Performance Evaluation of Table Driven and Buffer Adaptive WLANs

Transcription

1 6th WSEAS Internatinal Cnference n CIRCUITS, SYSTEMS, ELECTRONICS,CONTROL & SIGNAL PROCESSING, Cair, Egypt, Dec 9-3, 7 57 Perfrmance Evaluatin f Table Driven and Buffer Adaptive WLANs IMAM MAHMUD TAIFUR RAHMAN AL-WAZEDI, AHMED K. ELHAKEEM Department f Electrical and Cmputer Engineering Cncrdia University Sir Gerge Williams Campus. 55 St. Catherine W., Mntreal, Quebec, Canada H3G W CANADA Abstract: - The randm access cntrl (MAC) technique f standard WLANs is called the distributed crdinatin functin (DCF) [3]. DCF is a carrier sense multiple access based n cllisin avidance (CSMA/CA) scheme with binary sltted expnential backff. This expnential backff makes the system mre cmplex and fairness [3]-[3] amng the statins is a majr cncern. This paper shws ne pssible evlutin f WLANs where expnential backff is nt emplyed. In the new techniques herein users transmit randmly but adapt themselves t traffic cnditins, thus imprving thrughput and delay while guarantying fairness. Users in the first technique are cntrlled by a table which is derived frm traffic measurements (Table Driven). In the secnd, users transmissin activities are functin f their buffer cntents. Key-Wrds: - MAC, Table driven WLANs, Buffer adaptive WLANs Intrductin Wireless lcal area netwrks (WLANs) have been widely deplyed fr the past decade. Their perfrmance has been the subject f intensive research. In [] imprvement f thrughput and fairness is shwn by ptimizing the backff. [] uses a measure called the average idle interval which des nt cnsider the number f cllisins. In [], the authrs prpsed a MAC layer based WLAN technique in which they gave higher pririty t access the channels s as t imprve the thrughput and the channel utilizatin. [], [] discuss the fairness prblem f the expnential backff. [3] Prpses a technique based n cllisin avidance and fairness t imprve the channel utilizatin. Few WLAN standards have been adpted e.g. IEEE 8. [4] which uses cllisin avidance scheme with binary sltted backff. [4] Expresses hw the thrughput deterirates with increasing the number f ndes. [5] Uses an analytic mdel t study the channel capacity i.e., maximum thrughput when using the basic access (tw-way handshaking) methd. [6] Cnsiders three kinds f CSMA/CA prtcls, which include Basic, Stp-and-Wait and 4-Way Handshake CSMA/CA, and intrduce a theretical analysis fr them. Cali in [7] pinted ut that depending n the netwrk cnfiguratin, DCF may deliver a much lwer thrughput cmpared t the theretical limit. In [8] a cntentin based MAC prtcl named fast cllisin reslutin is presented. [] Prpses a mdel named DCF+ which shws the fairness perfrmance. [] Presents the perfrmance evaluatin f the decentralized nature f cmmunicatin between ndes in IEEE 8., in presence f hidden ndes. [3] Shws the perfrmance evaluatin f Multihp Ad Hc WLANs. Thus extensive research has been cnducted n WLANs []-[3]. Fairness index was nly discussed in [], [], [7] and []. This paper tries t investigate simultaneusly the fur perfrmance indexes i.e. thrughput, delay, delay variance and fairness which are nt cnsidered in previus studies []-[3]. In the table driven technique we cnsider bth idle perids and number f cllisins (Table driven technique) which shws the actual lad n the netwrk. In the secnd, we emply buffer adaptive technique which guaranties fairness and prvide smaller delay variance.. The IEEE 8. MAC Prtcl PACKET TRANSMISSIO N WIRELESS CHANNEL ACK COLLISI ON IDLE BACKOFF Fig. IEEE 8. MAC mechanism SLOT PACKET TRANSMISSIO N Fig. shws ne f many transmissin scenaris pssible with the IEEE 8. DCF mde. In this mde a nde with a packet t transmit initializes a backff timer with a randm value selected unifrmly frm the range [, CW-], where CW is

