Performance Evaluation of Collision Avoidance Schemes … · Performance Evaluation of Collision...

Technical Report # CSSE12-09

Performance Evaluation of Collision Avoidance Schemes inAd-hoc NetworksSuryakant Bhandare∗, Saad Biaz∗, Prathima Agrawal∗

Auburn University, Auburn, AL, 36849 USA

ABSTRACT

IEEE 802.11 Distributed Coordination Function (DCF) is a popular protocol used for the physical and MAC layers inmost ad hoc networks. DCF employs carrier sense multiple access with collision avoidance (CSMA/CA) and a binaryslotted exponential back off. It has been observed that the hidden and exposed terminal problems among stations degradeDCF’s performance in terms of throughput and fairness. Hence, the effectiveness of the IEEE 802.11 DCF mechanismin ad hoc networks has attracted many research studies. Many collision avoidance schemes were proposed to address theIEEE 802.11 DCF weaknesses as the performance of the MAC layer directly impacts the performance of higher-layerprotocols and hence the entire network. An evaluation of representative schemes under the same conditions will be helpfulin understanding the limitations and strengths of these schemes. This paper surveys various collision avoidance schemes,classify them based on their mechanism, and then provide a comparative study of representative schemes based on theirperformance. They are evaluated on a chain topology, a pair topology, and a random topology with static environment tomeasure their throughput, fairness, collision probabilities, and delay. Based on the evaluation, we conclude that GDCF(Gentle DCF) is the best scheme that has lesser collisions with improved throughput and fairness. A comparison with thelegacy CSMA/CA suggests that these proposed schemes do tend to be promising and would inspire future researchers whoare interested to find solutions to the age-old collision and fairness issues in ad-hoc networks.

∗CorrespondenceSuryakant Bhandare, Auburn University, Auburn, AL, 36849 USA.E-mail: [email protected] Biaz, Auburn University, Auburn, AL, 36849 USA.E-mail: [email protected] Agrawal, Auburn University, Auburn, AL, 36849 USA.E-mail:[email protected]

1. INTRODUCTION

A wireless ad hoc network is a congregation of wirelessstations that can configure themselves to form a networkwithout the assistance of any infrastructure. Wirelessstations sharing the same communication channel shouldget a fair chance to access the medium. Fairness is a keyrequirement a MAC protocol must provide.

Collisions of frames on a shared single channel areineluctable: single-channel wireless ad hoc networkssuffer from the hidden sender/receiver and exposedsender/receiver problems. A MAC protocol must provideeffective mechanisms for collision avoidance and faircontention resolution [1].

Figure 1 illustrates the hidden terminal problem.Stations A and H are the senders. Station H is in StationB’s range but cannot hear Station A. When both StationsA and H want to send frames to their common neighborStation B, collisions occur at B such that B hears onlygarbled frames. In this scenario, ”sensing the common

Figure 1. Hidden Terminal Problem

channel before an attempt to access the channel” doesnot prevent collisions and leads to a network performancedegradation. This scenario is referred to as the hiddenterminal problem. Station H is a hidden terminal from

Performance Evaluation of Collision Avoidance Schemes in Ad-hoc Networks S. Bhandare et al.

Station A. The hidden node problem causes unfairness inwireless networks as nodes are unable to coordinate theiraccesses [2].

To understand the exposed terminal problem, considerStation E in the range of Station A, but not in that ofreceiver Station B as shown in Figure 2. With the regularcarrier sensing mechanism, Station E will unnecessarilydefer from accessing the shared channel when A attemptsto transmit to B. Thus Station E is prevented fromtransmitting. Station E is the exposed terminal whosetransmission to C is deferred.

Figure 2. Exposed Terminal Problem

Collisions present an important design challenge forwireless local area networks (LANs). At the time ofcollision, considerable radio resources are wasted becausethe source node keeps transmitting the garbled packet. Theperformance degradation due to these collisions tends toworsen as the frame size increases. Hence, the amountof resources wasted on an ad-hoc networks dependsupon the size of the packet. Moreover, collisions causeretransmissions. This consumes a significant amount ofenergy, thus reducing the lifetime of battery-poweredwireless devices. The problem of collisions is worsein a multi-hop environment than in a single-hopenvironment [3]. Collision Avoidance is implementedin the medium access control (MAC) layers. A MACprotocol should provide an effective mechanism to sharethe available limited spectrum resources while deliveringhigh throughput and fairness to all the stations [4].

In the past few years, multiple MAC protocols weredeveloped to improve throughput, fairness, delay andcollision avoidance. A MAC protocol can be contentionor reservation based. Contention based MAC protocolsare used for applications requiring bursty traffic under alight network load and low delay. A reservation basedMAC protocol is generally used for applications with real-time requirements providing high QoS (Quality of Service)under heavy network load [5, 6]. They are typically usedin centralized network architectures. A classic unfairnessissue arises in a wireless mesh network when a nodenearer to the access point keeps sending its own traffic to

the access point gateway while starving the farther nodesfarther [7].

Research on contention based MAC protocolsstarted with ALOHA in the 1970s. Later, MACA [8],MACAW [9], FAMA [10] and DFWMAC [11] wereproposed by incorporating the carrier sense multipleaccess (CSMA) technique as well as the RTS and CTShandshaking mechanism for collision avoidance (CA).

This paper surveys, categorize, and evaluates represen-tative popular collision avoidance approaches in wirelessLANs. This paper also addresses the common issues thatmust be considered to design a collision avoidance scheme.

The remainder of this paper is organized as follows:Section 2 reviews popular collision avoidance schemes.This is followed by a classification of the CA schemesin Section 3. We then discuss the schemes we selectedfor performance evaluation in Section 4. We present thesimulation settings and discuss the performance of thiswork in Section 5 and then conclude our work in Section 6.

2. SURVEY

IEEE 802.11g has been the most common standardfor wireless local area networks. This standard is anenhancement of previously defined standards IEEE 802.11and IEEE 802.11b.

IEEE 802.11g uses OFDM(Orthogonal Frequency-Division Multiplexing) in addition to CCK (Complemen-tary Code Keying) as its modulation method to supportdata rates of up to 54 MBit/s. It is compatible and inter-operable with 802.11b as it shares the medium access, linklayer control, and the frequency range of 2.4 - 2.485 GHzwith its predecessors.

This section provides a brief overview of the collisionavoidance techniques of IEEE 802.11g and other proposedschemes.

2.1. Evolution of Collision Avoidance Methods

ALOHA is the earliest MAC protocol proposed for wirelessnetworks [12, 13]. In pure ALOHA, the stations transmitwhenever they have data to send, regardless of the currentstate of the medium. An acknowledgement is required bythe sender to determine if the transmission was successful.If no acknowledgement is received, a collision is assumedto have occurred and the sender must retransmit after arandom delay. Since ALOHA does not use carrier sensing,a data packet is vulnerable to collisions if another nodeis transmitting at the same time. A subsequent variationcalled slotted ALOHA was introduced to improve channelefficiency.

The fundamental cause of ALOHA’s low channelutilization is that senders do not listen to other users’transmissions [13] before sending. This led to thedevelopment of Carrier Sense Multiple Access (CSMA),in which a station attempts to avoid collisions by listening

2

S. Bhandare et al. Performance Evaluation of Collision Avoidance Schemes in Ad-hoc Networks

to the carrier of other user’s transmission prior to itstransmission.

