USF Libraries
USF Digital Collections

A measurement-based admission control mechanism for wireless local area networks

MISSING IMAGE

Material Information

Title:
A measurement-based admission control mechanism for wireless local area networks
Physical Description:
Book
Language:
English
Creator:
Babu, Srinivas Bandi Ramesh
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla
Publication Date:

Subjects

Subjects / Keywords:
Quality of service
Jitter
Starvation
IEEE 802.11 DCF
IEEE 802.11e EDCA
Dissertations, Academic -- Computer Science -- Masters -- USF
Genre:
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Abstract:
ABSTRACT: As users become more comfortable using IEEE 802.11 Wireless Local Area Networks, the need for quality of service is becoming more important because of the lack of support in current standards and the increase of multimedia traffic over the Internet. The IEEE 802.11 working group has recognized this fact proposing the Enhanced Distributed Channel Access (EDCA), a priority-based distributed scheme meant to provide service differentiation. EDCA relies on either different Arbitrary Interframe Space (AIFS), or Contention Window (CW) parameters, or both to provide service differentiation. In this thesis, a performance evaluation of the EDCA using five different combinations of the above mentioned parameters is included and compared to the current DCF (Distributed Coordination Function) standard, which is used as the base case. Simulation results show that simpler schemes based on one parameter alone can provide good average service differentiation. However, only multiparameter schemes provide the average and instantaneous high throughput and low delay values needed to support streaming applications. Starvation is a problem spanning all these schemes. It is especially more pronounced in schemes using combinations of parameters. In this thesis, a measurement-based admission control mechanism is proposed to overcome the above stated problems. The admission control mechanism uses an algorithm that admits a flow depending on the jitter values for high priority traffic and the throughput of the low priority traffic. It also allows the administrator to set the bandwidth sharing policy between the high priority traffic and low priority traffic. Results show that the admission control mechanism not only protects existing high priority flows from jitter and low priority flows from starvation, but also improves upon the network utilization.
Thesis:
Thesis (M.S.)--University of South Florida, 2005.
Bibliography:
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
Statement of Responsibility:
by Srinivas Bandi Ramesh Babu.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 65 pages.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001913757
oclc - 174266324
usfldc doi - E14-SFE0001403
usfldc handle - e14.1403
System ID:
SFS0025723:00001


This item is only available as the following downloads:


Full Text
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001913757
003 fts
005 20071016132450.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 071016s2005 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001403
040
FHM
c FHM
035
(OCoLC)174266324
049
FHMM
090
QA76 (ONLINE)
1 100
Babu, Srinivas Bandi Ramesh.
2 245
A measurement-based admission control mechanism for wireless local area networks
h [electronic resource] /
by Srinivas Bandi Ramesh Babu.
260
[Tampa, Fla] :
b University of South Florida,
2005.
3 520
ABSTRACT: As users become more comfortable using IEEE 802.11 Wireless Local Area Networks, the need for quality of service is becoming more important because of the lack of support in current standards and the increase of multimedia traffic over the Internet. The IEEE 802.11 working group has recognized this fact proposing the Enhanced Distributed Channel Access (EDCA), a priority-based distributed scheme meant to provide service differentiation. EDCA relies on either different Arbitrary Interframe Space (AIFS), or Contention Window (CW) parameters, or both to provide service differentiation. In this thesis, a performance evaluation of the EDCA using five different combinations of the above mentioned parameters is included and compared to the current DCF (Distributed Coordination Function) standard, which is used as the base case. Simulation results show that simpler schemes based on one parameter alone can provide good average service differentiation. However, only multiparameter schemes provide the average and instantaneous high throughput and low delay values needed to support streaming applications. Starvation is a problem spanning all these schemes. It is especially more pronounced in schemes using combinations of parameters. In this thesis, a measurement-based admission control mechanism is proposed to overcome the above stated problems. The admission control mechanism uses an algorithm that admits a flow depending on the jitter values for high priority traffic and the throughput of the low priority traffic. It also allows the administrator to set the bandwidth sharing policy between the high priority traffic and low priority traffic. Results show that the admission control mechanism not only protects existing high priority flows from jitter and low priority flows from starvation, but also improves upon the network utilization.
502
Thesis (M.S.)--University of South Florida, 2005.
504
Includes bibliographical references.
516
Text (Electronic thesis) in PDF format.
538
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
500
Title from PDF of title page.
Document formatted into pages; contains 65 pages.
590
Adviser: Miguel Labrador, Ph.D.
653
Quality of service.
Jitter.
Starvation.
IEEE 802.11 DCF.
IEEE 802.11e EDCA.
690
Dissertations, Academic
z USF
x Computer Science
Masters.
773
t USF Electronic Theses and Dissertations.
4 0 856
u http://digital.lib.usf.edu/?e14.1403



PAGE 1

A Measurement-Based Admission Control Mechanism for W ireless Local Area Netw orks by Srini v as Bandi Ramesh Bab u A thesis submitted in partial fulllment of the requirements for the de gree of Master of Science in Computer Science and Engineering Department of Computer Science and Engineering Colle ge of Engineering Uni v ersity of South Florida Major Professor: Miguel Labrador Ph.D. K en Christensen, Ph.D. De we y Rundus, Ph.D. Date of Appro v al: No v ember 3, 2005 K e yw ords: quality of service, jitter starv ation, IEEE 802.11 DCF IEEE 802.11e EDCA cCop yright 2005, Srini v as Bandi Ramesh Bab u

PAGE 2

A CKNO WLEDGEMENTS I w ould tak e this opportunity to e xtend my sincere thanks to Dr Miguel Labrador for introducing me to this project and pro viding me an opportunity to w ork on the same. His ideas and constant support throughout the course of this thesis ha v e k ept me moti v ated and made my stay at USF a v ery pleasant and memorable one. I w ould also e xtend my sincere thanks to Dr K en Christensen and Dr De we y Rundus, for being on my committee and for their v aluable comments and suggestions. It w ould ha v e been impossible to complete this thesis without the support e xtended by my f amily and friends. It w ould tak e more than w ords to e xpress my thanks to them for their constant support and moti v ation.

