A novel pause count backoff algorithm for channel access


Published on

Published in: Education, Technology
  • Be the first to comment

  • Be the first to like this

No Downloads
Total views
On SlideShare
From Embeds
Number of Embeds
Embeds 0
No embeds

No notes for slide

A novel pause count backoff algorithm for channel access

  1. 1. International Symposium on Computer Science and its Applications A Novel Pause Count Backoff Algorithm for Channel access in IEEE 802.11 based Wireless LANs Hao-Ming Liang1, Sherali Zeadally2, Naveen K Chilamkurti3, Ce-Kuen Shieh1 joeliang@hpds.ee.ncku.edu.tw;szeadally@udc.edu; n.chilamkurti@latrobe.edu.au 1 Institute of Computer and Communication Engineering, Department of Electrical Engineering, National Cheng Kung University 3 Department of Computer Science and Computer Engineering, La Trobe University Melbourne, Australia-3086 ²Department of Computer Science and Information Technology University of the District of Columbia 4200, Connecticut Avenue, N.W. Washington, DC 20008 Abstract Sense Multiple Access with Collision Avoidance (CSMA/CA) mechanism and contend for channel This paper proposes a novel backoff algorithm to access. This contention behavior results in collisions enhance the Distributed Coordination Function (DCF) among stations and therefore additional time is spent to function in IEEE 802.11 based wireless networks. The recover from collisions. Many research [2-3] focuses proposed algorithm, known as Pause Count Backoff on transmission quality and efficiency for DCF (PCB) algorithm, observes the number of backoff functions because DCF is mandatory supported in counter pauses during the channel access contention IEEE 802.11 devices. and sets the appropriate contention window, based on As mentioned previously, there are unavoidable the estimated results. We evaluate the performance of overheads such as header and collisions within the the proposed PCB algorithm using simulation tests and network. In [4] the authors show that the efficiency of we compare its performance with other proposed IEEE the network decreases when there are more active 802.11 backoff algorithms (Exponential Increase stations within the system. A Markov chain model is Exponential Decrease (EIED), and Adaptive Enhanced used to represent the behavior of a DCF function and a Distributed Coordination Function (AEDCF)). Our throughput formula is proposed. From the analysis in results demonstrate that PCB outperforms other [4], the Binary Exponential Backoff (BEB) algorithm backoff algorithms in various network conditions. is the key factor that influences system efficiency. In Moreover, the fairness index and end-to-end delay of 802.11 standards, the size of the Contention Window PCB are also much better than those obtained with (CW) is doubled when transmission fails and reset to other channel access algorithms. initial value CWmin once transmission is successful. This reset behavior becomes very unstable when numerous stations are contending within wireless 1. Introduction channel. This can cause more collisions and decreases the whole system utilization. IEEE 802.11 [1] is the dominant technology used In prior research efforts [5-6], the authors discuss in Wireless LANs. In IEEE 802.11 standard, a MAC techniques to reduce the collision rate by controlling protocol supports two coordination functions. One of the contention window size in their backoff algorithms them is Point Coordination Function (PCF) which to improve the throughput performance. The basic provides a polling-based service and is only available concept central to these methods is the estimation of in an infrastructure network mode. The Access Point the system load using the transmissions status and the (AP) acts as the administrator and determines channel appropriate CW size is computed. The size of CW access by sending a poll frame to each station in the could be additive or multiplicatively increased because network. The other transmission function defined in a of collisions and additive or multiplicatively decreased MAC protocol is the Distributed Coordination as a result of successful transmissions. However, the Function (DCF) which is a contention-based service estimation method is based on partial observations, and is available both in Ad Hoc and infrastructure such as that each station uses its own status of network modes. In DCF, the stations adopt a Carrier transmissions to represent the whole system. The status978-0-7695-3428-2/08 $25.00 © 2008 IEEE 163DOI 10.1109/CSA.2008.38
