t. Also let (t) e the proalty that at least one of the (n ) remanng mole statons would transmt. As one mportant approxmaton, f we now assume s constant and ndependent each other. Then the two-dmensonal process {s(t), (t)} forms a dscrete-tme Marov chan shown n Fg.. In ths Marov chan, let m e the maxmum acoff stage (whch also means the maxmum retransmsson count), and let m' e the maxmum count of the exponental ncrease of CW from CW mn + to CW max +. Note that the value of m may e larger than that of m'. As represented n Fg., once CW reaches CW max +, the CW shall reman at the value untl t s reset. Therefore, we have W = 2 W m m W = 2 W > m. () where W=(CW mn +), and 2 m' W=(CW max +). Fgure. Marov chan model of acoff wndow scheme The only non-null one-step transton proaltes of ths Marov chan are P {,, } = [2, W ] [, m] PW {, W, } = [, m] P {,, + } = [, W 2] [, m] P{,,} = ( ) / W [, W ] [, m ] (2) P {,,} = / W [, W ] [, m] P{, m, } = / W [, W ] These transton proaltes account for ) the fact that the acoff tme decrement s stopped and the acoff counter ncreases y one (except when t s already at ts maxmum), when the channel s sensed usy; 2) the fact that at the egnnng of each slot tme the acoff tme s decremented; 3) the fact that a new pacet followng a successful pacet transmsson starts at acoff stage, and thus the acoff s ntally unformly chosen n the range [, W -]; 4) the stuaton when an unsuccessful transmsson occurs at acoff stage -, the acoff stage ncreases, and the new ntal acoff value s unformly chosen n the range [, W -]; 5) the need to reset CW to CW mn at the maxmum acoff stage, and to restart the acoff stage for a new frame after. Let, = lm P{ s( t) =, ( t) = }, [, m], [, W ] e t the statonary dstruton of Marov chan. Frst we note that = < m, (3),, from whch we can show easly = m. (4),, Accordng to the gloal alance euatons and transtons n the chan, we have m ( ) j, + j m, =, W = +, W 2 (5) W P {,, } = Pst {( + ) =,( t+ ) = () st =,() t = }.

4 , m ( ) j, + j = m, = + W ( ) + < < W, +, m ( ) j, + j = m,, = +,2 W So accordng (5), (6) and (7), we have (6) ( ) (7) j j, ( W j)( ), j = = W ( ) W Smlarly, we have ( ) (8), +, < m, = W W = + + ( ) < m,< < W (9),,,, + W, +, + ( ) < m, = W Comnng to (8) and (9), we otan, j j, ( W j)( ) j= = ( m, W ) () W ( ) Fnally, accordng to (4) and (), all the values, can e expressed as functons of the value, and the proalty. therefore,, can e determned y mposng the normalzaton condton for statonary dstruton and can e expressed as functons of the proalty. m W =, () = = The proalty τ that a staton transmts n a randomly chosen slot tme can now e expressed as m m m+ τ =, =, =,. (2) = = In the statonary state, a staton transmts a pacet wth proalty τ. So we have = ( τ ) n. (3) Euatons (), (2), and (3) represent a nonlnear system n the two unnowns τ and, whch can e solved y numercal technues. B. Throughput Analyss Let P tr e the proalty that there s at least one transmsson n the consdered slot tme. And let P s e the proalty that a transmsson s successful, gven the proalty P tr. So we have = ( τ ) n. (4) Ptr tr n n nτ( τ) nτ( τ) Ps = =. (5) n P ( τ ) Now we are ale to express the normalzed system throughput S as the rato. E[Payload Informaton n a slot tme] S = E[Length of a slot tme] PP s tre[ P] (6) = ( Ptr ) σ + PsPtrTs + ( Ps ) PtrTc Here, T s and T c are the average tme the channel s sensed usy ecause of a successful transmsson or a collson respectvely. The E[P] s the average pacet length and σ s the duraton of an empty slot tme. Let H = PHYhdr + MAChdr e the pacet header, and δ e the propagaton delay. Let T c e the average tme the channel s sensed usy y each staton durng a collson. Accordng to the standard [], after transmttng a DATA/RTS frame, the staton shall wat for an ACKTmeout/CTSTmeout nterval. If the response does not occur durng the ACKTmeout/CTSTmeout nterval, the staton shall conclude that the transmsson of the DATA/RTS has faled. Accordng to [3], n the asc access case, we have Ts = DIFS + H + E[ P] + δ + SIFS + ACK + δ, (7) Tc = DIFS + H + E[ P] + δ + SIFS + ACK and n the RTS/CTS access case. We have Ts = DIFS + RTS + δ + SIFS + CTS + δ + SIFS + H + EP [ ] + δ + SIFS+ ACK+ δ, (8) Tc = DIFS + RTS + δ + SIFS + CTS We defned the Goodput (G) of the WLAN n ths paper as the sum of the end-to-end throughput n WLANs. That s, when two mole statons communcate wth each other through the AP, the goodput s the end-to-end transmsson rate etween these two mole statons. When a mole staton communcates wth a wred staton through the AP, the system throughput s the end-to-end transmsson rate etween the mole staton and the AP. As we assume aove, that all mole statons wor n saturaton mode. In our analyss, we assume the UDP protocol s used. As TCP s a self-adaptng protocol, we cannot mae t wor n saturaton mode. However, n Secton V, we shall prove that our new protocol can also mprove the networ performance usng TCP protocol y smulaton. In our scenaro, mole statons n the WLAN communcate wth each other through the AP. We assume there are n (n>)pars of mole statons n the WLAN. Each par contans one MS and one MR. The MS sends UDP data pacets to the MR n a saturaton mode through the AP. In such case, the goodput of the WLAN s just the forwardng rate of the AP. As we dscussed aove, the AP can only acheve on the average /(n+) of the system throughput. So we have S PP s tre[ P] GStandard = =. (9) n+ ( n+ )[( Ptr) σ + PsPT tr s + ( Ps ) PT tr c] As for our new protocol, the AP can send a frame