2 6th WSEAS Internatinal Cnference n CIRCUITS, SYSTEMS, ELECTRONICS,CONTROL & SIGNAL PROCESSING, Cair, Egypt, Dec 9-3, 7 58 the cntentin windw in terms f time slts. After a nde senses that the channel is idle fr an interval called (DCF interframe space), it begins t decrease the backff timer by ne fr each idle time slt bserved n the channel. When the channel becmes busy due t ther ndes transmissin ativities the nde freezes its backff timer until the channel is sensed idle fr anther. When the backff timer reaches zer, the nde begins t transmit. If the transmissin is successful, the receiver sends back an acknwledgement (ACK) after an interval called the. Then, the transmitter resets its CW t CW min. In case f cllisins the transmitter fails t receive the ACK frm its intended receiver within the specified perid, it dubles its CW subject t maximum value CW max, chses a new backff timer, and starts the abve prcesses again. Surce Destinatin Other RTS CTS DATA NAV (DATA) NAV (RTS) NAV (CTS) Defer Access ACK Fig. RTS/CTS access mechanism in DCF In 8., DCF als prvides a mre efficient way f transmitting data frames that invlve transmissin f special shrt RTS and CTS frames prir t the transmissin f actual data frame. As shwn in fig., an RTS frame is transmitted by a nde, which needs t transmit a packet. When the destinatin receives the RTS frame, it will transmit a CTS frame after interval immediately fllwing the receptin f the RTS frame. The surce statin is allwed t transmit its packet nly if it receives the CTS crrectly. Nte that all the ther statins are capable f updating their knwledge abut ther ndes transmissin duratin by receiving a certain field in RTS, CTS, ACK, and packets transmissin called netwrk vectr allcatin (NAV). This helps t cmbat the hidden terminal prblems. In fact, a nde that is able t receive the CTS frames crrectly, can avid cllisins even when it is unable t sense the data transmissins directly frm the surce statin. If a cllisin ccurs with tw r mre RTS frames, much less bandwidth is wasted when CW Backff Starts cmpared with the situatins where larger data frames in cllisin, thus justifying the case fr RTS, CTS mde f peratin. Accrding t the prtcl new users typically get the access befre existing users wh may cllide with each ther leading t fairness prblems [3] [3]. 3. System Analysis Fr The Ideal Standard Case Withut Backff Let p be the transmissin prbability f each nde and M be the number f active statins. Assuming n backff and each user tries t transmit randmly in each slt fllwing the perid, nly ne user tries his RTS which is then fllwed by CTS and a successful packet, the prbability f successful transmissin, is thus given by the fllwing M P s = Mp( p) () The prbability f an idle slt is M P = ( p).() and prbability f unsuccessful transmissin RTS (cllisin)is P = P P (3) c s Let i be the number f idle perids (cycles) t success shwn in fig 3 and j be the number f idle slts in each idle perid lengths ( W, W,... ). S the efficiency ( η ) is given by equatin (7). It is easily seen that the average length f each idle perid except the last ne befre packet success is given by ( P ) = W = W =... WI = E( j) j Pc j= P Pc W = W =... WI = slts. (4) ( P ) The last idle perid has an average f P Ps WI = slts.... (5) ( P ) The average number f cycles is given by, I I = ip i i= ( P ) =.(6) P s All cycles leading t n success (RTS heard but n CTS) will each have a cst f W i +T RTS +T +T Slt +T secnds. j

3 6th WSEAS Internatinal Cnference n CIRCUITS, SYSTEMS, ELECTRONICS,CONTROL & SIGNAL PROCESSING, Cair, Egypt, Dec 9-3, 7 59 First Idle Perid W =3 Last Idle perid befre success W I Packet transmissin ACK RTS Cllisin... RTS CTS Packet transmissin... Idle Slts Idle Slts Previus Transmissin Perid Current Transmissin Perid (Frm start t end f packet success) Fig 3. Transmissin Activity n the Wireless Channel Next Transmissin Perid η = T Paylad ( W + W +.. WI ) TSlt + ( I ) { TRTS + T + T } + T + TRTS + TCTS + TACK + 3T + TPaylad + WITSlt (7) The number f cllisins is C = I. This C and W I are calculated frm different values f M and p and stred in tw tables (nt shwn fr space cnsideratin). S fr particular values f M and p there is a particular value f C and W I. Frm equatin (7) the efficiency η can be calculated fr different values f M and p as in fig. 4. Table depicts the prbabilities at which the maximum efficiency ccurs fr different values f M. Fig 4. Efficiencies fr different prbabilities and different number f statins Table. Optimum Efficiencies fr different prbabilities and different number f statins N f Statins Prbability Optimum Efficiency N f Statins Optimum Efficiency Prbability Table Driven WLANs In this new prtcl, if the ndes sense that the channel is idle fr an interval called (DCF interframe space), they try t send RTS f a packet with a prbability p which is dependent n the traffic cnditin i.e. the number and activities f the ndes as fllws. The users cntinuusly mnitr the channel in each idle slt fllwing the, idle perid. If the previus slt is idle, it calls a unifrm randm generatr (,). If the value f this generatr is less than r equal t p, it tries t start its RTS transmissin in the given next slt. If the value is larger than p, the users persist n listening and repeats RTS transmissin trials as stated. Hwever if the channel is sensed busy the user defers his transmissin till the next idle perid heard. The ndes measure the number f cllisins C = I and the length W I f the last idle perid sliding, (by mnitring the channel ver a large enugh windw) they can then use the tables frmulated in sectin 3 t btain the crrespnding p and M. Users having a nn-empty queue start by mnitring the channel fr the first n transmissin perids. This active user will average the length f the idle perid preceding the crrect packet transmissin ver n transmissin perids i.e. W I and C = I i.e. the average number f cllisin ver the same perid.

