Providing Service Guarantees in 802.11e EDCA WLANs with Legacy Stations

Providing Service Guarantees in 802.11e EDCAWLANs with Legacy Stations

Albert Banchs, Member, IEEE, Pablo Serrano, Member, IEEE, and Luca Vollero, Member, IEEE

Abstract—Although the EDCA access mechanism of the 802.11e standard supports legacy DCF stations, the presence of DCF

stations in the WLAN jeopardizes the provisioning of the service guarantees committed to the EDCA stations. The reason is that DCF

stations compete with Contention Windows (CWs) that are predefined and cannot be modified, and as a result, the impact of the DCFstations on the service received by the EDCA stations cannot be controlled. In this paper, we address the problem of providing

throughput guarantees to EDCA stations in a WLAN in which EDCA and DCF stations coexist. To this aim, we propose a techniquethat, implemented at the Access Point (AP), mitigates the impact of DCF stations on EDCA by skipping with a certain probability the

Ack reply to a frame from a DCF station. When missing the Ack, the DCF station increases its CW , and thus, our technique allows usto have some control over the CWs of the legacy DCF stations. In our approach, the probability of skipping an Ack frame is dynamically

adjusted by means of an adaptive algorithm. This algorithm is based on a widely used controller from classical control theory, namely aProportional Controller. In order to find an adequate configuration of the controller, we conduct a control-theoretic analysis of the

system. Simulation results show that the proposed approach is effective in providing throughput guarantees to EDCA stations inpresence of DCF stations.

Index Terms—WLAN, 802.11, 802.11e, EDCA, DCF, ACKS, legacy stations, throughput guarantees, control theory.

Ç

1 INTRODUCTION

THE Wireless LAN (WLAN) technology is nowadayswidely used for the Internet access. One of the short-

comings of traditional WLANs, based on the 802.11standard [1], is that they provide no means to offer serviceguarantees to users. This is a significant drawback, inparticular, due to the inherent resource limitation in radiosystems. This shortcoming has been identified by theresearch community, who has devoted considerable effortover the last decade to the design of wireless local areanetworks (WLANs) with Quality of Service (QoS) support.Along this effort, the Enhancements Task Group (TGe) wasformed under the IEEE 802.11 WG to recommend aninternational WLAN standard with QoS support. Thisstandard is called 802.11e [2] and will be included in theongoing new revision of the 802.11 standard [3].

The 802.11e standard defines two different accessmechanisms: the Enhanced Distributed Channel Access(EDCA) and the HCF Controlled Channel Access (HCCA).This paper focuses on the former. The EDCA mechanism of802.11e was designed as an extension of the DistributedCoordination Function (DCF) mechanism of the legacy 802.11

standard. One of the key design goals of the EDCAmechanism was the backward compatibility with the legacyDCF mechanism. Following this goal, EDCA was designedsuch that legacy stations using DCF could operate in an802.11e WLAN under EDCA.

One of themain problems of the EDCAmechanism is that,although legacy DCF stations can interoperate in a WLANunder EDCA, they substantially degrade the performance ofthe WLAN and preclude the provisioning of serviceguarantees to the EDCA stations. Indeed, as we have notedin [4], [5], the fact that DCF (in contrast to EDCA) competeswith predefined contention parameters that cannot bemodified prevents controlling the aggressiveness of DCFstations. As a result, if EDCA stations competing againstaggressive DCF stations are to receive service guarantees,they will need to behave aggressively as well, and this willseverely degrade the overall WLAN performance.

Some effort in the literature has been devoted to theanalysis of WLANs in which EDCA and DCF stationscoexist (see, e.g., [6], [7], [8], [9]). Additionally, a number ofproposals have been made to improve the performance ofEDCA in presence of DCF stations, namely [10], [11], [12] inaddition to our previous works [4], [5].1 The main drawbackof [10], [11], [12] is that they require introducing modifica-tions into the DCF or the EDCA stations. In contrast, ourproposal of [4], [5] leaves the EDCA and DCF stationsuntouched, which represents a major advantage from adeployment perspective.

Following our previous works [4], [5], in this paper, weaddress the problem of providing throughput guarantees toEDCA stations in a WLAN with legacy DCF stations. Totackle this, we propose the Dynamic ACK Skipping (DACKS)

IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 8, AUGUST 2010 1057

. A. Banchs is with the Departamento de Ingenieria Telematica, UniversidadCarlos III de Madrid, Avda de la Universidad, 30, E-28911 Leganes(Madrid), Spain, and also with IMDEA Networks.E-mail: [email protected].

. P. Serrano is with the Departamento de Ingenieria Telematica, UniversidadCarlos III de Madrid, Avda de la Universidad, 30, E-28911 Leganes(Madrid), Spain. E-mail: [email protected].

. L. Vollero is with the CIR—Centro Integrato di Ricerca, UniversitaCampus Bio-Medico di Roma, Via Alvaro del Portillo, 21, 00128 Roma,Italy. E-mail: [email protected].

Manuscript received 21 May 2009; revised 6 Nov. 2009; accepted 3 Dec. 2009;published online 16 Mar. 2010.For information on obtaining reprints of this article, please send e-mail to:[email protected], and reference IEEECS Log Number TMC-2009-05-0185.Digital Object Identifier no. 10.1109/TMC.2010.52.

1. In [13], the authors used a similar idea to that of [4], [5] for a differentpurpose, namely to provide service differentiation in a WLAN with DCFstations only.

1536-1233/10/$26.00 ! 2010 IEEE Published by the IEEE CS, CASS, ComSoc, IES, & SPS

technique, which mitigates the impact of legacy stations onan 802.11e WLAN under the EDCA mechanism byimplementing a small modification in the 802.11e AccessPoint (AP). The main contributions of the paper aresummarized as follows:

. We propose the DACKS technique. The key featureof the approach (as compared to our previous works[4], [5]) is that the system is dynamically controlledbased on the observed behavior of the WLAN. Inparticular, DACKS is based on a common controllerfrom control theory.

. We develop a model of a WLANwith DACKS understationary conditions. Based on this model, wedetermine the optimal configuration of the EDCAparameters in order to provide EDCA stations withthroughput guarantees.

. We develop a model for the transient response of aWLAN controlled by DACKS. With this model, weanalyze the dynamics of our system from a control-theoretic standpoint to tune the DACKS parameters.

A longer version of this paper, containing proofs of thetheoretical results as well as additional simulation results, isavailable in [14].

2 802.11 DCF AND 802.11E EDCA

DCF and EDCA execute a similar algorithm to transmittheir frames. In the following, we first present the 802.11eEDCA mechanism and then we describe the differencesbetween 802.11e EDCA and 802.11 DCF.

EDCA regulates the access to the wireless channel on thebasis of the channel access functions (CAFs). A station mayrun up to 4 CAFs, and each of the frames generated by thestation is mapped to one of these CAFs. Then, each CAFexecutes an independent backoff process to transmit itsframes. A CAF i with a new frame to transmit monitors thechannel activity. If the channel is idle for a period of timeequal to the arbitration interframe space parameter (AIFSi),the CAF transmits. Otherwise, if the channel is sensed busy(either immediately or during the AIFSi period), the CAFstarts a backoff process. The arbitration interframe space(AIFSi) takes a value of the form DIFS ! nTe, whereDIFS is the DCF interframe space, Te is the duration of anempty slot time, and n is a nonnegative integer.

Upon starting the backoff process, the CAF computes arandom integer value uniformly distributed in the range"0; CWi # 1$ and initializes its backoff time counter withthis value. The CWi value is called the contention windowand depends on the number of transmissions failed for theframe. At the first transmission attempt, CWi is set equal tothe minimum contention window parameter (CWmin

i ). Aslong as the channel is sensed idle, the backoff time counteris decremented once every time interval Te, and “frozen”when a transmission is detected on the channel.

When the backoff time counter reaches zero, the CAFtransmits. A collision occurs when two or more CAFs starttransmission simultaneously. An acknowledgment (Ack)frame is used to notify the transmitting CAF that the framehas been successfully received. The Ack is immediatelytransmitted at the end of the frame, after a period of time

equal to the SIFS (the short interframe space). If the Ack isnot received within a time-out given by the Ack Timeout,the CAF assumes that the frame was not receivedsuccessfully and reschedules the transmission by reenteringthe backoff process. The CAF then doubles CWi (up to amaximum value given by the CWmax

i parameter), computesa new backoff time, and starts decrementing the backofftime counter at an AIFSi time following the time-outexpiry. If the number of failed attempts reaches a pre-determined retry limit R, the frame is discarded.

After a (successful or unsuccessful) frame transmission,before transmitting the next frame, the CAF must execute anewbackoff process.As an exception to this rule, the protocolallows the continuation of an EDCA transmission opportu-nity (TXOP). A continuation of an EDCA TXOP occurs whena CAF retains the right to access the channel following thecompletion of a transmission and transmits several framesback-to-back. The period of time a CAF is allowed to retainthe right to access the channel is limited by the transmissionopportunity limit parameter (TXOP limiti).