5 mmedately after any mole staton sendng a frame. It means that as long as a mole staton can otan the wreless channel successfully, a frame wll e forward to the MR n the duraton 2*T s. So we have PP s tre[ P] Gnew =. (2) ( Ptr ) σ + PsPtr (2 Ts ) + ( Ps ) Ptr Tc From the comparson etween (9) and (2), t s easy to conclude that our new protocol wll mprove the throughput performance of the WLAN sgnfcantly. The performance evaluaton n Secton V wll elucdate ths further. Snce TCP s a drectonal protocol, all statons (mole statons and the AP) wll send the data, when mole statons communcate wth each other or wth wred statons usng TCP protocol. In such case, the AP s a ottlenec too. Our protocol can gve the AP more chances to get the channel, so as to mprove the performance of WLANs. The smulaton result can e found n Secton V. V. PERFORMANCE EVALUATION We shall frst valdate our method and our analyss y ns-2 smulaton [8]. Fg. 2 shows the WLAN under consderaton. There s one Server (S) n the wred networs and several Wreless Statons (WS) n the WLAN. The channel rate s set to Mps and the pacet sze s 24 ytes. Fgure 2 Networ smulaton scenaro We frst consder the scenaro where mole statons communcate wth each other usng UDP. We assume there are n pars of mole statons n the WLAN. Each par contans one MS and one MR. As shown n Fg.3, the smulaton results (ponts wth x ) of our new protocol agree very well wth our analytcal results (sold lnes), thus demonstratng the analytcal euaton (2) of goodput for our new protocol ased on Fg. s ute capale of capturng the real networ scenaros. Lewse, we fnd such agreement for the analytcal goodput euaton (9) for the standard DCF [2]. As we can see, the goodput performance of our new protocol s much etter (at least doule) than that of the standard DCF n WLANs. The more mole nodes n the WLAN s, the more ovous the advantage of our protocol. When the numer of mole staton s 3, the goodput of our protocol s aout 6 tmes larger than that of the standard DCF. Ths s ecause the AP s not a ottlenec n our protocol as dscussed n Secton IV. Goodput (Mps) standard sm new sm standard model new modle Numer of Mole Statons Fgure 3. Goodput performance compare for UCP par scenaro Farness Index standard DCF new protocol Numer of Mole Statons Fgure 4. Farness performance compare Farness performance s also a very mportant property n wreless protocol. Fgure 4 presents the farness ndex of oth the standard DCF and our modfed DCF. We shall use the commonly accepted farness ndex defned n [7] to measure the far andwdth allocaton capalty. That s, n 2 n 2 x, where FI s farness ndex, n s the = = F I = x n numer of contendng statons, x s the andwdth shared y the th contendng staton. As shown n Fg. 4, our modfed DCF mantans the same good farness performance as the standard DCF (whle mprovng the goodput performance of the WLAN greatly). The second scenaro s smlar to the frst scenaro except that TCP s used nstead of UDP. Whenever the MS sends data to the MR, the MR should return the ACK. As TCP s a self-adaptng protocol, all statons should e n ther unsaturated mode. If the MS/MR does not receve the ACK/DATA pacet, t wll reman slent and not contend for the channel. Such ehavor gves the AP more chance to otan the channel. Therefore, the goodput performance does not degrade as serously as usng UDP. As shown n Fg.5, our new protocol n such scenaro also can otan etter performance than the standard DCF. In our protocol, the AP can send one pacet wthout contendng the channel, after any mole statons sendng a pacet. Such process saves the contenton tme of the AP and reduces the proalty of contenton.