  2. 2. of transmissions and system load may have a positive expected before the ACK timeout timer, otherwise thecorrelation but is not sufficient to precisely set CW sender will consider this frame as lost and retransmitsvalue. As a result, such an imprecise calculation of CW it. For each failed transmission, the station doubles itssize for stations introduces a fairness issue. current CW size until CW reaches its maximum value In this work, we propose a new method to estimate (CWmax) (the default value of CWmax is 1024 in IEEEthe number of active stations and find an appropriate 802.11b). The station then performs the backoffvalue of CW within a wireless network. The procedure again and reduces the probability ofcountdown procedure in backoff mode is paused when collision in the next transmission by using a larger CWother stations use the wireless channel at the same (CWnew). After a successful transmission, the CW istime. Therefore, each pause represents more than one reset to the initial value (CWmin), which is equal to 32station using the wireless channel and the number of in IEEE 802.11b.pauses could give a sense of the system status. Theproposed method, known as Pause Count Backoff(PCB), counts pauses during the countdown procedureand sets an appropriate CW size for the currentcondition. Using simulation results, we show that theproposed method improves system utilization andreduces the end-to-end delay when compared withother previous schemes such as the DistributedCoordination Function (DCF), Exponential IncreaseExponential Decrease (EIED), and Adaptive EnhancedDistributed Coordination Function (AEDCF) Figure 1 Illustration of the CSMA/CA mechanismalgorithms. The fairness index of PCB is 2.2 Related Backoff Algorithmsapproximately 1 in most of the network simulationscenarios. The size of the contention window in the backoff This paper is organized as follows. Section 2 procedure can affect the overheads for network accessdescribes the DCF function and discusses the channels with contentions. A large CW results in aoverheads of transmissions in wireless channel. Brief long idle duration when there are only a few activereviews of related studies are presented in section 2. In stations in the system (although a large CW could leadsection 3, we describe our proposed PCB algorithm. to a lower collision rate). A small CW can enhance theSimulation results and a performance analysis of our channel utilization but the number of collisions couldproposed PCB algorithm are discussed in section 4. increase quickly if a small CW is used for many activeFinally, section 5 concludes the paper and presents stations. Thus, the backoff algorithm should adapt thefuture works. correct value of CW to fit the system status. Many past research efforts on enhanced algorithms have been2. Background and Related work proposed taking into account various considerations such as throughput maximization [5] and QoS2.1 IEEE 802.11 DCF Algorithm requirements [6]. In In [5], a scheme named EIED has been proposed, the Based on the 802.11 standard, DCF adopts a station sets the new contention window CWnew asCSMA/CA mechanism. In this algorithm, each station CWold multiplied by the parameter ri, (where ri is theneeds to sense a wireless channel before sending increment backoff factor) as a successful transmission.frames. Transmission is performed if the medium is When a collision occurs, a new value of CW is givenidle for a Distributed InterFrame Space (DIFS) period by CWold divided by the parameter rd where rd is theas shown in Figure 1a – otherwise the backoff decrement backoff factor. The advantage ofprocedure is activated. A backoff number is chosen exponential increase is that Exponential Increase (EI)randomly from the interval [0,CW-1]. The backoff can reduce the probability of continuous collidednumber is decremented by one for every idle timeslot frames when many stations contend a channelduring the countdown procedure. The station sends out concurrently. Moreover, ED (Exponential Decrease)frames when the backoff number is counted down to keeps the collision history of the previouszero as shown in Figure 1b. This frame may collide transmissions instead of resetting automatically towith other frames or be discarded due to wireless error CWmin. ED could prevent numerous collisions fromduring the transmission. To indicate a successful occurring, especially in a network with large number oftransmission, an acknowledgement from the receiver is stations. In [6], the authors propose an adaptive service differentiation algorithm for IEEE 802.11 WLANs. In 164
  3. 3. AEDCF each station calculates the collision rate it pauses during the countdown procedure in step 1 couldexperiences during a given interval. A high collision observe the number of active stations in the system. Inrate often indicates a small CW under heavy system Figure 2, node A observes three pauses during itsloading. In AEDCF, a station does not reset the CW backoff procedure. From the observation, node Aafter a successful transmission. The station sets a new paused resume paused resume pausedCW based on the current collision rate. In AEDCF, the node A backoff window timeauthors seek to determine the optimal size of the CWthat will minimize collisions and improve the system node B timeefficiency. The accuracy of the estimated system status node C timedetermines whether the CW of the