4 6th WSEAS Internatinal Cnference n CIRCUITS, SYSTEMS, ELECTRONICS,CONTROL & SIGNAL PROCESSING, Cair, Egypt, Dec 9-3, 7 5 Aided with these values the users btain the perating values f p and M and uses p t cntrl their activities fr the head f line packet in their queue. Active users cntinuusly mnitr the channel and use a sliding windw technique t estimate W I and I and hence btain M, p. Fr example the first sliding windw averages W I and C f the first n transmissin perids. The secnd windw averages W I and C f the l =,3,... n + transmissin perids. The third t (n+) transmissin perids. The sliding windw averaging prcess reflects the changing traffic, s transmissin activity f active users are dependent nly n the current traffic and nt n past histry. It is pssible that the tables releting ( M, p) t ( W I, I ) yield mre than ne pssibility fr M, p fr certain W I, I measurement values frm the sliding windw. In this case, the user average the btained values f M and use table t find the ptimum p at this averaged value f M. This table is btained frm Fig. 4 in an evident manner. The peratin f this table driven technique is similar t the DCF standard IEEE prtcl [4] except fr using this ptimized transmissin prbability p and discarding the timers and backff windws. The active users just estimate M, p frm the traffic cnditins (by sensing the channel) in a sliding windw fashin transmissin, ne perid after anther. We nte that ld and fresh users bth measure the traffic and bth adpts t same traffic cnditin fairly and btain same p. Hwever having same p des nt mean all users will repeatedly cllide in same slt. Since feeding a randm number generatr with p yields different slt number t start transmitting the RTS each time it is called. 5. Buffer Adaptive WLANs In the buffer adaptive technique, each nde s prbability f transmissin s trying is calculated based n the number f packets stred in the buffers. Bu The prbability f trying, p f each nde is, Bu max Where Bu is the number f packets stred in the buffers at each nde and Bumax is the user buffer capacity. The buffer adaptive technique is simple, it is similar t the IEEE 8. RTS/CTS, DCF mde except fr the eliminatin f timers and expnential backff. It is als similar t the table driven technique except fr the eliminatin f table cnstructin simply the users adjust their randm transmissin prbability p fullwing the standard perid, cntinuusly based n the queue size. This technique des neither need traffic measurement nr table establishments. 6. Simulatin Results Fr numerical calculatins the fllwing parameters are used: T Paylad =msec;phyheader=8bits;ack=bits +PHY header; RTS=6bits+PHY header; CTS=bits+PHY header; Channel bit rate= Mbits/s; Slt time (T Slt )= 5 µs; T =8 µs; T =8 µs. In the table driven technique, as per the standards, fllwing the bservance f each, users try t transmit with prbability p btained frm the abve table which is btained frm the traffic measurements. S all users with nn empty buffer try t transmit a packet with a prbability btained frm the table. If tw r mre statins try t transmit at the same time, cllisins ccur. If n statins transmit (Fig 3), the number f idle slts will increase. If ne statin is successful after certain number f idle and cllisin perid, the transmissin perid ends. As a result the ttal time fr ne successful packet transmissin include T, T, T RTS, T CTS, T Idle, T Paylad. The efficiency is calculated at the end f the simulatin at certain values f M, λ, p i.e., η = T Paylad N f Transmissin Perids ( n) Time in the whle simulatin ( n) Where Time is the ttal simulatin time which depends n T, T, T RTS, T CTS, T Slt, T Paylad. ( n) Initially Time = T and subsequently increased based n the user s activity, e.g. ( n) ( n) Time = Time + TSlt ; ( n) ( n) Time = Time + TRTS + T ( n) ( n) Time = Time + TRTS + T fr each successful packet fr each idle Slt CTS + T + T perid ; fr each cllisin + T T ; 3 + Paylad In the case f the buffer adaptive technique, the simulatin time is calculated in the same way as per user s activity abve. Hwever, n table is cnstructed and sliding windws dn t apply. In the table driven technique described in sectin 4 the active statins estimate the value f M. At certain traffic the curve shwn in fig. 5 is linear, that means the ffered and the estimated values f the number f active statins are the same.