6 Goodput (Mps) standard DCF new protocol Numer of Mole Statons Fgure 5. Goodput performance compare for TCP par scenaro Goodput (Mps) standard DCF our protocol Numer of Mole Statons Fgure 6. Goodput performance compare for TCP download scenaro Fnally, we consder the downloadng scenaro where mole statons n the WLAN constantly download data from the wred server as shown n Fg. 2 usng the TCP protocol. Ths downloadng scenaro also has ts unsaturated mode. As we now the TCP s a drectonal protocol, mole statons downloadng data from the wred staton should contend the channel to return the ACK pacet, whch also eneft the AP n our new protocol. The analytcal and smulaton results n Fg.6 show the goodput performance of our new protocol s etter than that of standard DCF. VI. CONCLUSIONS In ths paper, we have proposed a new MAC layer protocol for 82. ased WLANs to mprove the networ performance. Compared wth standard DCF, our new protocol can gve the AP more change to access the channel, so that the AP wll not e the ottlenec any more. Our theoretcal analyss proved that our protocol can mprove the networ performance. We have also done extensve smulatons to evaluate the performance of our new protocol n severcal verous scenaros. The results prove the correctness of our analyss. Both smulaton results and theoretcal analyss shows our protocol can enhance the performance of a WLAN largely. ACKNOWLEDGMENTS Ths research was supported n part y the Natonal Natural Scence Foundaton of Chna (NSFC) under grant No , y a grant from the Csco Unversty Research Program Fund at Communty Foundaton Slcon Valley, and y the Natural Scences and Engneerng Research Councl of Canada (NSERC) under Grant No. RGPIN REFERENCES [] IEEE standard for wreless LAN Medum Access Control (MAC) and Physcal Layer (PHY) specfcatons, ISO/IEC 882-:999(E), Aug [2] G. Banch, Performance analyss of the IEEE 82. dstruted coordnaton functon, IEEE Journal on Selected Areas n Communcatons, vol. 8, no. 3, pp , Mar. 2. [3] Hatao Wu, Yong Peng, Kepng Long, Shduan Cheng, and Jan Ma, Performance of relale transport protocol over IEEE 82. wreless LAN: analyss and enhancement, n Proc. IEEE InfoCom 2, New Yor, NY USA, June 23-27, 22. [4] F. Cal, M. Cont, and E. Gregor, Dynamc tunng of the IEEE 82. protocol to acheve a theoretcal throughput lmt, IEEE/ACM Transactons on Networng, 8(6): , Dec. 2. [5] Guanghong Wang, Yanta Shu, Lang Zhang, Delay Analyss of the IEEE 82. DCF, n Proceedngs of IEEE Internatonal Symposum on Personal Indoor and Mole Rado Communcatons. Bejng, Chna, 23. pp [6] L. Bonon, M. Cont, E. Gregor, " Desgn and Performance Evaluaton of an Asymptotcally Optmal Bacoff Algorthm for IEEE 82. Wreless LANs," n Proc. IEEE InfoCom 3, Hyatt Regency San Francsco, Calforna, USA, Aprl - 3, 23. [7] R. Jan, D. Chu, W. Hawe, A uanttatve measure of farness and dscrmnaton for resource allocaton n shared computer systems. Techncal Report TR-3, DEC Research Report, 984. [8] Ns-2 Networ Smulator,

Resource Allocation Optimization for Device-to- Device Communication Underlaying Cellular Networks

Resource Allocation Optimization for Device-to- Device Communication Underlaying Cellular Networks Resource Allocaton Optmzaton for Devce-to- Devce Communcaton Underlayng Cellular Networks Bn Wang, L Chen, Xaohang Chen, Xn Zhang, and Dacheng Yang Wreless Theores and Technologes (WT&T) Bejng Unversty