In the case of a single station running more than oneCAF, if the backoff time counters of two or more CAFs ofthe station reach zero at the same time, a scheduler insidethe station avoids the internal collision by granting the accessto the channel to the highest priority CAF. The other CAFsof the station involved in the internal collision react as ifthere had been a collision on the channel, doubling theirCWi and restarting the backoff process.

As it can be seen from the description of EDCA given inthis section, the behavior of a CAF depends on a numberof parameters, namely CWmin

i , CWmaxi , AIFSi, and

TXOP limiti. These are configurable parameters that canbe set to different values for different CAFs. The standarddraft groups CAFs by Access Categories (ACs), all theCAFs of an AC having the same configuration, and limitsthe maximum number of ACs in the WLAN to 4. AnEDCA station that wants to enter the WLAN must issue asignaling request indicating the AC that it wants to join. Ifadmitted, the EDCA station can join the WLAN with aCAF configured according to the parameters of thecorresponding AC. The parameters of each AC areannounced periodically by means of beacon frames.

A DCF station executes a very similar backoff process tothe one described above for an EDCA CAF, albeit withsome differences. One difference is the way the backoffcounter is managed. In EDCA, the backoff counter isresumed one slot time before the AIFS expiration, while inDCF, it is resumed after the expiration. Moreover, in DCF, astation transmits immediately when the counter decrementsto 0, while in EDCA, it transmits in the next slot time.2

Another key difference between DCF and EDCA is that,while in 802.11e EDCA, the contention parameters areconfigurable and can be set to different values for differentACs, in DCF, the values of these parameters are fixed by thestandard as follows:

. The AIFSi parameter in DCF is set equal to DIFS.

. The configuration of the CWmini and CWmax

i para-meters is predefined by the 802.11 DCF standard.

1058 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 9, NO. 8, AUGUST 2010

2. The reader is referred to [7] for further details about the backoffbehavior of EDCA and DCF.

We refer to the values given by the standard asCWmin

dcf and CWmaxdcf , respectively.

. Upon accessing the medium, DCF stations transmit asingle packet, and hence, do not use the TXOP limitiparameter.

While EDCA has been designed to allow coexistence withlegacy DCF stations, the fact that the contention parameterswith which DCF stations compete are fixed jeopardizes theprovisioning of service guarantees to EDCA stations. Therest of the paper is devoted to overcoming this limitation.

3 DACKS TECHNIQUE

As we have seen in the previous section, legacy DCFstations start the backoff process with a CW equal toCWmin

dcf . This initial CW is fixed by the standard to a smallvalue, and it only doubles after each failed attempt. Thesesmall CW values of DCF stations raise problems in aWLAN in which EDCA stations are to receive serviceguarantees. Indeed, no matter whether the CWs of theEDCA stations are configured with small or large values,the following drawbacks are observed when there are anonsmall number of stations in the WLAN:

1. If EDCA stations were configured with small CWvalues in order to give them a higher priority thanDCF stations, we would have both DCF and EDCAstations with small CWs and the resulting overallefficiency of the WLAN will be low, due to the factthat small CW values result in a high collision rate.

2. If EDCA stations were configured with large CWvalues in order to avoid the above problem, DCFstations would compete with smaller CWs thanEDCA and would consume most of the WLANresources, leavingEDCAstationswith little resources,and thus, failing to meet their service guarantees.

None of the above two alternatives is desirable, as inboth cases the service received by the EDCA stations isseriously degraded as a consequence of the impact of legacystations. Instead, it would be desirable to increase the CWof legacy stations; in this way, EDCA stations could receiveservice guarantees without compromising the overallefficiency. The DACKS technique achieves this goal withoutmodifying the legacy DCF stations.

DACKS is based on the following behavior of DCF: aftersending a packet, a DCF station waits for an Ack frame, andif the frame is not received within an Ack time-out, itassumes a collision and increases its CW . The central idea isthen the following: if the AP skips the Ack reply to legacy

DCF stations with a certain probability (hereafter referred toas Pskip), these stations will “see” a collision rate higher thanthe actual one, and will contend with larger CWs, resultingthis in a smaller impact on the EDCA stations.

The above behavior of DACKS is illustrated in Fig. 1. Inthe figure, the behavior of a DCF station in a WLANwithout DACKS is compared against the behavior of a DCFstation in a WLAN that uses the DACKS technique. It canbe observed that in the latter case, by skipping the Ackreply with some probability, DACKS achieves the objectiveof increasing the average CW with which the DCF stationcontends for channel access, and hence, reduces the numberof times that the DCF station transmits.

The challenge with the DACKS technique is the config-uration of the probability Pskip. This adds to the inherentdifficulty in 802.11e of configuring the EDCA contentionparameters in order toprovide thedesiredbehavior. In [4], [5]weproposed somealgorithms to computePskip statically. Themain drawbacks of a static configuration are the following:

. A static configuration has to compute the configura-tion assuming the worst case in which all DCFstations are constantly active. This requires a muchmore aggressive behavior than needed against DCFstations. In particular, when all DCF stations areactive, Ack frames need to be skipped with a highprobability to ensure the desired throughput guar-antees for EDCA. In contrast, if someDCF stations arenot active, a smaller skipping probability is enough toprovide EDCA stations with the desired service.

. Similarly to the above, a static configuration has toassume that all admitted EDCA stations are active,since this is the worst case to ensure the desiredguarantees. This assumption forces a high probabil-ity of skipping Ack frames, degrading, thus, DCFperformance. In the case some EDCA stations are notactive, the desired service could be provided whilereducing the degradation suffered by DCF.

We conclude from the above that a static configurationdegrades the performance of DCF stations unnecessarilywhen not all the (EDCA and DCF) stations are active. In thispaper, we propose an alternative scheme that, by dynami-cally adjusting the skipping probability to the currentbehavior of the WLAN, minimizes the disruption sufferedby the DCF stations.

4 EDCA CONFIGURATION

It follows from the above explanations that amajor challengefor anEDCAWLANwithDACKS is the configuration of both

BANCHS ET AL.: PROVIDING SERVICE GUARANTEES IN 802.11E EDCA WLANS WITH LEGACY STATIONS 1059

Fig. 1. DACKS technique.

the EDCA parameters and the DACKS skipping probability.In this section, we analyze the EDCA configuration,3 whilethe DACKS configuration is analyzed in the next section.

4.1 Scenario and AssumptionsIn the following, we describe the scenario considered in thispaper as well as the assumptions upon which our analysisis based:

. Our scenario consists of a WLAN, where EDCA andDCF stations coexist. Our goal is to provide EDCAstations in this scenario with throughput guarantees.

. We consider that each EDCA station executes onlyone CAF and joins a given AC i depending on itsthroughput requirements. We denote by Ri thethroughput guarantee given to the EDCA stationsof AC i.

. We assume that, over a time period, a station iseither constantly backlogged4 or does not transmitany traffic. We refer to the former as an active stationand the latter as inactive.

. We denote by Ndcf the number of active DCFstations in the WLAN and by Ni the number ofactive EDCA stations that belong to AC i.

. Following our previous results of [16], we use thefollowing configuration for the EDCA stations:AIFSi % DIFS and CWmin

i % CWmaxi , since [16]

shows (both analytically and via simulation) thatno other configuration provides better throughputperformance. We denote CWi % CWmin

i % CWmaxi .

. Following [17], we assume that backoff times aregeometrically distributed, i.e., a station at a givenbackoff stage transmits with a constant and inde-pendent probability in each slot time.

. Upon accessing the channel, both EDCA and DCFstations transmit a single packet of length l.

4.2 DCF Station Model

We start our analysis by computing the probability that aDCF station transmits at a randomly chosen slot time, !dcf ,as a function of the probability that a transmission attemptof a DCF station collides, cdcf .

Fig. 2 illustrates our model of a DCF station. The statesrepresent the backoff stage of the station, i.e., the number ofcollisions suffered by the current frame. At state 0, thestation’s CW is equal to CWmin

dcf , yielding the followingtransmission probability [15]:

!dcf;0 %2

CWmindcf ! 1

: "1$

Let m be the maximum backoff stage defined byCWmax

dcf % 2mCWmindcf . Note that in DCF, we have m < R [1].

At state i & m, theCW has been doubled i times, yielding thefollowing transmission probability:

!dcf;i %2

2iCWmindcf ! 1

: "2$

At state i > m, the CW has already reached CWmaxdcf ,

yielding

!dcf;i %2

2mCWmindcf ! 1

: "3$

In the rest of the paper, we use the following simplifyingapproximation for !dcf;i:

!dcf;i '2

2min"i;m$"CWmindcf ! 1$

% !dcf;02min"i;m$ : "4$

Following the above, we have that at state i, the stationtransmits in each slot time with probability !dcf;i. If thetransmission collides (which occurs with probability cdcf ),the station moves to the next state, and doubles its CW ifi < m. If it succeeds, the station goes back to the initial state0 and sets the CW equal to CWmin

dcf . When the stationreaches the maximum retry limit at state R, it moves back tostate 0 no matter if the transmission succeeds or collides.This leads to the state transition probabilities given in Fig. 2.