In [14], Tobagi and Kleinrock showed that the existenceof hidden terminals significantly degrades the performanceof CSMA. The hidden terminal problem arises whentwo senders who cannot hear each other attempt tocommunicate with a common neighbor. To resolve thisissue, they proposed the Busy Tone Multiple Access(BTMA) [14] protocol as an extension to CSMA. InBTMA, the available bandwidth is split into two separatechannels: a data channel and a control channel. A busy-tone is placed on the control channel by the receiver aslong as the receiver station sense messages on its datachannel. The other stations sense this busy-tone channeland determine the state of the data channel at receivers andif busy, the senders defer their transmissions. This resolvesthe hidden terminal problem. BTMA can also solve theexposed terminal problem. A sending station could justignore the carrier sense signal when there is no busy-tone.BTMA, with its split channels, is actually a complex systemfrom a hardware perspective. The other problem associatedwith BTMA is that it increases the spectrum requirementsof each user station [8].

To avoid the need for a continuous busy-tone signal, theSplit-channel Reservation Multiple Access(SRMA) [15]was proposed. This is the first protocol to proposehandshakes between sending and receiving stations. Thisprotocol requires a separate channel for transmitting thehandshake signals. This requirement was unnecessary asthe data transfer does not start until the handshakingcomplete. Hence, the same channel could be used for thehandshake signals too.

The BTMA and SRMA extensions to CSMA arenot efficient and hence the Multiple Access CollisionAvoidance (MACA) was proposed by Karn [8]. MACAavoids the use of physical carrier sensing. MACA uses theRTS (Request-to-send) and CTS (Clear-to-send) handshakebefore data exchange. In MACA, any station overhearingRTS would defer so that the transmitting station canreceive the expected CTS and any station overhearingCTS would defer to avoid colliding with the imminentdata transmission. Thus MACA overcomes the hidden andexposed terminal problems. If the sender does not receivea CTS packet in response to its RTS packet, the sender usesa randomized exponential backoff algorithm to backofffor a random time before retrying. Each station waits arandomly chosen interval and tries again and will doublethe average interval on each successive attempt.

MACA suffers from unfairness and hence Bharghavan etal. [9] proposed Multiple Access with Collision Avoidancefor Wireless (MACAW). MACAW improves MACA witha new message sequence employing additional controlpackets and a Multiplicative Increase and Linear Decreasebackoff algorithm to improve both throughput and fairness.Unlike MACA’s backoff algorithm, in MACAW the valueof the backoff counter is increased by a factor of1.5 instead of 2 for each collision, and decreased by

1 for each success to provide a gentle reduction ofthe backoff value. MACAW uses a RTS-CTS-DS-DATA-ACK message exchange sequence. A short data-send(DS) packet precedes DATA packet, which is to informneighbors that a data packet is about to be transmitted.This reduces the probability of the packet to collide.The following ACK packet improves the reliability of thesystem. An additional control packet RRTS (Request forRTS) was used to let the receiver contend for the sender toimprove fairness in cases when there are two receivers inthe vicinity of each other and only one can win. MACAWdoes not generally solve the exposed terminal problem.

Distributed Foundation Wireless Medium AccessControl (DFWMAC) [11] protocol supports the basicaccess method(CSMA/CA) and allows an option to use theRTS/CTS based access method. The binary exponentialbackoff (BEB) algorithm used in DFWMAC protocol helpsease contention when the network load increases. Thecontention window size is doubled after every unsuccessfultransmission till it reaches a fixed maximum value. Thecontention window size returns to the minimum value if adata packet is successfully transmitted.

The IEEE 802.11 is a widely used WLAN technologythat is partially based upon DFWMAC. It supports highspeed communications up to 54 Mbps in the unlicensedbands at 2.4 GHz and 5 GHz [16]. The IEEE 802.11standard specifies two MAC schemes: a mandatoryDistributed Coordination Function (DCF) and an optionalPoint Coordination Function (PCF). Out of these, DCF ismost widely implemented in WLAN technologies (basedon DFWMAC) because it is simple and offers an efficientbest-effort service. DCF uses carrier sense multiple accesswith collision avoidance (CSMA/CA). With CSMA/CA, astation transmits its frame only if the medium is sensedidle. If the medium is determined to be busy, the stationdefers until the end of the current transmission. After thedefer period is over, or prior to attempting to transmitagain immediately after a successful transmission, thestation will select a random backoff interval and willdecrement the backoff interval counter while the mediumis sensed idle [17]. A transmission is considered successfulwhen an ACK frame is received by the sender. Thecollision avoidance technique uses a random backoff priorto the transmission of each frame. The random backoffprocedure reduces the probability of collision, but cannotcompletely eliminate the collisions since it is quite possiblethat two or more stations finish their backoff proceduressimultaneously. As the number of contending stationsincreases, the number of collisions will likely increase.

Figure 3 [16] illustrates the basic access method of IEEE802.11 in DCF mode. If a station wants to transmit a framewhen there is no ongoing backoff procedure, it determineswhether the medium is idle or not. If the medium is sensedidle, the station immediately proceeds with its transmissionafter an idle interval equal to the DCF Inter Frame Space(DIFS). For a busy medium, the station defers its access

3


Defer Access Backoff after Defer

Figure 3. Basic Access Mechanism of IEEE 802.11 DCF

until the medium is determined to be idle for a DIFSinterval, and then it starts the backoff procedure.

For the backoff procedure, the station sets its backofftimer to a randomly selected backoff time based on thecurrent contention window size CW

Backoff Time (BT) = Random() * aSlotTime (1)

Random() is an integer randomly selected from anuniformly distributed interval [0, CW].

The station performs a backoff procedure after a DIFSidle time. The carrier sensing technique is adopted tocheck if there is any activity during each backoff slot.If the medium is idle during a particular backoff slot,then the backoff time is decremented by a slot time(BTnew = BTold - aSlotT ime). During a busy slot,the decrementing is suspended. After the medium isdetermined to be idle for a DIFS period, the decrementingresumes. The transmission resumes when the backoffcounter reaches zero. If the destination receives the packetsuccessfully without errors, an acknowledgement (ACK)packet is sent after a short inter-frame space (SIFS)idle period, by the destination. The contention windowCW of the source station resets to the initial (minimum)value CWmin with the acknowledgement. For everyunsuccessful transmission, the contention window CWsize is increased, beginning with the initial value CWmin,up to the maximum value CWmax (in the IEEE 802.11specification). If the previously received frame containserror then a station defers forEIFS (Extended Interframespace) duration instead of DIFS before transmitting aframe. This process is called binary exponential backoff(BEB), which resolves collisions in the contention cycle.

To improve channel efficiency for long packet transmis-sions or in heavy-contending WLAN environments, theIEEE 802.11 protocol can use a short Request To Send(RTS) control frame and a short Clear To Send (CTS)frame to reserve access to the channel before a data frametransmission. The mechanism is shown in Figure 4. TheRTS/CTS access method is used to alleviate the hiddenterminal problem. RTS and CTS packets include a fieldcalled Network Allocation Vector (NAV). The stationsoverhearing the RTS and/or the CTS packets are informedhow long they should defer access to the channel by theNAV.

According to the IEEE 802.11 standard, the transmitterdecides whether to use the RTS frame. The RTS frame isused when the size of the pending data frame is equal

Figure 4. RTS-CTS exchange mechanism of IEEE 802.11

to or larger than the RTS threshold value (2347 bytes).However, most 802.11 devices operating in infrastructure-based WLANs with access points (APs), do not use theRTS/CTS exchange by setting the RTS threshold to thelargest value. RTS and CTS frames are rarely observedon real world networks. Note that even with RTS/CTSmechanism, we may still have the hidden terminal problemwith IEEE 802.11 networks.

In scenarios wherein there is unevenly distributed trafficload, a station that senses less competing nodes has morechances of succeeding in sending packets and thus ofresetting its contention window to the minimum value. Astation which is successful has a shorter backoff timer thanother stations that were not successful. When the trafficload is high, these other stations may suffer excessivelylong access delays, severe throughput degradation andultimately, starvation as the traffic load increases.