PAGE 3

T ABLE OF CONTENTS LIST OF T ABLES iii LIST OF FIGURES i v ABSTRA CT vi CHAPTER 1 INTR ODUCTION 1 1.1 Background 1 1.2 Quality of Service 3 1.3 QoS in IEEE 802.11 WLANs 4 1.4 Moti v ation 4 1.5 Contrib ution of this Thesis 5 1.6 Or ganization of this Document 5 CHAPTER 2 B A CKGR OUND AND LITERA TURE REVIEW 6 2.1 The IEEE 802.11 Standard 6 2.1.1 The Distrib uted Coordination Function (DCF) 6 2.1.2 The Point Coordination Function (PCF) 8 2.2 Ov ervie w of QoS Mechanisms in WLANs 9 2.2.1 Priority-based QoS Support 9 2.2.2 F air Scheduling Based QoS Support 10 2.3 IEEE 802.11e 11 2.3.1 Enhanced Distrib uted Channel Access (EDCA) 11 2.4 Re vie w of Performance Ev aluation Studies of IEEE 802.11e EDCA 12 2.5 Admission Control in IEEE 802.11e 14 2.5.1 Distrib uted Admission Control (D A C) 15 2.5.2 T w o-Le v el Protection and Guarantee Mechanism for V oice and V ideo 15 2.5.3 V irtual MA C (VMA C) and V irtual Source (VS) Algorithms 16 2.5.4 Threshold-based Admission Control 16 2.5.5 Harmonica 17 2.5.6 Model Based Admission Control 17 2.6 ns-2.26 (Netw ork Simulator) 17 CHAPTER 3 PERFORMANCE EV ALU A TION OF IEEE 802.11E EDCA 18 3.1 Ov ervie w 18 3.1.1 Metrics Used for Performance Ev aluation 19 3.1.2 Classes of T raf c 21 i

PAGE 4

3.2 Implementation in ns-2.26 21 3.2.1 Physical Layer 21 3.2.2 Medium Access Control Layer 22 3.3 Simulation Scenario 22 3.3.1 Communication Model 24 3.4 Results 25 3.4.1 Service Dif ferentiation 25 3.4.2 Service V ariability 29 3.4.3 P ack et Loss Rates and Starv ation 33 3.5 Conclusion 33 CHAPTER 4 MEASUREMENT -B ASED ADMISSION CONTR OL MECHANISM FOR IEEE 802.11E 36 4.1 Problem Statement and Design Considerations 36 4.2 Ov ervie w and Design of the Admission Control Mechanism 37 4.2.1 Metrics Used for Admission Control 39 4.2.2 Algorithm Used for the Admission Control Mechanism 40 4.2.3 P arameters Set by Netw ork Administrator 42 4.2.4 Assumptions 42 4.2.5 Sample Scenario 43 4.3 Implementation in ns-2 45 4.4 Simulation Scenario 45 4.4.1 Communication Model 46 4.5 Results 47 4.5.1 Scenario One 47 4.5.2 Scenario T w o 54 4.6 Conclusion 59 CHAPTER 5 CONCLUSION AND FUTURE W ORK 61 REFERENCES 63 ii

PAGE 5

LIST OF T ABLES T able 3.1 Simulation P arameters Used in the AIFS, CW and PF Schemes 23 T able 3.2 Simulation P arameters Used in the AIFS+CW AIFS+CW+PF and DCF Schemes 23 T able 3.3 Mean and Standard De viation of Delay and Throughput of the AIFS, CW and PF Schemes in High Netw ork Load Conditions 32 T able 3.4 Mean and Standard De viation of Delay and Throughput of the AIFS+CW and AIFS+CW+PF and DCF Schemes in High Netw ork Load Conditions 32 T able 4.1 P arameters Used for the AIFS+CW+PF Scheme Used in the Admission Control Mechanism 46 T able 4.2 V alues for Jitter at T ime Instances when CBR Flo ws are Admitted in the Mechanism with Admission Control (Scenario 1) 48 T able 4.3 P ack et Loss Rates for Scenario 1 54 T able 4.4 V alues for Jitter at T ime Instances when CBR Flo ws are Admitted in the Mechanism with Admission Control (Scenario 2) 55 T able 4.5 P ack et Loss Rates for Scenario 2 59 iii

PAGE 6

LIST OF FIGURES Figure 1.1 W ireless User Accessing Multimedia Services via Internet 2 Figure 1.2 T ypical IEEE 802.11 Netw ork 3 Figure 2.1 IEEE 802.11 Distrib uted Coordination Function (DCF) 7 Figure 2.2 Classication of Distrib uted QoS Mechanisms [10 ] 10 Figure 2.3 IEEE 802.11 Enhanced Distrib uted Channel Access (EDCA) 13 Figure 2.4 Classication of Admission Control Mechanisms for EDCA 14 Figure 3.1 EDCA Simulation Scenario 23 Figure 3.2 DCF Simulation Scenario 24 Figure 3.3 A v erage Delay v ersus Load of AIFS, CW PF AIFS+CW AIFS+CW+PF and DCF Schemes 26 Figure 3.4 A v erage Throughput v ersus Load of AIFS, CW and PF AIFS+CW AIFS+CW+PF and DCF Schemes 27 Figure 3.5 Delay V ariation 30 Figure 3.6 Throughput V ariation 31 Figure 3.7 P ack et Loss Rate v ersus Load of AIFS, CW PF AIFS+CW AIFS+CW+PF and DCF Schemes 34 Figure 4.1 Bandwidth Allocation for High Priority and Lo w Priority T raf c 38 Figure 4.2 Flo wchart for the Admission Control Mechanism 41 Figure 4.3 Sample Scenario 44 Figure 4.4 Simulation Scenario for the Ev aluation of the Admission Control Mechanism 45 Figure 4.5 Instantaneous Delay and Throughput with and without Admission Control for High Priority Class and one V oice Flo w in Scenario 1 50 i v