Let us denote by Pi the probability that the station is atstate i. The probability of entering state i is equal to theprobability of being at state i# 1 and performing a failedtransmission. The probability of leaving this state is equal tothe probability of performing a (failed or successful)transmission. By forcing equilibrium between these twoprobabilities, we have

Pi#1!dcf;i#1cdcf % Pi!dcf;i: "5$

Following (4), we have

!dcf;i %!dcf;i#1=2; i & m!dcf;i#1; i > m;

!"6$

which yields

Pi %Pi#12cdcf ; i & mPi#1cdcf ; i > m:

!"7$

Applying the above recursively leads to

Pi %P0"2cdcf$i; i & mP02mcidcf ; i > m:

!"8$

By forcingP1

i%0 Pi % 1 and isolating P0, we obtain

P0 %1

Pmi%0 "2cdcf$

i !PR

i%m!1 2mcidcf

% 11#"2cdcf $m!1

1#2cdcf! 2mcm

dcf"1#cR#m

dcf$

1#cdcf

:"9$

With the above, we can compute the transmissionprobability of a DCF station as follows:


3. In the EDCA configuration proposed here, we focus only on EDCAstations with throughput guarantees. In [14], we extend the configuration toBest-Effort EDCA stations.

4. [15] refers to constantly backlogged stations as saturated. In the rest ofthe paper, we use the terms “constantly backlogged” and “saturated”indistinctly.

Fig. 2. Markov chain model of a DCF station.

!dcf %XR

i%0

Pi!dcf;i

%"1# 2cdcf$

"1# cm!1

dcf

#!

"1# cdcf$"1# "2cdcf$m!1$!"1# 2cdcf$cm!1

dcf

"1# cR#m

dcf

#

"1# 2cdcf$2mcm!1dcf

"1# cR#m

dcf

# !dcf;0;

"10$

which terminates our model of a DCF station.

4.3 Throughput Analysis

Based on the model of a DCF station presented above, wenow analyze the throughput performance of DCF and EDCAstations in the WLAN. Our analysis is based on thefollowing: 1) after each transmission, there is a slot time inwhich DCF stations have not yet decremented their backoffcounter and only EDCA stations may transmit, 2) we assumethat EDCA and DCF stations transmit with a constant andindependent probability in those slot times where they areallowed, and 3) when computing their transmission prob-abilities, we account for the fact that EDCA stations wait forone extra slot time after the backoff counter reaches 0.

Equation (10) gives the transmission probability of a DCFstation as a function of the collision probability. Thetransmission probability of the EDCA stations, whoseconfiguration satisfies CWi % CWmin

i % CWmaxi , can be

easily computed as follows:5

!i %2

CWi ! 3: "11$

Further, the collision probability of the DCF stations canbe expressed as6

cdcf % 1# Pack"1# !dcf$Ndcf#1Y

i

"1# !i$Ni

!; "12$

where Pack is the probability that, upon successfullyreceiving a packet from a DCF station, the AP sends thecorresponding Ack—i.e., the probability that the DACKStechnique does not skip this Ack:

Pack % 1# Pskip: "13$

With the above, we can compute the transmissionprobability of all the stations of the WLAN as follows:

. The transmission probability of the EDCA stations, !i,can be computed from their configuredCWi with (11).

. Given all !is, we can compute !dcf by solving thenonlinear equation formed by (10) and (12).7

Once all the transmission probabilities have been ob-tained, we can compute the probability Pt that a given slottime contains a transmission (either a success or a collision) asfollows: If the previous slot time was empty, all stations maytransmit, otherwise, only EDCA stationsmay transmit. Thus,

1# Pt % "1# Pt$"1# !dcf$NdcfY

i

"1# !i$Ni

! Pt

Y

i

"1# !i$Ni ; "14$

which yields

Pt %1# "1# !dcf$Ndcf

Qi "1# !i$Ni

1!Q

i "1# !i$Ni # "1# !dcf$NdcfQ

i "1# !i$Ni: "15$

With the above, we can proceed to compute the through-put experienced by an EDCA station of AC i, ri, and thethroughput experienced by a DCF station, rdcf , as follows:

ri %!icil

"1# Pt$Te ! PtTt"16$

and

rdcf %Pack!dcf"1# !dcf$Ndcf#1 Q

j "1# !j$Nj

$ %l

"1# Pt$Te ! PtTt; "17$

where Te is the duration of an empty slot time, Tt is theduration of a slot time with a transmission, and ci is theprobability that a transmission attempt of an EDCA station ofAC i collides,

ci % "1# Pt$"1# !i$Ni#1"1# !dcf$NdcfY

j6%i

"1# !j$Nj

! Pt"1# !i$Ni#1Y

j6%i

"1# !j$Nj :"18$

The duration of an empty slot time (Te) is fixed by thestandard, while the duration of a slot time that contains asuccess and a collision is equal to, respectively:8

Ts % TPLCP !H

C! l

C! SIFS ! TPLCP !ACK

C!DIFS;

"19$

Tc % TPLCP !H

C! l

C! EIFS; "20$

where TPLCP is the Physical Layer Convergence Protocol(PLCP) preamble and header transmission time, H is theMAC overhead (header and FCS), ACK is the size of theacknowledgment frame, and C is the channel bit rate.

Since the standard fixes the value of EIFS equal to thetime required to send an Ack, we have that the duration of acollision and a success are equal, and we can thus computethe duration of a slot time with a transmission as

Tt % Ts % Tc: "21$

With the above, we can compute, given the configurationof the CWi and Pack parameters, the throughput of each ofthe DCF and EDCA stations in the WLAN, whichterminates the throughput analysis. In the followingsections, we address the configuration of these parameters.

4.4 CWi ConfigurationWe now address the issue of calculating the optimalconfiguration of the WLAN. The goal of the optimal


5. Note that (11) differs from (1) as it takes into account that an EDCAstation transmits at the next slot time when its backoff counter reaches 0.

6. Note that, by collision, here, we understand both the case when thetransmission actually collides and the case when, even if there is not a realcollision, the Ack is omitted by the DACKS technique.

7. The reader is referred to [18] for a discussion on the uniqueness of thesolution.

8. Note that, in case of a skipped Ack, the slot time duration is given byTs, since stations update their NAV to the duration of a successfultransmission, and defer channel access during this time.

configuration is to provide the desired throughput guaran-tees while maximizing the overall throughput performance.

Upon changing the CWi configuration, the AP needs todistribute the new configuration to the stations by means ofsignaling. This signaling limits the frequency with which theCWis values can be updated. In contrast to theCWis, the Pack

parameter is local and its value need not be sent to thestations. As a result, Pack can be updated as frequently asneeded with no associated signaling cost. Following this, inthis paper, we make the following choices:

. The CWi parameters are statically set based oninformation that does not change frequently, andtherefore, does not trigger frequent updates oftheir values.

. The Pack parameter is configured based on adynamic algorithm that constantly updates its valuefollowing the observed behavior of the WLAN.

In the remaining of this section, we address theconfiguration of the CWi parameters, while the dynamicalgorithm that updates Pack is presented in the next section.

Following the above argumentation, the computation ofthe CWi configuration needs to be based on data that do notchange frequently. In particular, we use the following data:

. The number of EDCA stations admitted in theWLAN and their required throughputs. These dataare available at the AP since EDCA stations, prior toentering the WLAN, have to issue an admissioncontrol request with this information.

. The number of DCF stations present in the WLAN.This information is available as DCF stations need togo through an authentication/association processbefore they enter the WLAN.

In contrast to the above data, Pack is constantly updated,and therefore, cannot be taken into account in thecomputation of the CWis. This raises an issue since theoptimal CWi configuration actually depends on the settingof this parameter. In order to overcome this problem, theapproach that we take in this paper is to compute theconfiguration of the CWis considering that Pack is set to 0.This suboptimal solution has the following advantages:

. The first advantage is that the solution becomesoptimal when the WLAN is stressed with manythroughput guarantee requests from the EDCAstations. This is due to the fact that, when theWLAN is stressed, the DACKS technique forces DCFstations to reduce drastically their transmission rateby setting Pack % 0, thereby making the computedCWi configuration optimal.

. The other advantage of the proposed configurationis that it allows maximizing the number of through-put guarantee requests that can be admitted. Indeed,if a request cannot be admitted when Pack is set to 0,this means that the request can never be admitted.

To compute the optimal configuration, we start byimposing the following condition, which ensures that thethroughput will be distributed among stations proportion-ally to their requests [19]:

!i"1# !j$!j"1# !i$

% Ri

Rj; "22$

where Ri is the throughput guarantee of AC i.