2.2. Conclusions

In this section, we discussed the evolution of collisionavoidance schemes - from ALOHA to IEEE 802.11 DCF.We looked at the problems associated with these schemesthat led to the development of better schemes. We alsoreviewed some popular schemes used that address theproblem of collisions in wireless networks. In the nextsection, we will classify the schemes we surveyed basedon their characteristics.

3. CLASSIFICATION OF COLLISIONAVOIDANCE SCHEMES

3.1. Classification

IEEE 802.11 is a widely deployed standard for WLANsystems and still has multiple issues to deal with.The hidden and exposed terminal problem are someof the critical issues that must be addressed. Theseproblems directly affect throughput, fairness, and collisionavoidance performance. Researchers have since proposedseveral schemes/protocols to address these issues. Inthis paper, we classified the various collision avoidanceschemes for an ease of evaluation. This classificationis based on the mechanism used by these schemes totackle collisions. A scheme can fall into more than onecategory as its mechanism can be a function of more thanone variable. We broadly classify these schemes into fivecategories as follows:

4


1. Backoff Tuning: These schemes primarily mod-ify the backoff timers and/or adjust the size ofthe contention window. Some of the schemes thatcome under this category are Enhanced CollisionAvoidance (ECA) [18], Fast Collision Resolution(FCR) [4], Gentle DCF (GDCF) [19], Probabilitybased DCF (PDCF) [20], Neighbourhood BackoffAlgorithm (NBA) [21], Real-time MAC [22], Lin-ear/Multiplicative Increase and Linear Decrease(LMILD) [23], Dynamically tuned 802.11 [24],Neighbour Aware CA-MAC [25], and Logarithmicbased MAC [26]. Our evaluation focuses on theschemes in this category. We selected ECA, FCRand GDCF for evaluation, and deal with them indetail in Section 4.

2. Reservation Based/ Virtual Carrier Sensing: Theseschemes propose handshaking methods to tacklethe collision and hidden/exposed terminal prob-lems. These schemes use handshake messagesto emulate physical carrier sensing, and hencehave been termed as Virtual Carrier Sensingschemes. Some of the schemes in this category areMACA [8], MACAW [9], MACA-BI (MACA byInvitation) [27], MACA-P (with Enhanced Paral-lelism) [5, 6], Distributed Backoff Reservation andScheduling (DBRS [28], Sender’s TransmissionRange Cover (STRC), Receiver’s TransmissionRange Cover (RTRC), Sender’s Carrier-sensingRange Cover (SCRC) and Receiver’s Carrier-sensing Range Cover (RCRC) [29], SEEDEX [30],and DCF+ [31].

3. Power based: These schemes vary the transmissionpower depending upon the packet type. Some ofthese schemes advertise interference margins. Fewprotocols belonging to this category are PowerControlled Multiple Access (PCMA) [32], PowerControlled Dual Channel (PCDC) [33], Powercontrolled MAC (POWMAC) [34], Improved busy-tone scheme using DBTMA [35], STRC, RTRC,SCRC, and RCRC [29].

4. Out-of-Band Signalling: These schemes use aseparate channel for signalling. Usually this channelis a control channel used to indicate other userswhether the medium is busy or not. Some schemesin this category are Busy-tone Solution [14],DBTMA [3], and Improved busy-tone [35].

5. Seminal Ideas: The schemes are seminal ideas.Schemes that can be considered novel are statisticsbased MAC [36], Busy-tone solution [14], Grant-to-send [37], MACA [8], and SRMA [15].

Table I lists the above categories of collision avoidancealgorithms. Even though not all the proposed protocolscan guarantee an A grade performance, it is seen thatthey tend to address and improve one or more issues.Based on the issues and performance metrics they address,we can further classify them as shown in Table II. Weobserve that backoff tuning improves the throughput and

delivers a better fairness. The hidden and exposed terminalproblems are addressed by schemes that have a reservationmechanism that would notify neighbors that they need tostay quiet for a transmission that is about to commence.Power based MACs usually focus on throughput andspectrum efficiency. The schemes that use an extra channelfor sending control signals address the hidden terminalissue.

3.2. Conclusions

In this section, we discussed popular schemes proposedby researchers. We looked at the functionality of theseschemes and provided a classification based on the stepsthey adopt to avoid/mitigate collisions. We also lookedat the specific issues that a particular type of schemeaddresses. For this paper, we will focus on the schemes inthe back off-tuning category from which we selected threeschemes (ECA, FCR, and GDCF). We detail these schemesin the next section.

4. SELECTED SCHEMES

This section briefly overviews the collision avoidanceschemes we will evaluate. We looked at the issues theseschemes address and selected three schemes from theback off-tuning category of our classification. The selectedschemes are ECA, GDCF, and FCR. These schemes havebeen very popular in the recent years and many recentworks have cited them. This is the reason why theygot selected for evaluation. In addition, we make surethat these schemes address similar type of issues. Theseschemes are evaluated and compared with the performanceof IEEE 802.11 DCF with and without the RTS-CTSmechanism.

4.1. Enhanced Collision Avoidance

CSMA/ECA’s main feature is that it attempts to reducecollisions by using a deterministic back off aftersuccessful transmissions [18]. CSMA/ECA introduces anew parameter: a deterministic back off value (V ) aftersuccessful transmissions. V is defined as:

V = d(CWmin− 1)/2e (2)

For our evaluation, we set V to 15 and CWmin

to 31. The proposers of this work - Jaume Barcelo,Boris Bellalta, Cristina Cano, Anna Sfairopoulou, MiquelOliver -, show that CSMA/CA and CSMA/ECA deliver thesame throughput when the network can handle all theoffered traffic. When the offered load exceeds the networkcapacity, CSMA/ECA outperforms CSMA/CA.

They show that by choosing a deterministic back-off after a successful transmission and a random back-off otherwise, the system converges to a collision-freeoperation when the number of active stations is not greaterthan the value of the deterministic back-off. The scheme

5


Table I. Classification of CA Schemes

Backoff Reservation/Virtual Power Out-of-band OriginalTuning carrier Mechanism Based Signalling Ideas

ECA, FCR, GDCF, PDCF,NBA, DCF+, EIED, EILD,HBAB, MACAW, DFWMAC,Real-time MAC, LMILD,Dynamically tuned 802.11,Neighbor Aware CA-MAC,Logarithmic based MAC

MACA, MACAW,MACA-BI, MACA-P,DFWMAC, DBRS,STRC, RTRC, SCRC,RCRC, SEEDEX

PCMA, PCDC,POWMAC, Improvedbusy-tone schemeusing DBTMA, STRC,RTRC, SCRC, RCRC

Busy-toneSolution,DBTMA,Improvedbusy-tone

SRMA, Busy-tone solution,Grand-to-send, MACA,SELECT

Table II. Specific issues addressed by the CA scheme categories

Type of CA schemes

Backoff Reservation/Virtual Power Out-of-bandTuning carrier Mechanism Based Signalling

Issues addressed

Throughput X XFairness X X

Collisions X X XMean Delay X

Hidden-Terminal problem X XExposed-Terminal problem X

Spectrum Efficiency X XPower Efficiency X X

relies on the deterministic behavior to stabilize the system,in case of a successful transmission, and the randomness ofthe back-off to avoid collisions. They provide a theoreticalanalysis to show that when a deterministic back-off isused, two stations that successfully transmitted in theirlast transmission attempt cannot collide among them.This gradually reduces the number of collisions. After allstations have had successful consecutive transmissions, theoperation of the system is collision-free and deterministic.