PAGE 7

Figure 4.6 Probability Density Function for the Delay of High Priority T raf c and one V oice Flo w with and without Admission Control in Scenario 1 51 Figure 4.7 Instantaneous Delay and Throughput with and without Admission Control for Lo w Priority Class and one FTP Flo w in Scenario 1 52 Figure 4.8 Netw ork Utilization with and without Admission Control in Scenario 1 (left to right) 53 Figure 4.9 Instantaneous Delay and Throughput with and without Admission Control for High Priority Class and one V oice Flo w in Scenario 2 56 Figure 4.10 Probability Density Function for the Delay of High Priority T raf c and one V oice Flo w with and without Admission Control in Scenario 2 57 Figure 4.11 Instantaneous Delay and Throughput with and without Admission Control for Lo w Priority Class and one FTP Flo w in Scenario 2 58 Figure 4.12 Netw ork Utilization with and without Admission Control in Scenario 2 (left to right) 59 v

PAGE 8

A MEASUREMENT -B ASED ADMISSION CONTR OL MECHANISM FOR WIRELESS LOCAL AREA NETW ORKS Srini v as Bandi Ramesh Bab u ABSTRA CT As users become more comfortable using IEEE 802.11 W ireless Local Area Netw orks, the need for quality of service is becoming more important because of the lack of support in current standards and the increase of multimedia traf c o v er the Internet. The IEEE 802.11 w orking group has recognized this f act proposing the Enhanced Distrib uted Channel Access (EDCA), a priority-based distrib uted scheme meant to pro vide service dif ferentiation. EDCA relies on either dif ferent Arbitrary Interframe Space (AIFS), or Contention W indo w (CW) parameters, or both to pro vide service dif ferentiation. In this thesis, a performance e v aluation of the EDCA using v e dif ferent combinations of the abo v e mentioned parameters is included and compared to the current DCF (Distrib uted Coordination Function) standard, which is used as the base case. Simulation results sho w that simpler schemes based on one parameter alone can pro vide good a v erage service dif ferentiation. Ho we v er only multiparameter schemes pro vide the a v erage and instantaneous high throughput and lo w delay v alues needed to support streaming applications. Starv ation is a problem spanning all these schemes. It is especially more pronounced in schemes using combinations of parameters. In this thesis, a measurement-based admission control mechanism is proposed to o v ercome the abo v e stated problems. The admission control mechanism uses an algorithm that admits a o w depending on the jitter v alues for high priority traf c and the throughput of the lo w priority traf c. It also allo ws the administrator to set the bandwidth sharing polic y between the high priority traf c and lo w priority traf c. Results sho w that the admission control mechanism not only protects e xisting vi

PAGE 9

high priority o ws from jitter and lo w priority o ws from starv ation, b ut also impro v es upon the netw ork utilization. vii

PAGE 10

CHAPTER 1 INTR ODUCTION W ireless netw orks ha v e e xperienced an e xponential gro wth during the last fe w years. The adv ent of ine xpensi v e mobile computers with po werful CPUs and wireless netw orking equipment, wide deplo yment of standardized protocols, the necessity to o v ercome geographical constraints and the need to ha v e ubiquitous access to the Internet together ha v e propelled the gro wth of wireless netw orks. W ireless Local Area Netw orks (WLANs) with Internet access are being used in hospitals, schools, cof fee shops, museums, hotels, homes, airports, etc. Users are utilizing these last hop wireless access technologies as if the y were connected to wired local area netw orks, conducting b usiness as usual. Concurrently the use of the Internet for multimedia applications lik e streaming video and v oice has also increased tremendously Since these applications ha v e such high demands and v arying requirements, it is necessary that the netw ork guarantee service requirements lik e delay throughput and jitter [1 ]. In order to achie v e these, the netw ork should ha v e support for Quality of Service (QoS). QoS has achie v ed reasonable progress in wired netw orks, b ut currently there is little or no support for it in WLANs. It is v ery important that these multimedia applications ha v e end-toend QoS support. Users accessing these multimedia services via the Internet using WLANs ha v e to recei v e them through a combination of wired and wireless netw orks (Figure1.1). The lack of support for QoS in the wireless netw orks does not allo w for this. 1.1 Backgr ound The Institute of Electrical and Electronics Engineers (IEEE) started a project in 1990 to come up with an international WLAN standard identied as IEEE 802.11 [2 ]. The moti v ational f actors behind the project are listed belo w [2 ]: 1

PAGE 11

Figure 1.1 W ireless User Accessing Multimedia Services via InternetThe installation and maintenance costs of making changes in e xisting wired LAN infrastructures w as pro ving to be high. WLANs could pro vide a cheap alternati v e.It w as v ery dif cult to t old b uilding with traditional wired LAN infrastructures. WLANs w ould be v ery easy to install.W ired LANs are impractical for netw orks that are needed on a v ery short term basis. WLANs could be the perfect alternati v e. The standard intended ”to de v elop a Medium Access Control (MA C) and Physical Layer (PHY) specication for wireless connecti vity for x ed, portable and mo ving stations within a local area” [2 ]. The standard denes tw o dif ferent MA C schemes namely Point Coordination Function (PCF) and Distrib uted Coordination Function (DCF) for transporting time-bounded and asynchronous ser vices respecti v ely The DCF w as designed to be f air to all users and gi v es them an equal chance of accessing the netw ork. The PCF w as primarily designed for the transmission of delay-sensiti v e traf c and access to medium is controlled by a central authority kno wn as the Access Point (AP). A typical e xample of an IEEE 802.11 netw ork is sho wn in Figure 1.2. 2