We note that, with the above equation, if we assume thatthe value of a given CWi is known, we can compute thevalue of all the other CWis. From the throughput analysisof Section 4.3 and taking Pack % 0, we can then compute allthe throughputs.

With the above, we proceed as follows to find theoptimal CWi configuration: We conduct a numerical searchusing the golden section search method over the CWi of theAC with the lowest throughput guarantee (without loss ofgenerality, we assume it is AC 1). For each CW1 valueevaluated in the search, we compute the other CWis from(22), and from these, we compute r1. With the numericalsearch, we find thus the CW1 value that leads to the largestr1. In order to avoid a large degree of unfairness with DCF,we impose in the search that CW1 cannot be smaller thanCWmin

dcf . Once the search finds CW1, we then compute all theother CWis, which terminates the algorithm.

Note that a requirement that must be met by the CWi

configuration given by the above algorithm is that theresulting ris are larger than the corresponding Ris. If thiscondition is not satisfied, this means that there exists no setof CWi values that meets the desired throughput guaran-tees even when Pack is set to 0. In this case, the requestedguarantees cannot be satisfied and the request thattriggered this computation must, therefore, be rejected.9

Note also that the above computation has been performedconsidering that Pack is set to minimally disrupt EDCAstations (specifically,Pack % 0).As a consequence, ifPack is notproperlyadjustedand takes largervalues, there is the risk thatthe throughput guarantees are not met. In the followingsection, we present an algorithm that adjusts Pack to ensurethat the committed throughput guarantees arenever harmed.

5 DACKS CONFIGURATION

In this section, we present an algorithm that updates thePack parameter dynamically. We start by analyzing theconditions that must be met by the setting of Pack. Next, wepropose a system based on control theory that, followingthese conditions, dynamically adjusts Pack. In order toanalyze the overall controlled system, we develop alinearized model of the system. Based on this linearizedmodel, we conduct a stability analysis to determine theregion of the system parameters that guarantees a stablebehavior. Finally, we obtain the setting of the parameters ofthe controlled system within the stability region.

5.1 Pack Configuration

Our goals for the setting of the Pack parameter are thefollowing ones:

. Given theCWi configuration obtained in the previoussection, we want to ensure that backlogged EDCAstations see their throughput guarantees satisfied.

. As long as the throughput guarantees for EDCAstations are met, we want to minimize the through-put degradation of the DCF stations by setting Pack

as large as possible.


9. Note that this request can come either from an EDCA or a DCF station.In the latter case, the AP can reject the request by not completing theassociation process initiated by the station. Note that many of today’s APsalready apply similar policies to deny association of stations based, e.g., ontheir MAC address or on the AP’s current load.

Following the above, the main goal for the dynamicalgorithm that computes Pack is to set it to the largest possiblevalue that satisfies the throughput requirements of the EDCAstations. We build the algorithm around the probability Pt

that a randomly chosen slot time contains a transmission.Note that (16) can be rewritten as a function of Pt:

ri %!i"1# Pt$l

"1# !i$ "1# Pt$Te ! PtTr" $ : "23$

Our algorithm is based on the following two observations:

. Given the CWi configuration of AC i, there exists amaximum Pt;max;i value such that, as long asPt & Pt;max;i, the throughput guarantee of AC i ismet. This can be seen from (23).

. The larger the Pt we allow, the smaller theprobability of skipping an Ack frame needs to be.One of the goals that we have stated above wasprecisely to make the probability of skipping an Ackframe as small as possible, in order to minimize thedisruption suffered by the DCF stations.

With the above observations, our objective can beformulated as to finding the Pack configuration that yieldsa transmission probability equal to

Pt;max % minifPt;max;ig; "24$

since this is the Pt value that minimizes the degradationsuffered by the DCF stations while meeting the throughputguarantees of all EDCA stations.

Pt;max;i can be obtained by imposing ri ( Ri and isolatingPt from (23). Given the Pt;max;is, we can then compute from(24) the value of Pt;max. Note that this value is a constantthat depends only on the CWi configuration obtained in theprevious section.

The remaining challenge is to design an adaptive algo-rithm that, byobserving the transmissionprobabilityPt in thechannel, adjusts Pack such that the channel’s transmissionprobability is equal to Pt;max. Note that the key advantage ofthe proposed algorithm is that, by monitoring the WLAN’sbehavior, we can adjust the probability of skipping an Ack tothe minimum value that current conditions allow, and thus,we disrupt legacy stations as little as possible. Specifically,note the following:

. With our algorithm, Pack is adjusted dynamically tothe behavior of the DCF stations. Indeed, as onlythe DCF stations currently active contribute to Pt,these are the only ones taken into account whenadjusting Pack.

. Pack is also dynamically adjusted to the behavior ofthe EDCA stations. Indeed, if some of the EDCAstations are not active, those do not contribute to Pt,and therefore, the setting of Pack is not unnecessarilypenalized because of them.

Following the above, we next design an algorithm basedon control theory that adjusts Pack as a function of the Pt

observed in the channel with the goal of forcing that this Pt

equals the target Pt;max.

5.2 DACKS Control SystemOur goal is to design a control law that drives thetransmission probability Pt to the desired target value Pt;max

computed in (24). To this aim, we build the closed-loopcontrol system illustrated in Fig. 3, which consists of thefollowing blocks:

. H"z$ represents the WLAN system. The system iscontrolled by Pack and its output is the occupation ofeach slot time (where an output of “1” means that aslot time is occupied and “0” that it is empty). Weconsider that this occupation function is given by theaverage transmission probability of the WLANsystem, Pt, added to some noise of zero mean,which we represent by N .

. C"z$ is the controller module. It takes the error,given by Pt;max # Pt, as input, and computes fromthis error the control signal.

. In order to eliminate the noise fed from N into thecontrol signal, we introduce (following the designguidelines of [20]) a low-pass filter F "z$ to eliminatethis undesired noise. The resulting control signalfree from noise is the probability of replying a framefrom a DCF station with an Ack, Pack.

For the transfer function of the controller C"z$, in thispaper, we focus on a very simple controller from classicalcontrol theory, namely the Proportional Controller [21]:

C"z$ % Kp: "25$

For the low-pass filter F "z$, we use a simple exponentialsmoothing algorithm of parameter " [22],

Fout)n* % "Fin)n* ! "1# "$Fout)n# 1*; "26$

where Fin and Fout are the input and output signals of thefilter, respectively.

Since the output of the filter F "z$ is the probability Pack,we need to enforce that it stays between 0 and 1. We do thisby setting

Pack)n* % max"0;min"1; Fout)n*$$; "27$

which generates the following clipping error:

e)n* % max"0;min"1; Fout)n*$$ # Fout)n*: "28$

In order to eliminate this error, we follow the strategy of[23] of subtracting the error of the previous sample into theinput of the following one. With this, (26) is rewritten as

Fout)n* % ""Fin)n* # e)n# 1*$ ! "1# "$Fout)n# 1*: "29$

In the analysis of the rest of this section, we assume thatFout keeps always in the range "0; 1$, andwe neglect the effectof the clipping error. With this assumption, F "z$ behaves asa first order filter with the following transfer function:

F "z$ % "

1# "1# "$z#1: "30$


Fig. 3. Block diagram of the controlled system.

It can be seen from the above that our control systemrelies on two parameters, namely Kp and ". The rest of thissection is devoted to analyzing the system with the goal offinding an appropriate setting for these parameters.

5.3 Transient Analysis of 802.11In the system illustrated in Fig. 3, we need to characterizethe WLAN transfer function H"z$. To this aim, the transientresponse of an 802.11 WLAN system has to be studied.While 802.11 has been widely analyzed under stationaryconditions (including our analysis presented in Section 4),its transient response to changing conditions has receivedmuch less attention. Indeed, although a number of papershave studied different aspects of the transient response of802.11 [24], [25], [26], to the knowledge of the authors, oursis the first attempt to analyze the transient response of thecomplete 802.11 protocol under general conditions.10

In our analysis, we will assume that the number of activeDCF stations and the number of active EDCA stations areconstant. Note that, with this assumption, the effect of allEDCA stations can be captured with the probability that aslot time contains the transmission of at least one EDCAstation. We denote this probability by Pedca. We furtherassume in what follows that stations never reach themaximum CW , which leads to the following simplifiedexpressions for (8) and (10):11

Pi % P0"2cdcf$i; "31$

!dcf % "1# 2cdcf$"1# cdcf$

!dcf;0: "32$

To model the transient behavior of the WLAN, our goalis to compute the probability that a DCF station transmits ata slot time n, !dcf )n*, given the transmission probability ofthe DCF station in the previous slot time, !dcf )n# 1*, and theprobability Pack. Note that in stationary conditions, we willhave !dcf )n# 1* % !dcf )n*.