4.2. Gentle DCF

GDCF is a modification of IEEE 802.11 DCF. In 802.11DCF, the contention window is reset to the initialvalue after each successful transmission, which essentiallyassumes that each successful transmission is an indicationthat the system is under a lighter load. GDCF takesa more conservative measure by halving the contentionwindow size after a fixed number of consecutive successfultransmissions [19]. They show that this gentle reductiondecreases the probability of collision when the load is high.

According to the collision resolution process of GDCF,if there are c consecutive successful transmissions, GDCFwill halve the contention window CW and select aback off timer value uniformly from [0, CW]. Then, thecounter for recording the number of continuous successfultransmissions is reset to zero. Otherwise, GDCF increasesthe counter for the number of consecutive successful

transmissions and maintains the same contention window.On a collision, GDCF will double the contention windowand select a random backoff value similar to DCF’s butGDCF also resets the counter for recording the numberof consecutive successful transmissions. The idle statebehavior of GDCF is the same as DCF’s. ChonggangWang, Bo Li, and Lemin Li [19] showed that the optimalvalue of counter c is in the range of 4 and 8 and is nearlyindependent of the number of nodes when there are morethan 10 nodes. When the counter c is in the optimal rangeof 4 and 8, GDCF improves throughput by about 15%to 20% for large networks but just a small improvementfor small networks. They also show that better fairnesscould be obtained by using a higher value for c. For oursimulations, we set c to 8.

4.3. Fast Collision Resolution

FCR (Fast Collision Resolution) is a novel MAC protocolproposed by Younggoo Kwon, Yuguang Fang and HaniphLatchmanand. It is an efficient contention-based MACprotocol for WLANs. The rationale behind this algorithmis to speed up the collision resolution by activelyredistributing the back off timers for all active nodes.The algorithm also reduces the average number of idleslots by using smaller contention window sizes for nodeswith successful packet transmissions. The back off timeris decremented exponentially when a fixed number of

6


consecutive idle slots are detected [4]. The key features ofFCR are:

1. A smaller initial (minimum) contention windowsize CWmin than IEEE 802.11’s;

2. A larger maximum contention window sizeCWmax than IEEE 802.11 MAC;

3. An increase in the contention window size ofa station when it is in either collision state ordeferring state;

4. A back off timer that reduces exponentially fasterwhen a prefixed number of consecutive idle slotsare detected.

The FCR algorithm works as follows. When a stationfollowing FCR’s backoff procedure senses an idle slot, itdecrements its back off timeBT by a slot time. The stationwill transmit a packet when its back off timer reacheszero. If [(CWmin + 1) * 2 - 1] consecutive idle slots aredetected, the back off timer is decremented exponentially.After that, if idle slots are detected, the back off timer willbe halved until either it reaches zero or it senses a busyslot, whichever comes first. On collision, the contentionwindow size of the station is increased and a randomback off time is chosen. For a successful transmission, thecontention window is reduced to CWmin and a randomback off time BT value is accordingly chosen. When astation detects the start of a new busy period, either due toa collision or because of the occupation of the medium, thestation will increase its contention window size and pick anew random back off time.

In the FCR algorithm, the station which successfullytransmitted has the smallest CW size and a smaller backoff timer, hence it has a higher probability of grabbingthe medium, while others have relatively larger CW sizesand larger back off timers. After a number of successfulpacket transmissions for one station, another station maywin the contention and this new station will then havehigher chance of getting access to the medium. Table IVshows the parameter values specific to FCR.

Table III presents the highlights of the selectedalgorithms. Since we are specifically looking into the backoff-redistribution category, we only consider how thesealogrithms process the contention window and back offtimer.

4.4. Conclusions

This section presented the schemes selected to beevaluated: IEEE 802.11 CSMA, IEEE 802.11 CSMA withRTS-CTS, ECA, GDCF, and FCR. We also gave a briefoverview of the functionality of these schemes. Thefollowing section will analyze the performances of theseschemes under various scenarios.

Figure 5. Topologies: a) Chain b) Pair c) Random

5. PERFORMANCE EVALUATION

In this section, we describe the experiments to evaluatethe performance of the selected schemes. As wirelesssystems are complex, it is not always possible to capture allrelevant features with just one topology while performingthe analysis. For an extensive evaluation, we consider threescenarios:

1. Chain Topology2. Pair Topology3. Random Topology with varying node density

The topologies mentioned above are shown in Figure 5.We evaluated these schemes on OMNeT++ [38] withINETMANET Framework, since the current versionprovides significant support for comprehensive simulationsetups for studying IEEE 802.11 systems. We use theIEEE 802.11g model available in this framework. TheMAC layer is operating at a bit rate of 54Mbps for dataframes and a basic bit rate of 2Mbps for RTS/CTS andACK frames. Each node has a transmitter power of 2mWand a receiver sensitivity of -85dBm that results in atransmission range of 250m. Table IV shows the parametersettings for this analysis. The nodes are allowed to startsending at the same time so they get a fair chance tocontend from the beginning of the simulations. To isolatethe performance of MAC layer from that of higher layers,we use a constant bit rate UDP traffic as TCP adds itsown congestion avoidance/rate control in the performanceanalysis of 802.11. We compare the performances in termsof throughput, fairness, number of collisions, and meanpacket delay. These metrics are defined as follows:

1. Throughput: Throughput is measured as thenumber of bits received over the total simulationtime. When we mention aggregate throughput, wemean the average of throughputs measured forall packet sizes. The bit rate of IEEE 802.11networks is distance dependent: the farther are twostations, the lower the bit rate. We also observethat the actual data throughput is generally lessthan half of the rated speed i.e. a system workingat a rate of 54 Mbps would yield only at most27 Mbps in real throughput. This is because802.11 uses a collision avoidance techniquerather than the collision detection method. Wiredsystems can detect a collision, but wireless cannot,

7


Table III. Highlights of the Selected Backoff Redistribution Schemes

Backoff Redistribution Schemes

IEEE 802.11ECA GDCF FCR

CSMA/CA

Event

Transmission Success

Reset CW to initialvalue, select a randombackoff timer valueuniformly from [0,CW]

Reset backoff counterto 15 (deterministicbackoff)

Count the numberof consecutivesuccessfultransmission tillthe threshold (8),don’t alter CW

Reset CW to initialvalue, select a randombackoff timer valueuniformly from [0,CW]

Collision

Double CW andselect a randombackoff timer valueuniformly [0, CW]


Double CW, select arandom backoff timervalue uniformly [0,CW], reset counter ’c’


Channel Idle

Decrease backofftimer by one slot



Decrease backofftimer by one slot till 7consecutive idle slots

Channel BusySuspend backoff Suspend backoff Suspend backoff Suspend backoff

Consecutive Idle slots

same as any idle slotdetection



Halve backoff timerwhen 7 consecutiveidle slots are sensed

Consecutive Success

same as any success-ful transmission


Halve the CW andselect a backoff timervalue from [0,CW] if’c’ consecutive suc-cessful transmissionshave happened


thus, with CSMA/CA, the sender waits for anacknowledgment from the receiver to determine ifthe packet was transmitted successfully. The highestthroughput measured in our simulation was 23Mbps when there was only a single sender-receiverpair.

2. Fairness: Fairness among stations is an importantissue in wireless networks. The fairness index canshow if resources are fairly allocated to each station.Here, we are using Jain’s fairness index formula[39]. Jain’s fairness index is calculated as

IJ =(∑n

i=1 ri)2

n∑n

i=1 r2i

(3)

where r is the throughput achieved on connectioni among n competing hosts. Jain’s fairness indexalways lies between 0 and 1. A fairness index of1 indicates that it is the most fair algorithm.

3. Collision Rate: The collision performance isevaluated using the collision rate. Collision ratemeasures the percentage of packets that areinvolved in collisions. This is measured at thereceiver. For calculations, we consider the packetsthat were sent originally and omit the retransmittedpackets.