5 6th WSEAS Internatinal Cnference n CIRCUITS, SYSTEMS, ELECTRONICS,CONTROL & SIGNAL PROCESSING, Cair, Egypt, Dec 9-3, Offered lad λ= Estimated N f statins 3.5 Efficiency Windw Size=5 Windw Size= Windw Size= Offered N f statins The table driven technique can be cnsidered as a lad adaptive system. That means it has the capability t adapt t the input traffic as quickly as pssible. Figure 6 shws a case where the input traffic suddenly increase frm 5 packets/sec t packets/sec. In this case the thrughput( η Input traffic rate(λ) ) (Fig. 6) is shwn t fllw the ffered traffic λ. Fig. 7 shws the efficiency curve fr different ffered lads fr the table driven technique fr different windw sizes. This shws that the efficiency rises and becmes saturated at higher values f the lad. The windw size has small effects n the efficiencies at different lads. Thrughput and Traffic ( λ ) Fig 5. Estimated M at a certain Traffic Input Traffic λ Thrughput N f Transmissin Perids Fig 6. Thrughput and Input Traffic crrespnding t the number f Transmissin perids. Fig. 8 depicts the packet delay crrespnding t different lads fr the table driven technique fr different windw sizes. The windw sizes have little effect n the packet delay crrespnding t different lads Offered Traffic λ Fig 7. Efficiency crrespnding t different ffered traffic using different windw size Delay (sec) Windw Size=5 Windw Size= Windw Size= Offered Traffic λ Fig 8. Delay crrespnding t different ffered traffic using different windw size Efficiency Bu/3 Bu/ Bu/ Offered Traffic λ Fig 9. Efficiency at different lads fr Buffer adaptive system fr different capacities

6 6th WSEAS Internatinal Cnference n CIRCUITS, SYSTEMS, ELECTRONICS,CONTROL & SIGNAL PROCESSING, Cair, Egypt, Dec 9-3, 7 5 As the efficiency rises with the increased lad, excess numbers f packets are left in the buffers. This results in a large packet delay at higher lads. Fig. 9 and fig shws the efficiency and the delay perfrmance at different lads fr the buffer adaptive technique. Frm packets/sec t 4 packets/s, the efficiency f the buffer adaptive technique and the table driven technique are mre r less the same. Beynd 4 packets/s, in the buffer adaptive technique the efficiency becmes very small, because all statins transmit their packets with higher prbability which results in high amunt f cllisin and n success. In these figures efficiency and the delay are calculated fr different buffer capacities, such as, 3,, 3. Fig. shws that the delay perfrmance degrades as the capacity f the buffer increases. Hwever the efficiency increases as the buffer capacity increases (fig. 9). Fairness is anther imprtant issue. T express this, we take the fairness index defined in [] t measure the fair packet capacity allcatin. That is, FI n ( x ) i= i ( xi ) = n n i=, where FI is the fairness index, n is the number f statins, x i is the packets transmitted by the i th active statin during the simulatin time (current traffic in which the ffered traffic λ is same fr all statins). Fig. shws that the fairness index decreases as the windw size is decreased fr the case f table driven technique. Fig. shws the fairness index f the buffer adaptive technique. Frm this we can bserve that the statins can be fairly perated when the buffer capacity is made high. 6 5 Bu/3 Bu/ Bu/3. Delay (sec) 4 3 Fairness Index Bu/8 Bu/3 Bu/ Offered Traffic λ Fig. Delay curves at different lads fr Buffer adaptive system fr different capacities N f Statins Fig. Fairness Index f the Buffer adaptive technique fr different buffer capacities..998 Fairness Index Windw size=5 Windw size= Windw size= N f Statins Fig. Fairness Index fr the table driven technique fr different windw sizes Fig 3. Fairness Index fr different number f statins fr different cases Fig. 3 shws a cmparisn f the Fairness index between the buffer adaptive technique and the