The key approximation uponwhich we base our transientanalysis is the following. We assume that the relationshipbetween the state probability Pi and the transmissionprobability !dcf given by (31) and (32), which has beenderived under stationary conditions, also holds duringtransients. Specifically, we assume that at a given slot timen# 1, we have

Pi)n# 1* % 1# 2!dcf )n# 1* # !dcf;0!dcf )n# 1* # 2!dcf;0

& '

+ 2!dcf )n# 1* # !dcf;0!dcf )n# 1* # 2!dcf;0

& 'i

;

"33$

where !dcf )n# 1* is the transmission probability at thisslot time.

Given Pi)n# 1*, the state probabilities at the next slottime n can be computed as follows: if the station does nottransmit at time n# 1, it stays in the same state at time n; if

it transmits successfully, it moves to state 0; if it collides, itmoves to state i! 1. This yields

Pi)n* % Pi)n# 1*"1# !dcf;i$ ! Pi#1)n# 1*!dcf;i#1cdcf ; i > 0

"34$

and

P0)n* % P0)n# 1*"1# !dcf;0$ !X1

i%0

Pi)n# 1*!dcf;i"1# cdcf$;

"35$

where cdcf , the probability that a transmission at slottime n# 1 collides, is given by

1# cdcf % "1# Pedca$"1# !dcf )n# 1*$Ndcf#1"1# Pack$: "36$

With the above, we can compute !dcf )n* as follows: Bydefinition,

!dcf )n* %X1

i%0

Pi)n*!dcf;i: "37$

Applying (35) and (34) to Pi)n* in the above equation,we have

!dcf )n* % P0)n# 1*"1# !dcf;0$!dcf;0

!X1

i%0

Pi)n# 1*!dcf;i"1# cdcf$!dcf;0

!X1

i%1

Pi)n# 1*"1# !dcf;i$!dcf;i

!X1

i%1

Pi#1)n# 1*!dcf;i#1cdcf!dcf;i:

"38$

Recombining the above terms and considering that!dcf;i % !dcf;i#1=2, we obtain

!dcf )n* %X1

i%0

Pi)n# 1*!dcf;i #X1

i%0

Pi)n# 1*!2dcf;i

! "1# cdcf$!dcf;0X1

i%0

Pi)n# 1*!dcf;i

! cdcf2

X1

i%0

Pi)n# 1*!2dcf;i;

"39$

where the first term of (38) has been integrated into the firsttwo sums of the above equation.

The termP

Pi)n# 1*!dcf;i is, by definition, equal to!dcf )n# 1*. The term

PPi)n# 1*!2dcf;i can be expressed as

follows:

X1

i%0

Pi)n# 1*!2dcf;i %X1

i%0

1# 2!dcf )n# 1* # !dcf;0!dcf )n# 1* # 2!dcf;0

& '

+ 2!dcf )n# 1* # !dcf;0!dcf )n# 1* # 2!dcf;0

& 'i !dcf;02i

$ %2;

"40$

which, solving the series, yields

X1

i%0

Pi)n# 1*!2dcf;i %2!dcf )n# 1*!2dcf;0

3!dcf;0 # !dcf )n# 1*: "41$


10. In particular, Cali et al. [24] analyze a dynamic 802.11 protocol, whichis different from the standard one; Challa et al. [25] analyze the start-up of asimplified version of the protocol in which CWmin % CWmax and Foh andZukerman [26] analyze the recovery time under a disaster scenario. None ofthese analyses models the transient behavior with a transfer function thatcan be used for a control theory study.

11. We note that while this assumption was not necessary in the previoussections, it is needed here to make the transient analysis (which is morecomplex) tractable.

Finally, combining all the above, we obtain the follow-ing equation that describes the system behavior undertransient conditions:

!dcf )n* % !dcf )n# 1* ! "1# cdcf$!dcf;0!dcf )n# 1*

# "1# cdcf=2$2!dcf )n# 1*!2dcf;0

3!dcf;0 # !dcf )n# 1* ;"42$

where cdcf is a function of Pack given by (36).Note that by imposing stationary conditions (i.e.,

!dcf )n# 1* % !dcf )n*), the above equation results in (32).

5.4 Linearized Model

The above transient analysis has resulted in a nonlinearrelationship between !dcf and Pack. In order to analyze theproblem from a control-theoretic standpoint, we need toobtain a linear relationship that can be captured by atransfer function. To achieve this, we linearize (42) aroundthe stable point of operation of the system.12

The stable point of operation of the WLAN can beobtained from forcing !dcf )n# 1* % !dcf )n* in (42) andisolating !dcf . We express the perturbations around thispoint as !dcf !!!dcf . When these perturbations are small,they can be approximated by

!!dcf )n* '@!dcf )n*

@!dcf )n# 1*!!dcf )n# 1* ! @!dcf )n*

@Pack!Pack; "43$

where !dcf )n* is the right-hand-side expression of (42).The above expression provides a linear relationship

between !dcf )n* and Pack; however, in order to obtain H"z$,we need to find a linear relationship between Pt)n* and Pack.We do this as follows:

!Pt)n* '@Pt)n*

@Pt)n# 1*!Pt)n# 1* ! @Pt)n*

@Pack!Pack; "44$

where

@Pt)n*@Pt)n# 1* %

@Pt)n*@!dcf )n*

@!dcf )n*@!dcf )n# 1*

@!dcf )n# 1*@Pt)n# 1* %

% @!dcf )n*@!dcf )n# 1*

"45$

and

@Pt)n*@Pack

% @Pt)n*@!dcf )n*

@!dcf )n*@Pack

: "46$

With the above, we have the following expression for therelationship between !Pt and !Pack:

!Pt)n* % H1!Pt)n# 1* !H2!Pack; "47$

where the expressions for the coefficients H1 and H2 arecomputed from (45) and (46) in [14, Appendix I].

By doing theZ-transformof the above equation,we obtain

!Pt"z$ % H1!Pt"z$z#1 !H2!Pack"z$ "48$

from where by isolating !Pt"z$=!Pack"z$, we finallyobtain H"z$:

H"z$ % H2

1#H1z#1: "49$

5.5 Stability Analysis

We now study the system when it suffers perturbationsaround its point of operation and analyze the conditionsthat guarantee local stability.

Fig. 4 illustrates the linearized model when workingaround the stable operation point:

Pt % Pt !!Pt; "50$

Pack % Pack !!Pack: "51$

Note that, as compared to the model of Fig. 3, in Fig. 4,only perturbations around the stable operation point areconsidered.

The closed-loop transfer function of the system of Fig. 4is given by

T "z$ % C"z$H"z$F "z$1! z#1C"z$H"z$F "z$

: "52$

Substituting (25), (30), and (49) into the above yields

T "z$ % Kp"H2

"1# "1# "$z#1$$"1#H1z#1$ ! z#1Kp"H2; "53$

which can be rewritten as

T "z$ % Kp"H2z2

z2 ! a1z! a2; "54$

with

a1 % Kp"H2 #H1 # "1# "$; "55$a2 % H1"1# "$: "56$

A sufficient condition for stability is that the poles of theabove polynomial fall within the unit circle jzj < 1. This canbe ensured by choosing coefficients fa1; a2g of the character-istic polynomial that belong to the stability triangle [28]:

a2 < 1; "57$a1 < a2 ! 1; "58$a1 > #1# a2: "59$

Equation (57) is met given that 1# " < 1 and H1 < 1.Equation (59) is met given that

# 1# a2 % #1#H1 !H1" < 1#H1 ! " < a1: "60$

Equation (57) imposes the following restriction:

Kp"H2 #H1 # "1# "$ < H1"1# "$ ! 1 "61$


12. A similar approach was used in [27] to analyze RED from a control-theoretic standpoint.

Fig. 4. Block diagram of the linearized system.

from which we obtain the following restriction on Kp:

Kp <2# "

"

& '1!H1

H2: "62$

As long as the configuration of Kp is smaller than theabove expression, the system is guaranteed to be stable.However, H1 and H2 in the above expression are a functionof the number of active DCF stations, Ndcf , and the behaviorof the EDCA stations, given by Pedca. These values are notknown a priori and may vary with time.

In order to assure stability, we need to find some upperbound for Kp that guarantees stability independent of Ndcf

and Pedca. This bound is given by Theorem 1 (in [14,Appendix II]), which shows that as long as Kp is configuredsmaller than Kmax

p , the system will be stable, Kmaxp being a

constant value given by the following expression:

Kmaxp % 2# "

"

& '1!Hmin

1

Hmax2

; "63$

where the expressions for Hmin1 and Hmax

2 are given in [14,Appendix II]. This terminates the stability analysis.

5.6 Parameter SettingThe stability analysis conducted in the previous sectionprovides a range for the parameters values, where the systemis guaranteed to be stable. In this section, we propose specificrules for setting the parameters " and Kp within this range.Theproposed rulesaimat 1) ensuring that the systembehavesstably while reacting quickly to changes and 2) eliminatingfrom the system the noise caused by the oscillations of Pt. Inthe following, we first fix " and then, with the given value of", we setKp such that these two objectives are met.