Table IV. Parameters used for Evaluation

Parameters Values

Slot time 20µsSIFS 10µsDIFS 50µsBit Rate 54Mbps

CWmin3 for FCR

31 for others

CWmax2047 for FCR

1023 for others

Retry Countern = 0 to 9 for FCR

n= 0 to 7 for othersRTS 160 bitsCTS 112 bitsACK 112 bitsPHY Header Length 192 bits

4. Mean Packet Delay: Mean packet delay is theaverage of the end-to-end delay of all receivedpackets.

8


5.1. Chain Topology

The topology used to evaluate the schemes is as shown inFigure 6. It is a four-node chain topology. This topologypresents the long-term fairness issues [40]. Long-termfairness is the fairness measured over a long period oftime. The unfairness issue arises because the two emitterscannot hear eachother and one receiver is pressed withmore collisions than the other receiver [41].

As shown in Figure 6, Nodes A and C can communicatewith node B but both (A and C) cannot hear eachother. Aalways senses the medium free, whereas any transmissionfrom C to D always succeeds if it manages to start, ascollisions cannot happen at D. The sender who starts firstdetermines the overall performance of this network.

Figure 6. Four node Chain Topology

We measure the throughput, fairness, delay andcollision rate on Flow F1 (A to B) and Flow F2 (C to D)for the collision avoidance schemes selected for evaluation.The nodes are 230m apart. Each sender sends 100k packetsto its destination over a simulation period of 50 seconds.We simulated this for 20 runs. The metrics are recorded atthe destination. The parameters for evaluation are given inTable V.

Table V. Settings used for Chain Topology

Parameters Values

Data Type Constant Bitrate UDPMessages generated per second 2000

Message Length 100-1400 bytesSimulation Time 50 seconds

5.1.1. ThroughputThe throughputs for all the selected schemes is shown

in Figure 7. FCR achieves the best throughput, deliveringan aggregate throughput of 5.17 Mbps, followed byGDCF with 4.98 Mbps. The worst scheme is IEEE802.11 with RTS-CTS, with an aggregate throughtput ofonly 2.13 Mbps. When compared with the legacy IEEE802.11 CSMA/CA, FCR achieved an improvement of 43%whereas RTS-CTS scheme showed a 41% degradation .

Figure 8 plots the cumulative distribution of throughputsachieved for each scheme. The RTS-CTS scheme performsworst: 80% of all flows achieve less than 2.8 Mbps.Furthermore, FCR and GDCF both shows a similar

0

2

4

6

8

10

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

Throug

hput

[Mbp

s]

Packet Size [bytes]

802.11 CSMA802.11 RTS-CTS

ECAGDCFFCR

Figure 7. Relationship between throughput and the packet sizes

0

20

40

60

80

100

0 1 2 3 4 5 6 7 8 9 10 11 12

CumulativeDistributionofThroughput

Throughput [Mbps]


ECAGDCFFCR

Figure 8. CDF of throughput received by flows

distribution, with 80% of all flows receive less than 7Mbps.

5.1.2. FairnessIn Figure 9, ECA shows better fairness than all the other

schemes. It follows almost the same trend as the legacyscheme as shown in the plot. ECA shows 91.66% fairnessfor the chain, close to IEEE 802.11 CSMA/CA’s 88.66%.We observe that FCR delivers higher throughputs, but atthe cost of fairness (58.47%). The unfairness of FCR canbe clearly seen in Figure 10, wherein D receives almost 8Mbps while B receives 2 Mbps. GDCF is shown to have abalanced performance in terms of throughput and fairnessamongst the nodes.

The cumulative distribution in Figure 11 shows that80% of all flows have a fairness index less than 0.65 withFCR. FCR is obviously badly performing. On the otherhand, ECA flows achieve a very good fairness: 80% offlows in ECA have a fairness less than 0.93, which is nearlysame as IEEE 802.11 basic access scheme.

9


0

0.2

0.4

0.6

0.8

1

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

FairnessIndex

Packet Size [bytes]


ECAGDCFFCR

Figure 9. Fairness-Index as a function of Packet-size

Figure 10. Throughput achieved at each node for each scheme

0

20

40

60

80

100

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Cum

ulativeDistributionofThroughput

Fairness Index


ECAGDCFFCR

Figure 11. CDF of Jain’s Fairness-Index

5.1.3. CollisionsFigure 12 plots how the collision rate varies with packet

size. FCR experiences a very drastic drop in performancebecause of collisions. Although the plot shows that FCRhas lesser collisions for higher packet sizes, it does notshow a very consistent performance for all packet sizes,like GDCF. GDCF maintains an aggregate collision rateof 5.08% which is a massive reduction of -48% whencompared with the legacy IEEE 802.11 CSMA/CA. Herethe aggregate collision is calculated by averaging thecollision rate for each size of packet.

0

5

10

15

20

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

CollisionRate[%]

Packet Size [bytes]


ECAGDCFFCR

Figure 12. Collision Rate v/s Packet Size

0

0.1

0.2

0.3

0.4

0.5

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

Delay[seconds]

Packet Size [bytes]


ECAGDCFFCR

Figure 13. Mean Packet Delay vs Packet Size

5.1.4. Packet DelayThe packet-delay performance of these schemes as a

function of packet-size is shown Figure 13. As expected,the reservation mechanism of RTS-CTS introducesadditional delay in packet transmission, delivering a meandelay of 278 msec. This is an increase of approximately191% compared to IEEE 802.11 CSMA/CA. We alsoobserve that GDCF offers the lowest delay by reducing themean delay by 53%.

5.1.5. SummaryHere we summarize the performance of all the schemes

on a chain topology. A brief summary is shown in Table VI.The bold numerals are the minimum or maximum achievedvalues for the measured metrics. It was observed that ECAhas the best fairness and FCR achieves the best throughput.But overall, GDCF achieves the best performance withlowest delays, lowest collision rate, and a relatively goodthroughput, and fairness performance.

5.2. Pair Topology

This is a critical scenario where strong unfairness appears.It is based on the asymmetry between the pairs, and

10


Table VI. Summary of Performance for Chain Topology

Scheme Mean Delay[ms] Collision Rate [%] Throughput [Mbps] Fairness[%]802.11 CSMA/CA 95.67 9.83 3.61 88.66802.11 RT-CTS 278.48 6.72 2.13 70.95ECA 86.21 9.18 3.69 91.66GDCF 44.86 5.08 4.98 83.35FCR 48.68 5.867 5.17 58.47

on the use of the EIFS delay. In this scenario, threepairs of communicating nodes are considered as shown inFigure 14. We consider that the adjacent senders are withintransmission range. Here, A and E cannot hear each other.But as C is able to sort an understanding with both A andE, its flow should achieve a higher throughput than theother senders.

Figure 14. Three Pair Topology

The parameters for evaluation are given in Table VII.We used the same simulation settings used for the chaintopology.

Table VII. Settings used for Pair Topology

Parameters Values


Message Length 100-1400 bytesSimulation Time 50 seconds

5.2.1. ThroughputThe throughput performance for a three - pair topology

is shown in Figure 15. GDCF shows the best throughputperformance over the other schemes. It delivers anaggregate throughput of 3.83 Mbps, which is significantlybetter than the other schemes. The worst performeramongst the schemes is IEEE 802.11 with RTS-CTS, thatdelivers an aggregate throughtput of only 1.58 Mbps.A reservation mechanism is bound to deliver a lowerthroughput because of the delay it introduces. The delayperformance of IEEE 802.11 with RTS-CTS confirms this.

Figure 16 plots the throughput achieved by each flow foreach scheme. ECA, with its deterministic back off process,tries to stabilize the network and thus introduces fairness.