PAGE 12

Figure 1.2 T ypical IEEE 802.11 Netw ork 1.2 Quality of Ser vice According to [3 ], Quality of Service (QoS) can be dened as follo ws: A measur e of performance for a tr ansmission system that r eects its tr ansmission quality and availability of service If a gi v en netw ork can guarantee the requirements specied by a particular application, it can be said that the netw ork pro vides support for QoS. Some of the requirements might include throughput, end-to-end delay and v ariability in delay (jitter). A prerequisite for QoS is service dif fer entiation Service Dif ferentiation is the ability of a netw ork to identify the dif ferent types of traf c and prioritize them accordingly Ho we v er service dif ferentiation does not mean QoS, since the netw ork can dif ferentiate between applications b ut may still not satisfy the specied QoS requirements. In this thesis, the focus is on pro viding service dif ferentiation in a distrib uted manner while looking at important metrics for multimedia applications. 3

PAGE 13

1.3 QoS in IEEE 802.11 WLANs There is limited or no support for QoS or service dif ferentiation in the IEEE 802.11 standard. DCF has limitations in terms of QoS or service dif ferentiation because it pro vides only best-ef fort service, i.e. it has no dif ferentiation mechanism to pro vide better service to multimedia applications compared to data applications [4 ].: PCF w as designed k eeping in mind time-bounded applications and w as supposed to pro vide support for QoS. But, it too has the follo wing limitations [4]:PCF cannot handle QoS requirements for dif ferent types of traf c.The access point which polls the stations itself has to contend for the channel.Once PCF gi v es permission to a station to transmit, it is dif cult to control the amount of time the station will transmit. These are the main reasons why the PCF specication w as rarely implemented in real products. A common problem for both PCF and DCF is that neither of them ha v e an y kind of Admission Control [4]. In order to o v ercome these limitations, the IEEE is coming up with a ne w standard named IEEE 802.11e. This standard, which is described in Chapter 2, is meant to pro vide support for service dif ferentiation using a priority-based mechanism. A number of modications ha v e been suggested to support QoS. These modications will be re vie wed in Chapter 2 as well. 1.4 Moti v ation A lot of research has been done on e v aluating IEEE 802.11e WLANs. Most of the studies ha v e focused on a v erage and not on instantaneous throughput and delay metrics, specically important metrics for multimedia applications. One important mechanism utilized man y times in the past to pro vide QoS guarantees in wired netw orks is Admission Control. Admission Control algorithms ha v e also been in v estigated in 4

PAGE 14

WLANs before, ho we v er the y ha v e not been able to trade of f netw ork utilization for jitter for high priority traf c and starv ation for lo w priority traf c in an ef cient manner 1.5 Contrib ution of this Thesis This thesis includes a thorough performance e v aluation of the IEEE 802.11e proposed standard and proposes an Admission Control mechanism for it. The most important contrib utions of this thesis are:Dif ferent priority based schemes possible with IEEE 802.11e EDCA are e v aluated and compared with le gac y IEEE 802.11 DCF .Starv ation and jitter are identied as the k e y problems in the service dif ferentiation mechanism.A ne w measurement-based admission control mechanism has been proposed that tak es into consideration jitter and starv ation as important QoS metrics. 1.6 Or ganization of this Document This chapter pro vides a brief introduction and e xplains the moti v ation and contrib ution of this thesis. The ne xt chapter re vie ws in brief the background related to this thesis. It also re vie ws the research that has been or is being performed on similar lines. The third chapter tak es a look at the performance e v aluation of the IEEE 802.11e proposed standard and the results pertaining to it. The fourth chapter e xplains the ne w measurement-based admission control mechanism proposed and presents the results of its e v aluation. The fth chapter gi v es a brief conclusion and e xplains the future scope of the project. 5

PAGE 15

CHAPTER 2 B A CKGR OUND AND LITERA TURE REVIEW This chapter describes the IEEE 802.11 and the IEEE 802.11e standards. It re vie ws some wellresearched and published w orks on the performance analysis of the IEEE 802.11e standard and also methods for pro viding Admission Control in the IEEE 802.11e standard. 2.1 The IEEE 802.11 Standard W ireless netw orks using the IEEE 802.11 standard can be congured in tw o w ays: infrastructure mode and ad-hoc mode [4 ]. In the infrastructure mode, all transmissions between the wireless stations go through a central station kno wn as the Access Point (AP). In the ad-hoc mode all stations can communicate with each other without the need for an AP The IEEE 802.11 standard denes specications for the Physical (PHY) and Medium Access Control (MA C) layer for both the infrastructure and the ad-hoc mode to form a WLAN [4 ]. This thesis will be looking only at the MA C specications for the infrastructure mode. The main aim of the IEEE 802.11 MA C layer is to pro vide access control to the wireless medium in an ef cient and f air manner to all the users. The function incorporates functions such as access coordination, addressing, frame check sequence generation and security [4 ]. In order to achie v e access coordination, tw o functions ha v e been dened in the IEEE 802.11 MA C specication, the Distrib uted Coordination Function (DCF) and the Point Coordination Function (PCF). Implementation of DCF is mandatory PCF is an optional function. Due to its comple xity it is not implemented. 2.1.1 The Distrib uted Coordination Function (DCF) In order to re gulate access to the shared wireless medium, the DCF uses Carrier Sense Multiple Access with Collision A v oidance (CSMA/CA) [4 ]. Figure 2.1 illustrates the functioning of DCF 6