The parameter " of the low-pass filter is fixed as follows:The goal of the low-pass filter is to eliminate the fluctua-tions introduced to the system by Pt. Since, with atransmission probability of Pt;max, there is approximatelyone transmission every Pt;max samples, the frequency thatneeds to be filtered out is approximately equal to 2#=Pt;max.Following this reasoning, we impose as design criterion thatthe low-pass filter reduces this frequency by a factor GF :

jF "2#=Pt;max$j % GF : "64$

With the above, the problem of configuring " is reformu-lated as to finding the value that satisfies (64). Combining thiswith (30) yields

"

1# "1# "$)cosw! j sinw*

((((

((((2

% G2F ; "65$

where

w % 2#

Pt;max: "66$

Operatingon the above,weobtain a secondorder equationfrom which we can isolate ":

" %#"1# cosw$ !

))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))))"1# cosw$2 ! 2

"G#2

F # 1#"1# cosw$

q

G#2F # 1

;

"67$

which terminates the setting of ".

Given the above " setting, we next address the config-uration of the parameterKp in order tomeet the two goals setat the beginning of this section. We start by analyzing thesetting of Kp following stability considerations.

From a stability standpoint, we have a trade-off betweensystem stability and speed of reaction to changes. The largerKp, the fastest the system reacts to changes; however, ifKp ischosen too large, the system becomes unstable (as we haveseen in the previous section). In order to determine the righttrade-off between these two effects in the setting of theKp parameter, we follow the Ziegler-Nichols rules [21],which are widely used to configure proportional controllers.According to these rules, we impose that this parametercannot be larger than one-half of the maximum value thatguarantees stability,

Kp & Kstabilityp %

Kmaxp

2: "68$

In addition to the above, Kp also needs to be setaccording to the objective of eliminating the noise fromthe system. The noise caused by the fluctuations of Pt

around frequency w is amplified into the input signal Pack

by jC"w$F "w$j. In order to avoid that this noise causes toolarge oscillations on the input signal, we impose as a designcriterion that this gain is no larger than a factor GCF :

jC"w$F "w$j % KpGF & GCF : "69$

Isolating Kp from the above equation, we obtain thelargest Kp allowed by the considerations on noise:

Kp & Knoisep % GCF

GF: "70$

Finally, based on the above, we configure Kp as followsto guarantee that the two objectives set at the beginning ofthis section on stability and noise are met:

Kp % min"Kstabilityp ;Knoise

p $: "71$

Note that the configuration proposed above depends onthe setting of two parameters, GF and GCF . To provideappropriate filtering and attenuate noise, these parametersshould take small values. Furthermore, to allow sufficientlylarge Knoise

p values, (70) imposes GCF , GF . Followingthese considerations, in this paper, we take GCF % 10#2 andGF % 10#4.

6 PERFORMANCE EVALUATION

In order to evaluate the performance of DACKS, we haveperformed an exhaustive set of simulation experiments. Forthe simulations, we have extended the simulator used in[16], [29]; this is an event-driven simulator that closelyfollows the details of the MAC protocol of 802.11 EDCA. Forall tests, we have taken a fixed frame payload size of 1,000bytes and the system parameters of the IEEE 802.11bphysical layer [30]. For the simulation results, average and95 percent confidence interval values are given (note that inmany cases, confidence intervals are too small to beappreciated in the graphs). Sections 6.1-6.9 focus on a singleEDCA Access Category (AC 1) and saturated conditions,while the experiments of Sections 6.10 and 6.11 extend theevaluation to more than one AC and nonsaturation,respectively. Additional simulation results and moredetailed explanations can be found in [14].


6.1 Throughput GuaranteesIn our first experiment, we evaluated the ability of DACKSto provide throughput guarantees to the EDCA stations. Tothis aim, we considered a scenario with Nedca EDCAstations, all belonging to the same AC (AC 1), and Ndcf

DCF stations. The EDCA stations were given a throughputguarantee of 300 Kbps. We took Nedca % Ndcf % N andvaried N from 2 to the maximum number of stationsallowed by our admission control algorithm. The results ofthis experiment are illustrated in Fig. 5. Analytical resultsare represented with lines, and simulations with pointswith error bars. A horizontal line is used to show theguaranteed throughput. We conclude from the figure thatthe proposed DACKS technique is effective in providingthroughput guarantees, since EDCA stations never have athroughput below 300 Kbps. Additionally, we observe thatanalytical results follow simulations closely, which vali-dates our analytical model.

6.2 Number of DCF StationsIn the experiment of the previous section, the number ofEDCA stations has been taken equal to the number of DCFstations. In order to evaluate the performance of DACKS inscenarios with different numbers of EDCA and DCFstations, we performed the following experiment: We fixedthe number of EDCA stations (Nedca) to 5 (low load), 10(medium load), and 15 (high load) stations, and varied thenumber of DCF stations (Ndcf ) from 2 to 20. The resultingthroughputs for EDCA and DCF stations (the latter in asubplot), obtained analytically and via simulation, are givenin Fig. 6. Results confirm the effectiveness of DACKS undera variable number of DCF stations.

6.3 Total ThroughputIn addition to providing throughput guarantees, one of ourgoals is also to optimize the overall throughput performance.In order to assess the performance of the CWi configurationproposed in Section 4.4, we compared the total throughputobtained with our CWi setting against the result ofperforming an exhaustive search over CWi and choosingthe best configuration. Specifically, in the exhaustive search,we evaluated all possible CWi values, choosing for each onethe largest Pack that ensured the desired throughputguarantees, and took theCWi value that provided the largest

total throughput. The results of this experiment are given inTable 1 as a function ofNedca % Ndcf % N . We can see that thetotal performance achieved by our configuration followsclosely the one resulting from the exhaustive search. Basedon these results, we conclude that our scheme is effective inoptimizing the overall throughput performance.

6.4 WLAN without DACKSIn order to assess the benefits gained from DACKS, wecompared its performance against aWLANwithout DACKSconfigured according to the two following strategies:

. Standard configuration: EDCA stations are configuredwith the CWi setting recommended by the standard[2] for voice traffic, which is the one that gives thehighest priority to EDCA over DCF.

. Optimal configuration: For eachN value, we configureEDCA stations with the CWi setting that maximizestheir throughput, which leads to maximizing theadmissibility region.

Results on the total throughput and the throughput ofEDCA and DCF stations are given in Fig. 7. We firstobserve that DACKS outperforms the strategies withoutDACKS in terms of total throughput. Looking at the perstation throughputs, we see that the three approaches givesimilar throughput to EDCA stations, while DACKSprovides a substantial larger throughput to DCF stations.The reasons for this improvement are further analyzed inthe next experiment.

We further observe that DACKS allows admitting moreEDCA stations while meeting the throughput guarantees.Indeed, up to 16 stations can be admittedwithDACKS,whileonly 13 and 9 stations can be admitted with the optimal andstandard configurations, respectively. We conclude that


Fig. 5. Throughput guarantees. Fig. 6. Number of DCF stations.

TABLE 1Total Throughput (in megabits per second) for theProposed Algorithm and the Exhaustive Search

DACKS benefits both DCF stations (by providing them withmore throughput) and EDCA stations (by increasing thenumber of stations that can be admitted).

6.5 Collision RateThe reason for the performance improvement achieved withDACKS is that, although DACKS wastes some time in theretransmission of successful frameswhoseAcks are skipped,a WLAN without DACKS wastes much more time incollisions. Indeed, in a WLAN without DACKS, the aggres-siveness of DCF stations cannot be controlled, and as aconsequence, EDCA stations need to behave aggressively aswell, which results in many collisions. In order to illustratethis behavior, Fig. 8 shows the collision rate with DACKS forthe same scenario as theprevious experiment andcompares itagainst the collision rate for the strategies without DACKS.This result confirms that the collision rate with DACKS isindeed much smaller than that with the other approaches.

6.6 StabilityOne of the objectives of the configuration setting computedin Section 5 is to ensure that the system is stable. In order toevaluate the stability of our configuration, we analyzed theevolution of the control signal (Pack) over time and comparedit against a configuration with Kp set to a value 100 timeslarger. Fig. 9 depicts the time plots for our configuration(straight line) and for the configuration with larger Kp

(dotted line) for a scenario withN % 15. We observe from thefigure that with our configuration, Pack oscillates stablyaround the average value, while the configuration withlargerKp shows an unstable behavior with large oscillationsof Pack that go from 0 (where DCF stations are starved) to 1(where DCF stations are uncontrolled). These results confirmthe effectiveness of our configuration to ensure stability.

6.7 Changing ConditionsIn addition to stability, another objective of the configurationsetting computed in Section 5 is to ensure that the systemreacts quickly upon changes. In order to study the system’stransient response to changes, we performed the followingexperiment: Initially, we had the system operating withNedca % Ndcf % 5. At some time instant (t % 200 seconds), weintroduced 10 additional DCF stations in the system(Ndcf % 15). At some later instant (t % 300 seconds), weintroduced 10 further additional EDCA stations (Nedca % 15),which (in contrast to the previous case) triggered thecorresponding configuration update. Fig. 10 depicts the timeplot of the throughput of one EDCA station. As a benchmarkto assess the response of our system, we compare theinstantaneous throughput with DACKS against that of asystem where Pack is immediately changed to a fixed newvalue upon the stations’ arrival. We observe from the figurethat DACKS reacts quickly and smoothly to the changes. Thisand the previous experiments confirm the proposed config-uration setting in terms of stability and response to changes.