0

1

2

3

4

5

6

7

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400Th

roug

hput

[Mbp

s]Packet Size [bytes]


ECAGDCFFCR

Figure 15. Relationship between throughput and the packetsizes

Figure 16. Throughput achieved at each node for each scheme

Figure 17 plots the cumulative distribution of aggregatethroughputs for each scheme. It is clear from thedistribution that the use of RTS-CTS mechanism degradesthe performance of the system. About 80% of all flowsreceive less than 2 Mbps throughput for IEEE 802.11 withRTS-CTS. About 80% of all flows in GDCF receive morethan 2 Mbps.

5.2.2. FairnessFor the three pair topology, the fairness provided by

IEEE 802.11 CSMA/CA and IEEE 802.11 CSMA/ECAdeliver similar fairness of about 88% fairness. Figure 18shows that ECA delivers a better fairness comparable toIEEE 802.11 CSMA/CA. GDCF and FCR trade off fairnessfor the throughput, but this trade-off is less when we

11


0

20

40

60

80

100

0 1 2 3 4 5 6 7 8 9 10 11 12

CumulativeDistributionofThroughput

Throughput [Mbps]


ECAGDCFFCR

Figure 17. Relationship between throughput and the packetsizes

consider the fairness of other schemes. They are only 5%lesser fair than the other schemes.

0.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

FairnessIndex

Packet Size [bytes]


ECAGDCFFCR

Figure 18. Fairness-Index as a function of Packet-size

Figure 19 plots the cumulative distribution of fairnessindex over various packet sizes. FCR introduces heavy

0

20

40

60

80

100

0.7 0.75 0.8 0.85 0.9 0.95 1

Cum

ulat

ive

Dis

tribu

tion

ofTh

roug

hput

Fairness Index


ECAGDCF

FCR

Figure 19. CDF of Jain’s Fairness-Index

0

5

10

15

20

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

CollisionRate[%]

Packet Size [bytes]


ECAGDCFFCR

Figure 20. Collisions Rate v/s Packet Size

unfairness. About 80% of the flows experience fairness lessthan 0.84 with FCR. The maximum fairness achieved by allthese schemes is 0.9.

5.2.3. CollisionsFigure 20 illustrates the effect of the size of the

packets on the collisions rate. FCR is the worst performerwhen it comes to collisions in a pair topology. FCR’scollision rate is almost double the collision rate of IEEE802.11 CSMA/CA. GDCF shows a very steady collisionperformance throughout the packet-size variation. In fact,it shows a better performance than IEEE 802.11 CSMA/CAby nearly halving the overall number of collisions.

5.2.4. Packet DelayThe packet delay of these schemes as a function of

packet-size is plotted on Figure 21. As expected, theRTS-CTS handshake degrades the delay performance. Theaverage delay is 260 msec which has a bad impact onthe overall throughput performance. But GDCF is muchfaster, as was expected. It shows a mean delay of 95 msec,nearly comparable to FCR’s 97 msec. Both GDCF andFCR perform better than the legacy, reducing the overalldelay by about 22

5.2.5. SummaryTo summarize the behavior of the schemes in a pair

topology, GDCF has shown promising performance withrespect to delay, collisions rate, throughput. and fairness. Abrief summary is shown in Table VIII. The bold numeralsare the minimum and maximum achieved values for eachmetric.

5.3. Random Topology

This is an arbitrary topology where stations are uniformlydistributed. Each station can either be a sender or areceiver. The random placement algorithm places a stationrandomly such that one station has at least one station inits range. This ensures that the resource-wasting scenario

12


Table VIII. Summary of Performance for Pair Topology

Scheme Mean Delay[ms] Collision Rate [%] Throughput [Mbps] Fairness[%]802.11 CSMA/CA 122 7.65 3.13 88.18802.11 RT-CTS 260 6.97 1.58 87.12ECA 115 5.35 3.26 87.83GDCF 95 3.10 3.83 83.92FCR 97 14.16 3.42 83.04

0

0.1

0.2

0.3

0.4

0.5

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

Delay[seconds]

Packet Size [bytes]


ECAGDCFFCR

Figure 21. Mean Packet Delay v/s Packet Size

Table IX. Settings used for Random Topology

Parameters Values


Message Length 100-1400 bytesNumber of hosts 10,30 and100Simulation Time 10 secondsPlayground Area 600m x 600m

is avoided and that the network is fully connected. Weconsider 3 sub-scenarios for a random topology:

1. Low Density Random topology of 10 stations2. Medium Density Random topology of 30 stations3. High Density Random topology of 100 stations

The parameters set for evaluation purpose are providedin Table IX. We know that half of the stations are sendersand the remainder are receivers.

5.3.1. ThroughputThe throughput performance for the random topology

based on density is shown in Figures 22, 23, and 24. GDCFdelivers the best throughput, with an aggregate throughputof 1.56 Mbps, 0.26 Mbps, and 0.1 Mbps for low, mediumand high density topology respectively. GDCF performs42% better than the basic access method for low densityand 13% better for medium density. Also observed is

that IEEE 802.11 with RTS-CTS performs terribly, with adegradation of around 40% in all the three cases.

0

0.5

1

1.5

2

2.5

3

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

Throug

hput

[Mbp

s]

Packet Size [bytes]


ECAGDCFFCR

Figure 22. Throughput vs packet sizes (Low Density)

Figures 25, 26, and 27 plot the cumulative distributionof aggregate throughputs in low, medium, and highdensity topology respectively. The distribution indicatesthat GDCF performs best and the IEEE 802.11 with RTS-CTS mechanism performs worst.

5.3.2. FairnessThe fairness for the three densities is plotted in

Figures 28, 29, and 30. As the number of nodes increases,the fairness of the IEEE 802.11 CSMA/CA basic access

0

0.1

0.2

0.3

0.4

0.5

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

Throug

hput

[Mbp

s]

Packet Size [bytes]


ECAGDCFFCR

Figure 23. Throughput vs packet sizes (Medium Density)

13


0

0.1

0.2

0.3

0.4

0.5

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

Throug

hput

[Mbp

s]

Packet Size [bytes]


ECAGDCFFCR

Figure 24. Throughput vs packet sizes (High Density)

0

20

40

60

80

100

0 1 2 3 4 5 6 7 8

Cum

ulat

ive

Dis

tribu

tion

ofTh

roug

hput

Throughput [Mbps]


ECAGDCF

FCR

Figure 25. Throughput CDF (Low Density)

0

20

40

60

80

100

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Cum

ulat

ive

Dis

tribu

tion

ofTh

roug

hput

Throughput [Mbps]


ECAGDCF

FCR

Figure 26. Throughput CDF (Medium Density)

method is better than the rest, but the performances ofECA and GDCF are almost as good. GDCF yields a51% fairness for low density, 28.81% fairness for mediumdensity, and 28.81% fairness for high density.

Figures 31, 31, and 31 show the cumulative distributionof fairness indexes. As mentioned earlier, the IEEE 802.11

0

20

40

60

80

100

0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4

Cum

ulat

ive

Dis

tribu

tion

ofTh

roug

hput

Throughput [Mbps]


ECAGDCF

FCR

Figure 27. Throughput CDF (High Density)

0.2

0.25

0.3

0.35

0.4

0.45

0.5

0.55

0.6

0.65

0.7

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

FairnessIndex

Packet Size [bytes]


ECAGDCFFCR

Figure 28. Fairness-Index as a function of Packet-size (LowDensity)

0

0.1

0.2

0.3

0.4

0.5

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

FairnessIndex

Packet Size [bytes]


ECAGDCFFCR

Figure 29. Fairness-Index as a function of Packet-size (MediumDensity)

basic access method has more fairness in all the threerandom scenarios, similar to GDCF’s.