PAGE 16

Figure 2.1 IEEE 802.11 Distrib uted Coordination Function (DCF) using CSMA/CA. This mechanism primarily uses tw o dif ferent techniques to re gulate medium access, namely the Interframe Space (IFS) and the back-of f algorithm. In DCF each station checks if the medium is idle. If idle, the station can transmit immediately Else, it will w ait for the current transmission to complete and then will w ait for a further DCF IFS (DIFS) interv al, after which it runs a back-of f mechanism. The back-of f time or Back-of f Interv al (BI) is calculated using the equation belo w [5 ]: BI = r andom() SlotT ime where random () is a pseudorandom inte ger dra wn from a uniform distrib ution o v er the interv al [0,CW]. CW stands for Contention W indo w SlotT ime is dened as the ”length of time that a transmitting station w aits before attempting to retransmit follo wing a collision” [6]. When the rst transmission attempt is made by the station, the v alue of CW is at its minimum, also kno wn as CWmin. W ith each unsuccessful transmission, the v alue for CW is doubled until it reaches the maximum possible v alue for CW kno w as CWmax. CWmin and CWmax are predened. CW is reset to CWmin after e v ery successful transmission. The back of f time is decremented only when the medium is sensed to be idle. If medium is sensed to be b usy then the back of f time v alue is 7

PAGE 17

frozen. When the v alue of the back of f time reaches zero, the station is allo wed to transmit. If by coincidence, the back of f time for tw o stations reaches zero at the same time, then a collision might occur This process is clearly illustrated in Figure 2.1. In order to notify the sender that the pack et has been successfully recei v ed, the recei v er sends an Ackno wledgement frame (A CK). If the sender recei v es the A CK before a pre-dened time-out v alue, then the sender considers the transmission to be successful, else the frame sent is assumed to be lost and is retransmitted. Control pack ets, such as the Request to Send (R TS) and Clear to Send (CTS) pack ets used to a v oid the hidden terminal problem and pro vide reliability use a smaller DIFS called Short IFS (SIFS) to get priority access to the medium. One of the biggest dra wbacks of DCF is that it supports only best-ef fort service. Streaming application lik e real-time audio and video are time-bound and ha v e stringent requirements lik e bandwidth, delay and jitter guarantees. W ith DCF all stations contend and gain access to the channel with an equal priority No le v el of service dif ferentiation is pro vided to mak e sure that streaming applications get more priority than data applications. 2.1.2 The P oint Coordination Function (PCF) In order to o v ercome the inability of DCF to pro vide an y le v el of service dif ferentiation the PCF w as introduced. It can be used only when the WLAN is being used in the infrastructure mode. When a wireless netw ork is set in an infrastructure mode, an Access Point (AP) within the netw ork has to be dened. This AP acts as a point coordinator PCF di vides channel access time into periodic interv als called Beacon Interv als (BI). Each BI is di vided into tw o phases: a contention-free period (CFP) and a contention period (CP). The AP is supposed to maintain a list of all the wireless stations that are supposed to be part of the infrastructure netw ork. The AP polls the stations to gi v e them access to the medium. A wireless station can gain access to the medium and start transmissions only when it has been polled by the AP Though PCF w as de v eloped to support time-bounded applications, it has serious limitations [7 8 9 ]. The limitations are listed as follo ws: 8

PAGE 18

PCF does not support dif ferent le v els of priority that are associated with the v aried QoS requirements of dif ferent types of traf c.PCF cannot control the transmission time of a wireless station in all conditions, thus introducing v ariable transmission time. This in turn creates problems for the AP in pro viding guaranteed delay and jitter performance for the other wireless stations in the netw ork.In order to poll a wireless station, the AP itself has to contend for the channel rst. This might introduce unnecessary delays that may result in being able to pro vide QoS guarantees.PCF lik e DCF has no admission control mechanism. When hea vy traf c loads are introduced, there is considerable de gradation in the performance of both the functions. 2.2 Ov er view of QoS Mechanisms in WLANs T o o v ercome the deciencies of the DCF and the PCF in terms of pro viding support for QoS, a number a mechanisms ha v e been suggested. This document re vie ws only the mechanisms designed for DCF i.e. distrib uted mechanisms. According to [10 ], distrib uted QoS mechanisms for WLANs can be classied as in Figure 2.2. 2.2.1 Priority-based QoS Support One method of pro viding support for priority-based QoS is to use the Inter Frame Space (IFS). Either ne w IFS v alues can be used or e xisting ones can be used. It is suggested in [11 ] that e xisting IFS v alues, i.e. DIFS and PIFS be used for prioritization. PIFS is used for high priority traf c and DIFS for lo w priority traf c. Under high load conditions, the mechanism sho ws increase in a v erage access delay and also suf fers from high pack et losses. But, the proposed mechanism can meet the QoS requirements of v oice and video e v en at v ery high netw ork loads. [12 13 ] suggest ne w IFS v alues named Arbitrary IFS (AIFS) be used for prioritization. The v alues for AIFS are to be greater than that of DIFS. This could create problems because e xisting DCF-a w are frames w ould get more priority than QoS-a w are high priority frames. [14 ] also proposed a similar idea to dif ferentiate between priority classes. 9