Fig. 7. WLAN without DACKS (main plot: total throughput; subplots:throughput per EDCA/DCF station).

Fig. 8. Collision rate.

Fig. 9. Stability.

Fig. 10. Changing conditions.

6.8 Validation of the Transient ModelOne of the main contributions of this paper is the transientanalysis of 802.11 presented in Section 5.3. In order tovalidate the model proposed, we performed the followingexperiment: We had 10 DCF stations in the WLAN, and atslot time 200, five of the stations left. Fig. 11 illustrates theevolution of the total transmission probability in thechannel, Pt, according to our transient model and simula-tions. For the simulations, the total probability is computedby taking into account the backoff stage of each station andthe corresponding transmission probability at this backoffstage as given by (3). We observe that simulation resultsfollow our model; although there is a large degree ofvariability in the simulations, caused by the inherentrandomness of Pt, the results given by our model fallwithin the confidence intervals of simulation results, whichconfirms the validity of the model.

6.9 Inactive StationsOne of the design goals of the proposedDACKS scheme is itsability to dynamically adapt to the number of activeDCF andEDCA stations. Specifically, the proposed scheme automa-tically adjusts Pack to the traffic actually transmitted in theWLAN, in order to avoid degrading unnecessarily thethroughput experienced byDCF stations. In order to evaluatethis feature,we performed the following experiment:WehadtheWLAN configured to supportNedca % Ndcf % 16 stations,with a throughput request of 300 Kbps for each EDCAstation, and had that only Nactive of the EDCA and DCFstations were active. To understand the benefit of adjustingPack dynamically, we compared DACKS against a staticconfiguration, where Pack was computed in order to providethe desired throughput guarantees with Nedca % Ndcf % 16.Fig. 12 illustrates the throughput of a DCF station resultingfrom this experiment. We observe that DACKS achieves theobjective of minimizing the disruption suffered by the DCFstations while the static configuration severely degrades theDCF throughput. We conclude that the proposed adaptiveDACKS approach outperforms very significantly the staticapproach proposed in [4].

6.10 Multiple ACsThe experiments performed so far involve one single EDCAAccess Category with throughput guarantees. To gaininsight into the performance of DACKS with more than

one AC, we conducted the following experiment: We hadfour ACs, with throughput guarantees of 300, 150, 75, and37.5 Kbps to AC 1, AC 2, AC 3, andAC 4, respectively. Fig. 13illustrates the throughput obtained by the EDCA stations ofthe different ACs. The results confirm the effectiveness ofDACKS under multiple ACs; in particular, the desiredthroughput guarantees are always met for all ACs.

6.11 Nonsaturated TrafficAll previous experiments have been performed with allstations saturated. In order to evaluate DACKS underdifferent traffic conditions, we repeated the experiment ofthe previous section under nonsaturation. Specifically, weconsidered the following traffic models:

1. EDCA stations of AC 1 and DCF stations weresaturated.

2. EDCA stations of AC 2 generated traffic at aconstant rate.

3. EDCA stations of AC 3 generated traffic following aPoisson process with an average rate equal to itsguaranteed rate.

4. EDCA stations of AC 4 generated traffic following aPareto process of shape 2.

The results obtained, illustrated in Fig. 14, show that ourtechnique is also effective under nonsaturated conditions.


Fig. 11. Validation of the transient model. Fig. 12. Inactive stations.

Fig. 13. Multiple ACs.

In particular, all the ACs see their desired throughputguarantees satisfied independent of their arrival process.

7 SUMMARY AND FINAL REMARKS

The EDCA mechanism of the IEEE 802.11e standard isbackward compatible, thereby allowing legacy DCFstations to interoperate in a WLAN working under theEDCA mechanism. However, the coexistence of EDCA andDCF stations in the same WLAN stations degradesperformance substantially. In particular, the presence ofDCF stations jeopardizes the service guarantees committedto the EDCA stations and degrades the overall efficiencyof the WLAN. The reason for this performance degrada-tion is that DCF stations compete with overly small CWsvalues, and these values cannot be modified since they arepredefined by the standard.

In this paper, we have proposed the DACKS technique toovercome the above problem. With DACKS, upon receivinga frame from a DCF station, the AP skips the Ack reply withsome probability. When missing the Ack reply, DCFstations assume that the transmitted frame collided anddouble their CW . This allows having some control on theaverage CWs used by the DCF stations and therebyovercoming the above problem which was caused by thelack of control on the CWs of the DCF stations.

One of the major challenges with the DACKS scheme isthe configuration of the probability of skipping the Ackreply. This probability should be configured in order topreserve the committed service guarantees to the EDCAstations while minimizing the disruption suffered by theDCF stations. We argue that these goals require the skippingprobability to be dynamically configured. Indeed, if theskipping probability was statically set, we would have tochoose a conservative configuration to avoid failing to meetEDCA service guarantees when all stations are active. As aresult, when some of the stations were inactive, the skippingprobability would be too high and DCF stations would seetheir throughput performance unnecessarily reduced.

The system proposed to dynamically tune the skippingprobability is based on the observation that, as long as theoverall transmission probability in the WLAN does notexceed a certain threshold, EDCA stations are guaranteed toreceive the committed service. Following this observation,the controller used by our system takes as input the observed

transmission probability and provides as output the skip-ping probability. The algorithm that we have chosen in thispaper to compute the output control signal based on themeasured input is based on a Proportional Controller.

One of the challenges of our DACKS system is theconfiguration of the gain of the proportional controller. Thishas been addressed in the paper in the following two steps:In the first step, we have conducted a performance analysisof our system under stationary conditions to obtain themaximum transmission probability in the WLAN thatguarantees EDCA stations receive the committed through-puts, which has been used as the reference signal of theDACKS controller. In the second step, we have conductedan analysis of our system under transient conditions. Basedon this analysis, we have studied our system from a control-theoretic standpoint and found the conditions that need tobe met in order to guarantee that our system is stable.Following considerations from control theory, we have thenset the gain of the Proportional Controller as a trade-offbetween stability and speed of reaction.

The proposed scheme has been exhaustively evaluatedby means of simulations. The performance evaluationconducted has shown the following:

1. DACKS is effective in providing throughput guar-antees to EDCA stations.

2. The chosen configuration maximizes the overallefficiency.

3. A WLAN with DACKS is more efficient than aWLAN that does not use the DACKS technique.

4. Our technique avoids disrupting DCF stations in casesome of the (EDCA or DCF) stations are not active.

5. Our closed-loop system behaves stably while react-ing quickly upon changing conditions.

Although the focus of this paper has been on providingEDCA stations with throughput guarantees, the proposedscheme can also be used to provide delay guarantees. Indeed,the key idea of DACKS is to regulate the DCF stations toensure that the transmission probability in the channel doesnot exceed a given value. Following this, the value of thetransmission probability that ensures the desired delayguarantees can be computed based on the model of [31],and then DACKS can be used to provide these guarantees.

ACKNOWLEDGMENTS

The authors thank Dr. Jose Felix Kukielka for havingcarefully read the manuscript. They are grateful to theanonymous referees for their valuable comments, whichgreatly helped in improving the paper. The research leadingto these results received funding from the EuropeanCommunity’s Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 214994 (CARMENproject). It was also partly funded by the Ministry ofScience and Innovation of Spain, under the QUARTETproject (TIN2009-13992-C02-01).

REFERENCES

[1] IEEE 802.11 WG, Information Technology—Telecomm. and Informa-tion Exchange between Systems. Local and Metropolitan Area Networks.Specific Requirements. Part 11: Wireless LAN Medium Access Control(MAC) and Physical Layer (PHY) Specifications, IEEE, Aug. 1999.


Fig. 14. Nonsaturated traffic.

[2] IEEE 802.11 WG, Amendment to Standard for Information Technology.LAN/MAN Specific Requirements—Part 11: Wireless LAN MediumAccess Control (MAC) and Physical Layer (PHY) Specifications:Medium Access Control (MAC) Enhancements for Quality of Service(QoS), Supplement to IEEE 802.11 Standard, IEEE, Nov. 2005.

[3] IEEE 802.11 WG, Information Technology—Telecomm. and Informa-tion Exchange between Systems. Local and Metropolitan Area Networks.Specific Requirements. Part 11: Wireless LAN Medium Access Control(MAC) and Physical Layer (PHY) Specifications, IEEE 802.11-REVma/D9.0, Revision of Standard 802.11-1999, IEEE, 2006.