14


0

0.1

0.2

0.3

0.4

0.5

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

FairnessIndex

Packet Size [bytes]


ECAGDCFFCR

Figure 30. Fairness-Index as a function of Packet-size (HighDensity)

0

20

40

60

80

100

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Cum


Fairness Index


ECAGDCFFCR

Figure 31. CDF of Fairness Index (Low Density)

0

20

40

60

80

100

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Cum


Fairness Index


ECAGDCFFCR

Figure 32. CDF of Fairness Index (Medium Density)

5.3.3. CollisionsFor the number of collisions, FCR is the worst

performer in all the three random scenarios. FCR hasalmost twice the number of collisions as the best performerin those scenarios. Figure 34, 35, and 36 shows the

0

20

40

60

80

100

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Cum


Fairness Index


ECAGDCFFCR

Figure 33. CDF of Fairness Index (High Density)

0

5

10

15

20

25

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

CollisionRate[%]

Packet Size [bytes]


ECAGDCFFCR

Figure 34. Collision Rate v/s Packet Size (Low Density)

0

10

20

30

40

50

60

70

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

CollisionRate[%]

Packet Size [bytes]


ECAGDCFFCR

Figure 35. Collision Rate v/s Packet Size (Medium Density)

collisions performance of the schemes in our randomtopologies.

FCR’s bad performance is clearly shown in thecumulative plots in Figure 37, 38, and 39.

15


0

20

40

60

80

100

120

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

CollisionRate[%]

Packet Size [bytes]


ECAGDCFFCR

Figure 36. Collision Rate v/s Packet Size (High Density)

0

20

40

60

80

100

0 1000 2000 3000 4000 5000

Cum


Collisions


ECAGDCFFCR

Figure 37. CDF of Average Collisions (Low Density)

0

20

40

60

80

100

0 2000 4000 6000 8000 10000 12000 14000

Cum

ulat

ive

Dis

tribu

tion

ofTh

roug

hput

Collisions


ECAGDCF

FCR

Figure 38. CDF of Average Collisions (Medium Density)

5.3.4. Packet DelayThe packet delay of these schemes as a function of

packet-size is shown on Figures 40, 41, and 42. The IEEE802.11 CSMA/CA with RTS-CTS experience higher delaysin all the three cases. FCR showed a relatively lower delayas compared with the other the schemes.

0

20

40

60

80

100

0 5000 10000 15000 20000 25000 30000

Cum


Collisions


ECAGDCFFCR

Figure 39. CDF of Average Collisions (High Density)

0

0.1

0.2

0.3

0.4

0.5

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

Delay[seconds]

Packet Size [bytes]


ECAGDCFFCR

Figure 40. Mean Packet Delay v/s Packet Size (Low Density)

0

0.2

0.4

0.6

0.8

1

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

Delay[seconds]

Packet Size [bytes]


ECAGDCFFCR

Figure 41. Mean Packet Delay v/s Packet Size (MediumDensity)

Figure 43, 44, and 45 clearly shows how the reservationscheme performs when we consider the mean delay.

5.3.5. SummaryA brief summary for the random topologies is shown in

Table X, XI, and XII. FCR shows high number of collisions

16


0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400

Delay[seconds]

Packet Size [bytes]


ECAGDCFFCR

Figure 42. Mean Packet Delay v/s Packet Size (High Density)

0

20

40

60

80

100

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Cum

ulativeDistribu

tionof

Throug

hput

Delay [seconds]


ECAGDCFFCR

Figure 43. CDF of Mean Delay (Low Density)

0

20

40

60

80

100

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Cum

ulativeDistribu

tionof

Throug

hput

Delay [seconds]


ECAGDCFFCR

Figure 44. CDF of Mean Delay (Medium Density)

but has a good delay performance. On the other hand,GDCF achieves comparatively higher throughput, lowercollisions, and a better fairness. The bold numerals are theminimum and maximum achieved values for each metric.

0

20

40

60

80

100

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Cum

ulativeDistribu

tionof

Throug

hput

Delay [seconds]


ECAGDCFFCR

Figure 45. CDF of Mean Delay (High Density)

6. CONCLUSIONS

In this paper, we first described some of the problemsassociated with a wireless network. We described thehidden/exposed terminal problem that leads to collisions.We also looked at the fairness issue and the limitationsof the throughput achieved by the IEEE 802.11 MAClayer. Then, we surveyed the schemes that address theseproblems and studied the evolution of these schemes andtheir limitations.

We also proposed a classification of collision avoidanceschemes to understand the variables that lead to theseproblems and the ways in which they can be resolved.We also study and evaluate three schemes (ECA, FCRand GDCF), selected from one of the five categories inour classification. To provide a rigorous evaluation, wesimulate them in a chain, pair, and random topologies.The metrics that we measured are throughput, fairness,collision rate, and mean packet delay. We compare theperformance of these schemes against the legacy IEEE802.11 DCF.

In our evaluation, we observed that GDCF performsbetter than the other protocols. GDCF delivers animproved throughput, an optimal fairness and a lowcollision rate. The wild oscillations of the BinaryExponential Back off (BEB) has been reduced in GDCF.BEB is a very aggressive approach. It reduces the backoff counter to its minimum value immediately upon asuccessful transmission. These large variations in the backoff counter after every successful transmission increasethe probability of contention. GDCF applies a minoradjustment to the back off computation by halving the backoff counter after a fixed number of consecutive successfultransmissions. This improves the efficiency of MAC layerand hence the entire network.

17


Table X. Summary of Performance for Low Density Random Topology


Table XI. Summary of Performance for Medium Density Random Topology


Table XII. Summary of Performance for High Density Random Topology

Scheme Mean Delay[ms] Collision Rate [%] Throughput [Mbps] Fairness[%]802.11 CSMA/CA 797.96 52.76 0.0656 19.81802.11 RTS-CTS 858.62 48.66 0.0416 9.51ECA 793.07 53.25 0.0657 18.87GDCF 731.48 63.64 0.09 12.34FCR 621.71 100.88 0.0429 10.15

REFERENCES

[1] Bharghavan V. Performance evaluation of algorithmsfor wireless medium access. Computer Performanceand Dependability Symposium, 1998. Proceedings.IEEE International 1998; 0:86 – 95.

[2] Wang X, Kar K. Throughput modelling andfairness issues in CSMA/CA based ad-hoc networks.INFOCOM 2005. 24th Annual Joint Conference ofthe IEEE Computer and Communications Societies.Proceedings IEEE, vol. 1, 2005; 23–34 vol. 1.

[3] Haas Z, Deng J. Dual Busy Tone Multiple Access(DBTMA) - A Multiple Access Control Schemefor Ad Hoc Networks. IEEE Transactions onCommunications, 2002; 975–985.

[4] Kwon Y, Fang Y, Latchman H. A Novel MACProtocol with Fast Collision Resolution for WirelessLANs. IEEE Infocom, 2003; 793–807.

[5] Acharya A, Misra A, Bansal S. MACA-P: A macfor concurrent transmissions in multi-hop wirelessnetworks. First IEEE International Conferenceon Pervasive Computing and CommunicationsPERCOM, 2003; 505–508.

[6] Bianchi G, Tinnirello I. Kalman filter estimationof the number of competing terminals in an IEEE

802.11 network. INFOCOM 2003. Twenty-SecondAnnual Joint Conference of the IEEE Computer andCommunications. IEEE Societies, vol. 2, 2003; 844 –852.

[7] Kongara H, Kondareddy Y, Agrawal P. Fairness andGateway Classification Algorithm (GCA) in mul-tihop Wireless Mesh Networks. 41st SoutheasternSymposium on System Theory, 2009. SSST 2009.,2009; 77 –81.