PAGE 19

Figure 2.2 Classication of Distrib uted QoS Mechanisms [10 ] Another priority-based method of pro viding QoS is based on using the Contention W indo w (CW), i.e. mak e changes to the back of f algorithm. [12 ] and [13 ] ha v e suggested w ays of modifying the minimum and maximum v alues for CW i.e. CWmin and CWmax such that the v alues of CWmin and CWmax for a lo w priority class are al w ays greater than the corresponding v alues for a high priority class. This w ould ensure that the Back of f Interv al (BI) for the high priority class is less than that of the the lo w priority class, i.e. a high priority frame w ould get f aster access to the medium when compared to a lo w priority frame. [15 ] and [16 ] ha v e suggested similar methods. Simulation results in [15 ] sho w v ery clear service dif ferentiation for the high-priority traf c, while that in [16 ] sho w reduced end-to-end pack et delays. The CW o v erlaps for the lo w priority and high priority classes in all the cases mentioned abo v e. 2.2.2 F air Scheduling Based QoS Support In opposition to priority-based mechanisms, where the channel access is bound by the priority of traf c classes, f air queuing mechanisms ha v e been suggested, which di vide the bandwidth of the medium f airly in a suggested ratio. [17 ] suggests partitioning netw ork resources among all the 10

PAGE 20

o ws in a pre-suggested ratio by re gulating the w ait time so that e v ery o w gets a f air opportunity to access the medium. QoS support using f air scheduling can be achie ving by modifying the back of f algorithm. [18 ] and [19 ] suggest one such method called Distrib uted W eighted F air Queuing (D WFQ). Another approach using the back of f algorithm called the Distrib uted F air Scheduling (DFS) is proposed in [20 ]. A mechanism that uses the IFS for f air scheduling has been proposed in [21 ] and [22 ]. This mechanism is named the Distrib uted Decit Round Robin (DDRR). F air scheduling mechanisms are out of the scope of this thesis. 2.3 IEEE 802.11e T o o v ercome the deciencies of the DCF and the PCF in terms of pro viding support for QoS, a ne w standard IEEE 802.11e is being proposed. The IEEE 802.11e draft introduces tw o ne w modes of operation [23 ]: an e xtension of the le gac y DCF kno wn as the EDCA (Enhanced Distrib uted Channel Access) and the Hybrid Coordination Function (HCF) Controlled Channel Access (HCCA) dened for infrastructure based wireless netw orks. This document focuses solely on the EDCA mechanism. 2.3.1 Enhanced Distrib uted Channel Access (EDCA) In order to introduce support for service dif ferentiation, EDCA denes four dif ferent transport modes kno wn as Access Cate gories (A C) [23 ]. Each A C beha v es lik e a virtual wireless station independently contending for access to the medium. Each A C will ha v e its o wn queue and MA C parameters. The logic behind this is essentially being able to pro vide quick er access to the medium to higher priority traf c. A higher priority A C will ha v e a smaller CW v alue, so that it gets access to the medium f aster than a lo wer priority A C. A CWmin and CWmax v alue is set for each A C. Therefore, the v alue of CW for a particular A C lies between these ranges. The CW v alues for each class may or may not o v erlap. 11

PAGE 21

Another w ay to pro vide service dif ferentiation is to use dif ferent DIFSs for each service class, lik e control pack ets in the re gular DCF scheme. F or this, an IFS other than DIFS is introduced, which is kno wn as the Arbitration IFS (AIFS). The v alue of AIFS has to be at least DIFS and v aries for each A C. When a transmission ends and the medium becomes idle, each A C will w ait for an AIFS time interv al and then run a Back of f procedure. The Back of f Interv al is dened as a v alue in the interv aln r, where is the CW dened for a particular A C. As such, each A C inside a station beha v es as a virtual station. When conicts arise within the dif ferent A Cs of the same station, the T ransmission Opportunity (TXOP) is gi v en to a higher priority A C. Lo wer priority A Cs can consider that the y lost their transmitted data due to collisions on the wireless medium and consequently increase their CW Of course, the CW of the A Cs may be the same for all classes, or it may be dif ferent. A third possible method to pro vide service dif ferentiation is a simple modication to the normal DCF scheme in which a simple multiplier is included in the equation that calculates the BI interv al, as sho wn in the equation belo w: r! #"$&%' )(*%,+.-0/12n3%'45 67%891:<;>=?This multiplier is called the Priority F actor (PF). So for e xample,? @Acan use a Priority F actor (PF) equal to one,rBPF=2, and so on, so thatrB @Aon a v erage w aits less time thanrB. The parameters, CW AIFS and PF can also be used in combinations. F or e xample, dif ferent AIFS and CW v alues for each class, so that higher priority classes ha v e smaller IFS and also CW interv als to choose the random number from. This combination mak es the service dif ferentiation more pronounced or stronger between classes. Finally the PF can also be used in combination with the AIFS and CW to emphasize the service dif ferentiation e v en further All the abo v e mechanisms ha v e been illustrated in Figure 2.3. 2.4 Re view of P erf ormance Ev aluation Studies of IEEE 802.11e EDCA A number of studies ha v e e v aluated the performance of the IEEE 802.11 EDCA. In [7], the ne w EDCA is compared with the le gac y 802.11 as re gards to service dif ferentiation. A scenario 12