7 6th WSEAS Internatinal Cnference n CIRCUITS, SYSTEMS, ELECTRONICS,CONTROL & SIGNAL PROCESSING, Cair, Egypt, Dec 9-3, 7 53 prtcl prpsed in []. We cnclude that the buffer adaptive technique yields the same fairness index as in [] which use mdified expnential backff. Fig. 4 shws a cmparisn between table driven technique and buffer adaptive technique alng with the prtcl prpsed in []. It can be bserved that, fr all the cases up t active statins the perfrmance is the same. Beynd that lad, the fairness f the table driven technique degrades. Fig. 5 shws the delay variance fr the buffer adaptive technique fr different buffer capacities. It is bserved that the delay variance increases in accrdance with the number f statins as well as the buffer capacity. Fig. 6 shws the delay variance f the table driven technique fr different windw sizes. The delay variance perfrmance is wrse than the case f buffer adaptive technique Windw size=5 Windw size= Windw size=5 Delay Variance Fig 4. Fairness Index fr different number f statins fr different cases Nw let us intrduce the Delay variance fr the tw different new techniques. Delay variance is n ( Di Daverage) calculated by i=, where DV is the delay variance, th i statin and all statins. Delay Variance DV = n D is the average packet delay f the i D is the average packet delay f average Bu/8 Bu/3 Bu/ Number f Statins Fig 5. Delay Variance fr the Buffer adaptive technique fr different buffer capacities N f Statins Fig 6. Delay Variance fr the Table driven Technique fr a different windw sizes 7. Cnclusin In this paper tw new techniques (Table driven and Buffer adaptive) were presented, mdeled and cmpared. Simulatin results shw that the table driven technique perfrms well fr faster lad variatins, where as the buffer adaptive technique des nt. But the buffer adaptive technique has FI (fairness index) and DV (Delay Variance) perfrmances better than the table driven technique making it suitable fr real time applicatin (vice, vide etc). The buffer adaptive technique des nt scale well at higher lads as cmpared t the table driven technique. The thrughput and delay perfrmances are better in the case f table driven technique making it suitable fr data applicatin. References: [] Xuejun TIAN, Xiang CHEN, Tetsu IDEGUCHI and Yuguang FANG, Imprving Thrughput and Fairness in WLANs thrugh Dynamically Optimizing Backff, IEICE Transactins n Cmmunicatins 5 E88-B(): [] Liang Zhang, Yantai Shu and Oliver Yang, Perfrmance Imprvement fr 8. Based Wireless lcal Area netwrks IEICE Trans. Cmmun.,vl. E9-B, NO.4 April 7.

8 6th WSEAS Internatinal Cnference n CIRCUITS, SYSTEMS, ELECTRONICS,CONTROL & SIGNAL PROCESSING, Cair, Egypt, Dec 9-3, 7 54 [3] V. Bharghvan, "Perfrmance evaluatin f algrithms fr wireless medium access," IEEE Internatinal Cmputer Perfrmance and Dependability Sympsium IPDS'98, pp.86 95, 998. [4] G. Bianchi, "Perfrmance analysis f the IEEE 8. distributed crdinatin functin," IEEE J. Sel. Areas Cmmun., vl.8, n.3, pp , March. [5] Y.C. Tay and K.C. Chua, "A capacity analysis fr the IEEE 8. MAC prtcl," ACM/Baltzer Wireless Netwrks, vl.7, n., pp.59 7, March. [6] J.H. Kim and J.K. Lee, "Perfrmance f carrier sense multiple access with cllisin avidance prtcls in wireless LANs," Wirel. Pers. Cmmun., vl., n., pp.6 83, 999. [7] F. Cali, M. Cnti, and E. Gregri, "Dynamic tuning f the IEEE 8. prtcl t achieve a theretical thrughput limit," IEEE/ACM Trans. Netw., vl.8, n.6, pp , Dec.. [8] Y. Kwn, Y. Fang, and H. Latchman, "A nvel MAC prtcl with fast cllisin reslutin fr wireless LANs," IEEE INFOCOM'3, April 3. [9] J. Weinmiller, H. Wesner. J.P. Ebert, and A. Wlisz, "Analyzing and tuning the distributed crdinatin functin in the IEEE 8. DFWMAC draft standard," Prc. MASCOT, San Jse, CA, Feb [] H. Wu, Y. Peng, K. Lng, S. Cheng, and J. Ma, "Perfrmance f reliable transprt prtcl ver IEEE 8. wireless LAN: Analysis and enhancement," IEEE INFOCOM', vl., pp , June. [] H.S. Chhaya and S. Gupta, "Perfrmance mdeling f asynchrnus data transfer methds f IEEE 8. MAC prtcl," ACM/Baltzer Wireless Netwrks, vl.3, pp.7 34, 997. [] G. Bianchi and I. Tinnirell, "Kalman filter estimatin f the number f cmpeting terminals in an IEEE 8. netwrk," IEEE INFOCOM'3, vl., pp , April 3. [3] Farshad Eshaghi, Ahmed Elhakeem, Yusef R. Shayan, Perfrmance Evaluatin f Multihp Ad Hc WLANs, IEEE Cmmunicatins Magazine, March 5.