[8] Karn P. MACA: A new channel access method forpacket radio. Computer Networking Conference, vol.9th, 1990; 134–140.

[9] Bharghavan V, Demers A, Shenker S, Zhang L.Macaw: a media access protocol for wireless LAN’s.Proceedings of the conference on Communica-tions architectures, protocols and applications, SIG-COMM ’94, ACM, 1994; 212–225.

[10] Fullmer C, Garcia-Luna-Aceves JJ. Floor acquisitionmultiple access (FAMA) for packet-radio networks.Proceedings of the conference on Applications, tech-nologies, architectures, and protocols for computercommunication, SIGCOMM ’95, ACM, 1995; 262–273.

[11] Diepstraten W, Ennis G, Belanger P. Dfwmac:Distributed foundation wireless medium accesscontrol. IEEE Document P802.11-93/190, 1993.

18


[12] Abramson N. THE ALOHA SYSTEM: Anotheralternative for computer communications. Proceed-ings of the November 17-19, 1970, Fall joint com-puter conference, ACM, 1970; 281–285.

[13] Kleinrock L, Tobagi F. Packet Switching in RadioChannels: Part I–Carrier Sense Multiple-AccessModes and Their Throughput-Delay Characteris-tics. IEEE Transactions on Communications 1975;23:1400–1416.

[14] Kleinrock L, Tobagi F. Packet Switching in RadioChannels: Part II–The Hidden Terminal Problem inCarrier Sense Multiple-Access and the Busy-ToneSolution. IEEE Transactions on Communications1975; 23:1417.

[15] Kleinrock L, Tobagi F. Packet Switching inRadio Channels: Part III–Polling and (Dynamic)Split-Channel Reservation Multiple Access. IEEETransactions on Communications 1976; 24:832 –845.

[16] Choi J, Yoo J, Choi S, Kim C. EBA: An enhancementof the IEEE 802.11 DCF via distributed reservation.IEEE Transactions on Mobile Computing 2004; .

[17] IEEE 802.11 Wireless LAN Medium Access Control(MAC) and Physical Layer (PHY) specifications.IEEE Standard, 2007.

[18] Barcelo J, Bellalta B, Cano C, Sfairopoulou A, OliverM. Carrier sense multiple access with enhanced colli-sion avoidance: A performance analysis. Proceedingsof the 2009 International Conference on WirelessCommunications and Mobile Computing: Connect-ing the World Wirelessly, 2009; 733–738.

[19] Wang C, Li B, Li L. A New Collision ResolutionMechanism to Enhance the Performance of IEEE802.11 DCF. IEEE 802.11 DCF, IEEE Transactionson Vehicular Technology 2004; 53:1235–1246.

[20] Wang C, Li B, Li L. A probability-based algorithmto adjust contention window in IEEE 802.11DCF. International Conference on Communications,Circuits and Systems, 2004. ICCCAS 2004. 2004;1:418 – 422.

[21] Taifour M, Nat-abdesselam F, Simplot-ryl D.Neighbourhood backoff algorithm for optimizingbandwidth in single hop wireless ad-hoc networks.Proceedings MobiWac ’05, 2005.

[22] Baldwin R, Davis N, Midkiff S, Raines R. Packetizedvoice transmission using RT-MAC, a wireless real-time medium access control protocol. SIGMOBILEMobile Computing and Communications Review July2001; 5:11–25.

[23] Deng J, Varshney P, Haas Z. A new backoff algorithmfor the IEEE 802.11 Distributed Coordination Func-tion. In Communication Networks and Distributed

Systems Modeling and Simulation, CNDS ’04, 2004;215–225.

[24] Cali F, Conti M, Gregori E. IEEE 802.11 protocol:Design and performance evaluation of an adaptivebackoff mechanism. IEEE Journal on Selected Areasin Communications 2000; 18:1774 –1786.

[25] Romaszko S, Blondia C. Neighbour -Aware, Colli-sion Avoidance MAC Protocol (NCMac) for MobileAd Hoc Networks. Wireless Communication Sys-tems, 2006. ISWCS ’06. 3rd International Symposiumon, 2006; 322 –326.

[26] Manaseer S, Ould-khaoua M. Logarithmic basedbackoff algorithm for MAC protocol in MANETs.Technical Report, University of Glasgow 2006.

[27] Talucci F, Gerla M, Fratta L. MACA-BI (MACA ByInvitation) - A Receiver Oriented Access Protocol forWireless Multihop Networks. In Proceedings IEEEPIMRC, 1997.

[28] Choi E, Lee W. Distributed backoff reservation andscheduling for collision mitigation in IEEE 802.11WLANs. International Conference on InformationNetworking, 2008, 2008; 1 –5.

[29] Shih KP, Chen YD, Chang CC. Adaptive Range-Based Power Control for Collision Avoidance inWireless Ad Hoc Networks. ICC, IEEE, 2007; 3672–3677.

[30] Rozovsky R, Kumar PR. SEEDEX: a MAC protocolfor ad hoc networks. Proceedings of the 2ndACM international symposium on Mobile ad hocnetworking & computing, MobiHoc ’01, ACM, 2001;67–75.

[31] Wu H, Peng Y, Long K, Cheng S, Ma J. Performanceof Reliable Transport Protocol over IEEE 802.11Wireless LAN: Analysis and Enhancement. INFO-COM ’02, 2002.

[32] Monks J, Bharghavan V, Hwu WM. A power con-trolled multiple access protocol for wireless packetnetworks. INFOCOM 2001. Twentieth Annual JointConference of the IEEE Computer and Communica-tions Societies. Proceedings. IEEE, vol. 1, 2001; 219–228.

[33] Muqattash A, Krunz M. Power Controlled DualChannel (PCDC) Medium Access Protocol forWireless Ad Hoc Networks. Proceeding of IEEEINFOCOM Conference, 2003.

[34] Muqattash A, Krunz M. POWMAC: a single-channelpower-control protocol for throughput enhancementin wireless ad hoc networks. IEEE Journal onSelected Areas in Communications 2005; 23:1067–1084.

19


[35] Wang P, Zhuang W. An Improved Busy-ToneSolution for Collision Avoidance in WirelessAd Hoc Networks . International Conference inCommunications (ICC), Istanbul, Turkey, 2006.

[36] Hu J, Raymond C. A statistics based design of MACprotocols with distributed collision resolution for adhoc networks. International Conference on WirelessNetworks, Communications and Mobile Computing,2005, vol. 2, 2005; 917 – 922.

[37] Choi J, Levis P. Grant to Send: Fairness and isolationin low-power wireless. Technical Report SING-06-01, Stanford Information Networks Group, 2006.

[38] Varga A. OMNeT++ 2010. URL http://www.omnetpp.org.

[39] Jain R, Chiu D, Hawe W. A Quantitative MeasureOf Fairness And Discrimination For ResourceAllocation In Shared Computer Systems. TechnicalReport DEC-TR-301, Digital Equipment Corporation1984.

[40] Bensaou B, Wang Y, Ko C. Fair medium access in802.11 based wireless ad-hoc networks. Proceedingsof the 1st ACM international symposium on Mobilead hoc networking & computing, MobiHoc ’00, IEEEPress, 2000; 99–106.

[41] Chaudet C, Dhoutaut D, Lassous I. Performanceissues with IEEE 802.11 in ad hoc networking. IEEECommunications Magazine 2005; .

20

http://www.omnetpp.org

http://www.omnetpp.org

Performance Evaluation of Collision Avoidance Schemes … · Performance Evaluation of Collision...

Documents

Transcript of Performance Evaluation of Collision Avoidance Schemes … · Performance Evaluation of Collision...