PAGE 22

Figure 2.3 IEEE 802.11 Enhanced Distrib uted Channel Access (EDCA) in v olving four v oice stations, tw o video stations and four data stations w as utilized. Calculating the aggre gated throughput of each traf c type for both protocols, it w as concluded that unlik e le gac y 802.11, EDCA can pro vide dif ferentiated channel access. In [24 ], the authors include a perfor mance e v aluation of EDCA and Elimination-Y ield Non-Preempti v e Medium Access (EY -NPMA). The y performed simulations using a custom tool and concluded that EY -NPMA performed better than EDCA in terms of mean pack et delay and medium utilization. Another performance e v aluation of EDCA is included in [25 ] using a multi-dimensional Mark o v model. The metrics considered for the e v aluation are saturation throughput, throughput ratios and access delays of o ws of distinct priorities under the R TS/CTS mode. The y also analyze the impact of AIFS and CW on prior itized traf c and conclude that EDCA pro vides signicant adv antage to high priority traf c. In [5 ], a performance e v aluation of the le gac y 802.11 with HCF (Hybrid Coordination Function) and EDCA is included utilizing a scenario of 802.11b/e access to an IP core netw ork through an Access Point in an infrastructure WLAN. Considering the a v erage pack et delay and maximum pack et delay the y conclude that EDCA and HCF combined signicantly impro v e QoS support in WLANs. Finally in [10 ] a comparison of priority-based and scheduling-based algorithms for QoS in WLANs 13

PAGE 23

Figure 2.4 Classication of Admission Control Mechanisms for EDCA is included. Based on a v erage and instantaneous v alues of throughput and delay the y sho w that pro viding QoS in WLANs is still an open problem. 2.5 Admission Contr ol in IEEE 802.11e As w as the case with DCF it is v ery lik ely that EDCA will pro v e to be the more popular channel access mechanisms in WLANs [26 ]. The reasons for this being:Distrib uted nature of EDCA.Easy to implement. K eeping this in mind, we shall be focusing only on the admission control mechanisms meant for EDCA. Admission control mechanisms for EDCA ha v e been classied as Measurement-based Admission Control and Model-based Admission Control (Figure 2.4). 14

PAGE 24

2.5.1 Distrib uted Admission Contr ol (D A C) The Distrib uted Admission Control (D A C) w as introduced and described in [27 ] and [28 ]. In this measurement-based mechanism, a central controller for the wireless netw ork announces a transmission b udget to each A C during a beacon interv al. T ransmission b udget can be dened as the additional amount of time a v ailable. In order to calculate the transmission b udget for each A C, the central controller measures the amount of time that each A C spends in transmission and then subtracts this v alue from the transmission limit. Also, e v ery station in the wireless netw ork sets a transmission limit for each A C. This v alue is also determined on the basis of the transmission time of the A C during the pre vious beacon period and the set transmission limit. Once an A C uses up its transmission limit, it can no longer introduce ne w o ws. Also, e xisting o ws are restricted from increasing their transmission time. D A C has a fe w limitations listed belo w [26 ]:T ransmission parameters are adjusted at e v ery beacon interv al resulting in v ariable netw ork performance.Scheme w orks well only for lo w to medium traf c loads. 2.5.2 T w o-Le v el Pr otection and Guarantee Mechanism f or V oice and V ideo The tw o-le v el protection scheme is an e xtension of the D A C scheme and has been proposed in [29 ]. The tw o le v els of protection are dened as follo ws:The rst le v el protects e xisting QoS o ws from ne w incoming or e xisting QoS o ws.The second le v el protects e xisting QoS o ws from the lo w priority or best ef fort traf c. The rst protection uses tw o schemes, namely the tried-and-kno wn and early-protection method. In the tried-and-kno wn scheme, a o w is introduced into the netw ork and the performance of the netw ork and the o w are considered o v er the ne xt fe w beacon interv als. If the performance does not meet the requirements, then the o w kills or rejects itself. In the early-protection method, a o w is admitted only if all the performance requirements are within pre-specied limits. 15

PAGE 25

Protection from lo w priority or best ef fort traf c can be achie v ed by increasing the contention windo w size and inter -frame space. Since the scheme has been adapted from the D A C, it has the same problems as the D A C. 2.5.3 V irtual MA C (VMA C) and V irtual Sour ce (VS) Algorithms VMA C and VS algorithms ha v e been proposed in [15 ] and [30 ]. The VMA C algorithm runs the applications and MA C processes virtually to decide whether a o w w ould get the required QoS parameters. The VMA C runs in parallel to the real MA C in e v ery station, only that it handles virtual pack ets instead of real ones. The VS is responsible for generating virtual pack ets that are similar to pack ets generated by real applications. The pack ets generated by the VC are passed on the VMA C. The main adv antage of this scheme is that the algorithms do not consume an y e xtra bandwidth. This is some what of fset by the f act that e v ery wireless station has to perform e xtra processing. 2.5.4 Thr eshold-based Admission Contr ol A mechanism suggested in [31 ] measures the traf c condition on the wireless medium. On the basis of these measurements, tw o admission control mechanism ha v e been suggested:Using Relati v e Occupied Bandwidth: Upper and lo wer bandwidth occupanc y thresholds are dened for the medium. If the amount of bandwidth occupied in the medium is less than the lo wer limit, then an A C with the highest priority is admitted. If the amount of occupied bandwidth is between the limits, then no action is tak en. If the amount of occupied bandwidth is more than higher limit, then the lo west acti v e A C is terminated.Using a v erage collision: In this case, instead of the bandwidth occupanc y a ratio called the a v erage collision ratio which is a ratio between the number of collisions to the number of transmissions is used. Thresholds similar to that of the rst case are applied. The actions dened are also the same. 16