[4] L. Vollero, A. Banchs, and G. Iannello, “ACKS: A Technique toReduce the Impact of Legacy Stations in 802.11e EDCA WLANs,”IEEE Comm. Letters, vol. 9, no. 4, pp. 346-348, Apr. 2005.

[5] A. Banchs, P. Serrano, and L. Vollero, “Reducing the Impact ofLegacy Stations on Voice Traffic in 802.11e EDCA WLANs,” IEEEComm. Letters, vol. 11, no. 4, pp. 331-333, Apr. 2007.

[6] G.-H. Hwang and D.-H. Cho, “Performance Analysis on Coex-istence of EDCA and Legacy DCF Stations in IEEE 802.11 WirelessLANs,” IEEE Trans. Wireless Comm., vol. 5, no. 12, pp. 3355-3359,Dec. 2006.

[7] G. Bianchi, I. Tinnirello, and L. Scalia, “Understanding 802.11eContention-Based Prioritization Mechanisms and Their Coexis-tence with Legacy 802.11 Stations,” IEEE Network, vol. 19, no. 4,pp. 28-34, Aug. 2005.

[8] P.E. Engelstad and O.N. Osterbo, “Analysis of the Total Delay ofIEEE 802.11e EDCA and 802.11 DCF,” Proc. IEEE Int’l Conf. Comm.(ICC ’06), June 2006.

[9] J. Villalon, P. Cuenca, and L. Orozco-Barbosa, “On the Capabilitiesof IEEE 802.11e for Multimedia Communications over Hetero-geneous 802.11/802.11e WLANs,” Telecomm. Systems, vol. 36,nos. 1-3, pp. 27-38, Nov. 2007.

[10] J. Majkowski and F.C. Palacio, “QoS Protection for IEEE 802.11e inWLAN with Shared EDCA and DCF Access,” Proc. Conf. Comm.Systems and Networks (CSN), Aug. 2006.

[11] H. Al-Mefleh and J.M. Chang, “A New ACK Policy to Mitigate theEffects of Coexisting IEEE 802.11/802.11e Devices,” Proc. IEEEINFOCOM, Apr. 2008.

[12] J. Villalon, P. Cuenca, L. Orozco-Barbosa, and A. Garrido,“B-EDCA: A QoS Mechanism for Multimedia Communicationsover Heterogeneous 802.11/802.11e WLANs,” Computer Comm.,vol. 31, no. 17, pp. 3905-3921, Nov. 2008.

[13] L. Vollero and G. Iannello, “ACK Skipping: Enabling QoS forMultimedia Communications in WiFi Hot Spots,” Int’l J. High-Performance Computing and Networking, vol. 4, nos. 1/2, pp. 23-30,July 2006.

[14] A. Banchs, P. Serrano, and L. Vollero, “Providing ServiceGuarantees in 802.11e EDCA WLANs with Legacy Stations,”technical report, Univ. Carlos III of Madrid, http://www.it.uc3m.es/banchs/papers/dynamic_acks.pdf, 2010.

[15] G. Bianchi, “Performance Analysis of the IEEE 802.11 DistributedCoordination Function,” IEEE J. Selected Areas Comm., vol. 18,no. 3. pp. 535-547, Mar. 2000.

[16] A. Banchs and L. Vollero, “Throughput Analysis and OptimalConfiguration of 802.11e EDCA,” Computer Networks, vol. 50,no. 11, pp. 1749-1768, Aug. 2006.

[17] F. Cali, M. Conti, and E. Gregori, “Dynamic Tuning of the IEEE802.11 Protocol to Achieve a Theoretical Throughput Limit,” IEEE/ACM Trans. Networking, vol. 8, no. 6, pp. 785-799, Dec. 2000.

[18] V. Ramaiyan, A. Kumar, and E. Altman, “Fixed Point Analysis ofSingle Cell IEEE 802.11e WLANs: Uniqueness and Multistability,”IEEE/ACM Trans. Networking, to appear.

[19] A. Banchs, X. Perez-Costa, and D. Qiao, “Providing ThroughputGuarantees in IEEE 802.11e Wireless LANs,” Proc. 18th Int’lTeletraffic Congress (ITC-18), Sept. 2003.

[20] B. Kristiansson and B. Lennartson, “Robust Tuning of PI and PIDControllers,” IEEE Control Systems Magazine, vol. 26, no. 1, pp. 55-69, Feb. 2006.

[21] G.F. Franklin, J.D. Powell, and M.L. Workman, Digital Control ofDynamic Systems, second ed. Addison-Wesley, 1990.

[22] A.K. Palit and D. Popovic, Computational Intelligence in Time SeriesForecasting: Theory and Engineering Applications. Springer-Verlag,2005.

[23] H. Chen and A.M. Haimovich, “Iterative Estimation andCancellation of Clipping Noise for OFDM Signals,” IEEE Comm.Letters, vol. 7, no. 7, pp. 305-307, July 2003.

[24] F. Cali, M. Conti, and E. Gregori, “IEEE 802.11 Protocol: Design andPerformance Evaluation of anAdaptive BackoffMechanism,” IEEEJ. Selected Areas Comm., vol. 18, no. 9, pp. 1774-1786, Sept. 2000.

[25] R.K. Challa, S. Chakrabarti, and D. Datta, “Modeling of IEEE802.11 DCF for Transient State Conditions,” J. Networks, vol. 2,no. 4, pp. 14-19, Aug. 2007.

[26] C.H. Foh and M. Zukerman, “Performance Evaluation of IEEE802.11,” Proc. IEEE Vehicular Technology Conf. (VTC), May 2001.

[27] C.V. Hollot, V. Misra, D. Towsley, and W.-B. Gong, “A ControlTheoretic Analysis of RED,” Proc. IEEE INFOCOM, Apr. 2001.

[28] K. Astrom and B. Wittenmark, Computer-Controlled Systems, Theoryand Design, second ed. Prentice Hall, 1990.

[29] P. Serrano, A. Banchs, and A. Azcorra, “A Throughput and DelayModel for IEEE 802.11e EDCA under Non Saturation,” WirelessPersonal Comm., vol. 43, no. 2, pp. 467-479, Oct. 2007.

[30] IEEE 802.11 WG, Information Technology—Telecomm. and Informa-tion Exchange between Systems. Local and Metropolitan Area Networks.Specific Requirements. Part 11: Wireless LAN Medium Access Control(MAC) and Physical Layer (PHY) Specifications: High-Speed PhysicalLayer Extension in the 2.4 GHz Band, Supplement to IEEE 802.11Standard, IEEE, Sept. 1999.

[31] A. Banchs and L. Vollero, “A Delay Model for IEEE 802.11eEDCA,” IEEE Comm. Letters, vol. 9, no. 6, pp. 508-510, June 2005.

Albert Banchs received the telecommunica-tions engineering degree from the PolytechnicalUniversity of Catalonia in 1997 and the PhDdegree from the same university in 2002. Since2003, he has been with the University Carlos IIIof Madrid. His PhD dissertation received thenational award for the best thesis on broadbandnetworks. He was a visiting researcher at theInternational Computer Science Institute, Berke-ley, in 1997, worked for Telefonica I+D in 1998,

and worked for NEC Europe Ltd., Germany, from 1998 to 2003. He is theauthor of more than 50 publications in peer-reviewed journals andconferences and has four patents (two of them granted). He is anassociate editor for IEEE Communications Letters and has been theguest editor for IEEE Wireless Communications and for ComputerNetworks. He has served on the program committees of a number ofconferences and workshops including IEEE INFOCOM, IEEE ICC, andIEEE GLOBECOM, and is the program committee chair for EuropeanWireless 2010. He is a member of the IEEE.

Pablo Serrano received the telecommunicationengineering degree and the PhD degree fromthe Universidad Carlos III de Madrid (UC3M) in2002 and 2006, respectively. He has been in theTelematics Department of UC3M since 2002,where he is currently an assistant professor. In2007, he was a visiting researcher in theComputer Network Research Group at theUniversity of Massachusetts Amherst, partiallysupported by the Spanish Ministry of Education

under a Jose Castillejo grant. His current research focuses on theperformance evaluation of wireless networks. He has more than20 scientific papers in peer-reviewed international journals andconferences. He also serves as a program committee member ofseveral international conferences, including IEEE GLOBECOM andIEEE INFOCOM. He is a member of the IEEE.

Luca Vollero received the MS degree intelecommunications engineering and the PhDdegree in computer science from the Universityof Naples “Federico II” in 2001 and 2005,respectively. He is an assistant professor in theLaboratory on Elaboration Systems and Bioin-formatics at the University Campus Bio-Medicoof Rome. His research interests include net-works modeling and simulations, design andevaluation of systems for mobility in heteroge-

neous networks, multimedia data elaboration, and image processing. Heis a member of the IEEE, the IEEE Computer Society, the IEEECommunications Society, and the ACM.


Providing Service Guarantees in 802.11e EDCA WLANs with Legacy Stations

Documents

Transcript of Providing Service Guarantees in 802.11e EDCA WLANs with Legacy Stations