proposed backoffalgorithm is appropriate or not. In prior approaches, node D timethese algorithms estimate the system load from the Figure 2 Estimating number of active stations with thenumber of collisions encountered. This is inadequate paused backoff counterbecause the station only observes its own transmission determines that there are more than three mobilestatus which does not represent the whole system stations concurrently contending for the wirelesscondition. The imprecise information affects fairness channel. To keep the scheme stable, Exponentialamong stations. In EIED and AEDCF algorithms, a Weighted Moving Average (EWMA) is applied tostation may inaccurately assign a small CW for a calculate an average pause count. The number ofhighly contended channel. This will result in this pauses observed directly correlates to the backoffstation having a higher channel access opportunity than counter value. A high backoff counter records moreother stations. When the station with a small CW pauses than a small backoff counter during thecollides with others, the station will set the CW to a countdown procedure. We propose the solution bysmall value once again due to the small CW used in the setting CW size to active stations in step 2. In step 2, alast transmission and a lower collision rate. policy with successful transmission results sets the proper contention window size for the backoff3. Pause Count Backoff (PCB) Algorithm procedure. While transmission is a success, CW is set to the average pause value multiplied by β which The main objective of our proposed algorithm is to relates to the collision rate. The method used toimprove the system performance of DCF with fairness compute β is given below:into consideration as well. Previous efforts have 2N (1) Pc ≅focused on improving system efficiency by adjusting CWthe size of CW from transmission results obtained. β = 1 / Pc (2)Although the transmission result correlates with the Pc represents the probability of collision for eachsystem status, it cannot precisely determine the system transmission in a station. N is the number of activestatus from a partial transmitting event. Inaccurate stations in the system. The new CW size can beestimation results in inefficiency and unfairness among derived from the number of active stations and β (asstations. Thus, our proposed PCB algorithm takes shown above). To minimize oscillations in the pauseglobal observations to estimate the system status, and count, the new CW is applied after an observationthen PCB sets an appropriate CW that matches the period which is determined by the number ofglobal system status. There are two steps in our transmission attempts. In this paper, we set theproposed PCB algorithm. The first step is the observation period to 10 transmission attempts in theestimation of CW and the second step involves setting station. PCB sets a proper contention window sizeCW. according to the system status when transmission is In step 1 of our proposed PCB approach, a method successful instead of resetting CW to CWminestimates the number of active stations. In DCF, the immediately or additively/ multiplicatively decreasingbackoff counter pauses if other stations transmit frames CW by a single transmission status. When transmissionat the same time and resumes countdown whilst a fails, the new CW should prevent further collisions inchannel is idle for a DIFS period. The concept is the next transmission to reduce the overhead ofillustrated in Figure 2. The proposed method observes retransmissions. In our proposed PCB algorithm, thethe number of pauses until the counter becomes zero. station sets a new CW to a large contention window toEach pause represents another station transmitting its avoid further collisions. The station could avoid theframes or more than two stations incurring a collision. next collision effectively by choosing a large backoffThe backoff counter is uniformly distributed and the counter, although such a large CW may result in moreparameter avg_paused_count for average number of idle timeslots during the backoff procedure. The idea 165
  4. 4. here is that the overheads of collided frames areusually larger than waiting for idle timeslots. The Control rate 1 Mbits/sparameter rd (where rd is a number for division) used in Data rate 11 Mbits/sthe pseudocode for the PCB algorithm below Slot_time 20 μsdetermines the size of the new CW. The new CWcomputed using rd and CWmax addresses the fairness SIFS 10 μsissue among stations in step 1 when a station does not DIFS 50 μscalculate the system status correctly. The new CW CWmin 32could reset the inappropriate CW and adjust CW again CWmax 1024after an observation period. Packet size 1000 bytes Pseudocode of PCB algorithmStep 1 : estimation • Goodput: avg _ pause _ count = (1 − α ) * avg _ pause _ count + α * current _ pause _ count Goodput is the most common performance metric that calculates total data amount received in a periodStep 2 : setting CW by a station. In general, a higher goodput always when collision occurs set CW new = CW max / rd indicates better efficiency in a system. In this paper, we use aggregate goodput to evaluate PCB and when transmissi on succeeds related backoff algorithms. if during observatio n period CW new = current CW • Fairness Index: else Fairness among stations is an important problem CW new = avg _ pause _ count * β in BEB study and has been discussed by many restart a new observatio n period researchers. Fairness index could show if resource is fairly allocated to each stations. Authors in [8]4. Performance evaluation derived a fairness index using the formula given below: This section evaluates the system performance ofthe proposed PCB algorithm with other existing (∑ G ) i i 2 (3) Fairness Index = n * ∑ (G ) 2backoff algorithms under different system loads. The i isimulation is performed using NS-2 version 2.28 Where n is the number of stations and Gi is thesimulator [7]. We use IEEE 802.11b based WLAN goodput of station i achieved. The value of thesetup and we assume transmissions without the fairness index is bounded to the interval [0, 1]. TheRequest to Send/Clear to Send (RTS/CTS) mechanism index is equal to 1 when all stations obtain the samein an ideal channel. Each mobile station establishes a goodput.Constant Bit Rate (CBR) flow with a 2 Mbps link to • Collision Rate:the base station. The number of stations is varied from Collision rate gives a probability that packets be5 to 50 and the duration of simulation is set to 30 discarded due to collisions in each transmission. Aseconds. The parameters used in the simulation are higher collision rate usually indicates heavy systemlisted in Table 1. The parameters ri and rd in EIED load and implies more overheads.algorithm are set to 2 as suggested in [5]. For AEDCF, • Average end-to-end delay:we set the observation period of the estimated collision End-to-End delay is the time it takes for a packetrate to 0.5 seconds and α to 0.8 used in [6]. In the to travel from sender to receiver. For some time-proposed PCB algorithm, we set the weight α to 0.9 to constraint applications, end-to-end delay is the mostobtain a smooth pause count. The parameter β is set to concern than other metrics. In this paper, we5 which mean the expected collision rate Pc of a station calculate the average end-to-end delay in the systemis around 20% in heavy network load by PCB by various backoff algorithms.algorithm. The parameter rd is set to 4 to get a largesize of new CW that is value 256 in IEEE 802.11b 4.1 Goodput Performanceafter collisions. We use the following performance metrics to Figure 3 shows the goodput performance results ofevaluate the performance of PCB with other previously various backoff algorithms for IEEE 802.11 WLANs.proposed algorithms: The efficiency of standard DCF performs worse (as Table 1 IEEE 802.11 b MAC and network parameters expected) when more stations contend for the channel. used in simulation Although the EIED algorithm takes an exponential 166
  5. 5. decrease CW policy instead of resetting to CWmin whenthere is a successful transmission, the curve decreaseswhen there are more active stations in the system. Thismeans that stations applying EIED and DCFalgorithms make decisions with unclear system statusand adjust the CW quickly from the result of a singletransmission. In contrast to EIED and DCF, thegoodput of AEDCF and PCB algorithms remains highwith respect to various system loads. Theseimprovements mean that stations using AEDCF and Figure 3. Aggregate goodput vs. number of stationsPCB algorithms adjust CW value appropriatelyaccording to the load variation within the network. Incase of a light system load, when there is a collision,PCB wastes some idle timeslots due to the large newvalue of CW. The graph of PCB has maximumgoodput at around 20 stations. When the number ofstations increases, the overheads of collision decreasesthe efficiency in PCB. Overall, the PCB algorithmobtains a high efficiency compared with other backoffalgorithms in various network conditions. AlthoughFigure 3 shows that AEDCF performs consistentlybetter compared to PCB, we demonstrate further below Figure 4. Fairness index vs. number of stationsother additional performance (e.g.,fairness, collisionrate, average end-to-end delay) benefits provided byPCB over AEDCF.4.2 Fairness Index The aggregate goodput in Figure 3 can representthe efficiency of a system. However, when designing abackoff algorithm, we should also consider anotherimportant criterion: fairness among stations. The worstcase scenario is when one station sets a very small CWdue to inaccurate calculations of the system status and Figure 5. Collision rate vs. number of stationsas a result always occupies the channel. The aggregategoodput is high but it is unfair to other stations. In simulation enlarges the goodput of stations achieved atFigure 4, we present the fairness index of each backoff the end of the experiments.algorithm among stations. Using the simulation setupdescribed in prior section, we executed the simulation 4.3 Collision Ratefor 30 iterations and we calculated a 95% confidenceinterval. From Figure 4, the proposed PCB algorithm Figure 5 presents the collision rate of varioushas the most stability and is close a fairness index of backoff algorithms. As expected, a high collision rateone when compared with other contention algorithms. usually indicates additional overheads and longer end-We also observe that the fairness index of AEDCF and to-end delays. From the simulation results, more frameEIED are low and oscillatory. This phenomenon means collisions occurred as the number of active stations insome stations occupy more channel capacity than other the system grows. The collision rate of a standard DCFstations due to a different understanding of the system is the highest among the compared algorithms due tostatus among stations. In the case of the AEDCF the resetting behavior for each successful transmission.algorithm, the estimated collision rate dominates the In the case of n=50, the collision rate is around 50%.new CW size. A slight difference in the estimated The PCB algorithm achieved a low collision rate whencollision rates among stations at the start of the compared with other backoff algorithms (as shown in Figure 5). According to the parameter β setting, the collision rate is around 20% as we expected in heavy system load in Figure 5. A low collision rate indicates that the station effectively reduces the retransmission 167
  6. 6. overheads in the system. However, a low collision ratedoes not mean the least overhead in the system. Even alarge CW can result in a low collision rate. In the meantime, the duration of idle channels causes the efficiencyof the system to decrease. Therefore, a good backoffalgorithm should also examine another metric such asend-to-end delay. Figure 7 Normalized delay vs. number of stations4.4 End-to-End Delay 6. References The variation of the mean end-to-end delay with [1] IEEE WG, “Part 11: Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY)number of active stations is presented in Figure 6. As Specifications,” IEEE 802.11 Standard, 1999.expected, the delay increases with the number of [2] Andras Veres, Andrew T. Campbell, Michael Barry, andstations. The objective of the PCB algorithm is that it Li_Hsiang Sun, “Supporting Service Differentiation inprecisely estimates the actual network status and sets Wireless Packet Networks Using Distributed Control,”the corresponding CW to minimize overheads in the IEEE Journal on Selected Areas in Communications, vol.system. In Figure 6, the PCB shows the advantage of 19, no. 10, October 2001.overhead reduction and obtains the lowest delay [3] Chonggang Wang, Bo Li, and Lemin Li, “A Newamong these backoff algorithms. In Figure 7, we Collision Resolution Mechanism to Enhance thepresent the gain on end-to-end delay by normalizing Performance of IEEE 802.11 DCF," IEEE Transaction on Vehicular Technology, vol. 53, no. 4, pp. 1235-1246,end-to-end delay of a standard DCF. The delay of the July 2004.PCB is around 20% less than that of a standard DCF in [4] Bianchi G., “Performance analysis of the IEEE 802.11the case of n = 50. However, the end-to-end delay is distributed coordination function,”IEEE Journal ontoo large for multimedia services. The reason for the Selected Areas in Communications, vol. 18(3), pp. 535-results is that 802.11b is throughput-oriented without 547, 2000.considering QoS requirements. [5] N.-O. Song, B.-J. Kwak, J. Song, and L. E. Miller, “Enhancement of IEEE 802.11 distributed coordination5. Conclusion and Future Work function with exponential increase exponential decrease backoff algorithm,” in Proc. IEEE VTC-Sprin, 2003, vol. 4, pp. 2775-2778. In this paper, we have proposed a PCB backoff [6] Lamia Romdhani, Qiang Ni, and Thierry Turletti,algorithm to improve the efficiency of IEEE 802.11 “Adaptive EDCF: Enhanced Service Differentiation forDCF function. Our PCB algorithm estimates system IEEE 802.11 Wireless Ad-Hoc Networks”, in Proc. IEEEstatus by using a pause count backoff counter and WCNC 2003, vol. 2, pp.1373-1378determines a proper contention window size that [7] Network Simulator 2, http://www.isi.edu/nsnam/ns .accurately matches current network conditions. We [8] R. Jain, D. Chiu, and W. Hawe, “A Quantitative Measurecompared the performance of PCB with past proposed of Fairness and Discrimination for Resource Allocation in Shared Computer Systems,” DEC Research Reportalgorithms such as IEEE 802.11 DCF, EIED, and TR-301, 1984.AEDCF. Our simulation results demonstrate that PCBoutperforms these previously proposed algorithms forvarious performance metrics and dynamically adapts tothe variations in a network. In the future, we plan toconsider an extension of this work to considerRTS/CTS mode and wireless transmission errors. Figure 6 End-to-end delay vs. number of stations 168