PAGE 26

2.5.5 Harmonica The Harmonica scheme has been suggested in [32 ]. In this scheme, the AP parameters related to the medium called as link layer quality indicator parameters (LQI). When a real time application requests for access to the medium, the Harmonica mechanism assigns it to a traf c class based on its requirements and then an admission control mechanism is applied to decide if the application should be gi v en access to the medium. The decision is based on the amount of bandwidth already occupied and the bandwidth requirements of the traf c class that the application has been assigned to. 2.5.6 Model Based Admission Contr ol A Mark o v chain model-based admission control mechanism has been suggested in [33 ]. Admission Control in this scheme is achie v ed by predicting the achie v able throughput for each o w A Contention-W indo ws-base d Admission Control mechanism is suggested in [34 ]. This mechanism changes the v alues of CW dynamically to achie v e admission control. 2.6 ns-2.26 (Netw ork Simulator) The Netw ork Simulator ns-2 has been e xtensi v ely used for the performance e v aluation of IEEE 802.11e EDCA and also for the admission control mechanisms. ns-2 is a discrete e v ent simulator that w as originally de v eloped at the La wrence Berk ele y National Laboratory (LBNL). Extensions from [35 ] ha v e been used to add the functionality of the EDCA to ns-2 ns-2 allo ws for simulation of both wired and wireless netw orks. It is a freely distrib uted, open source tool. It is implemented in C++ and used O Tcl as a command and conguration interf ace. ns-2 implementations for FreeBSD, Linux, Solaris, W indo ws and Mac platforms are a v ailable. F or more details about ns-2 see [36 ]. 17

PAGE 27

CHAPTER 3 PERFORMANCE EV ALU A TION OF IEEE 802.11E EDCA In this chapter we describe the performance e v aluation of the IEEE 802.11e EDCA (Enhanced Distrib uted Channel Access), the metrics and the scenarios used for the e v aluation and the results and conclusions. 3.1 Ov er view The IEEE 802.11e EDCA w as primarily designed to o v ercome the lack of support for QoS in IEEE 802.11 DCF (Distrib uted Coordination Function). As we re vie wed in Chapter 2, a number of studies including [7 24 25 5 10 ] ha v e done a performance e v aluation of the EDCA. But all these studies ha v e primarily only concentrated on the a v erage throughput and delay metrics. This is good enough to sho w that the scheme pro vides good service dif ferentiation, b ut not enough to sho w that the scheme can support streaming applications adequately In order to pro v e that streaming applications can be supported, it is necessary to consider instantaneous performance metrics lik e instantaneous throughput and delay These help in determining the violation of classes as well as the v ariability of the service. These metrics shall be considered in the performance e v aluation. As discussed in Chapter 2, there are three dif ferent parameters in EDCA that can be used to prioritize dif ferent types of traf c, i.e. to dene dif ferent types of Access Cate gories (A C), namely:AIFS (Arbitrary Inter Frame Space)CW (Contention W indo w)Priority F actor (PF) 18

PAGE 28

These parameters can be used singularly or in combination to dene dif ferent A Cs. In order to pro vide a particular le v el of priority to a particular traf c class, it has to be assigned to the corresponding A C. Fi v e dif ferent schemes using these parameters ha v e been e v aluated. These schemes are described belo w:Only AIFS is used to dene dif ferent A Cs. When this scheme is used, the v alues for the other parameters remain the same for all A Cs. Throughout the rest of the document, this scheme shall be referred to as AIFS.Only CW is used to dene dif ferent A Cs. This scheme shall be referred to as CW The v alues for the CW for the dif ferent A Cs do not o v erlap.Only PF is used to dene dif ferent A Cs. This scheme shall be referred to as PF .AIFS and CW are used in combination to dene dif ferent A Cs. The v alues for PF will remain the same for all A Cs when this scheme is used. This scheme shall be referred to as AIFS+CW .All the parameters are used in combination to dene A Cs. This scheme will be referred to as AIFS+CW+PF All the abo v e mentioned schemes will be e v aluated along with DCF which will act as the base case. 3.1.1 Metrics Used f or P erf ormance Ev aluation The metrics that ha v e been used for the performance e v aluation are described belo w:A ver a g e Thr oughput : This is the a v erage amount of information that a particular traf c o w is able to transmit successfully per second for the entire duration of the traf c o w The measurement used to represent a v erage throughput in this document is Kb/s (Kilo bits per second) or Mb/s (Me ga bits per second). This metric is v ery useful to sho w that the scheme pro vides good support for service dif ferentiation. The traf c o ws with the highest priority should get the highest amount of a v erage throughput while the traf c o ws with the lo west priority should get the least amount. A v ery lo w near zero, throughput means traf c starv ation for that particular o w or class of traf c. 19

PAGE 29

A ver a g e Delay : This is the a v erage amount of delay that MA C frames belonging to a particular traf c o w e xperience during the entire duration of the o w The measurement used to represent a v erage delay in this document is ms (milliseconds). This metric lik e the a v erage throughput metric is v ery useful in demonstrating that the scheme pro vides good support for service dif ferentiation. The traf c o ws with the highest priority should get the least amount of a v erage delay while the traf c o ws with the lease priority should get the least amount.Instantaneous Thr oughput : This is the amount of information that a particular traf c o w is able to transmit successfully per second at an y gi v en second during the duration of a o w The measurement used to represent instantaneous throughput in this document is Kb/s or Mb/s. F or streaming applications, it is not only important that the y are supported by good service dif ferentiation, b ut also important that there is no violation of classes and the v ariability in service is v ery less. This metric helps in sho wing the le v el of v ariability and violation of classes if an y .Instantaneous Delay : This is the amount of delay that MA C frames are e xperiencing per second at an y gi v en second during the duration of the o w The measurement used to represent instantaneous delay throughout this document is ms. This metric has the same signicance as that of Instantaneous Throughput. It is especially v ery useful in sho wing the amount of jitter that a particular traf c class is e xperiencing.Jitter : This is the amount of v ariability found in the delay of the MA C frames belonging to a particular traf c o w It this document, it is e xpressed in milliseconds and is calculated as the Standard De viation (SD) or v ariance of the delay .P ac k et Loss Ratio (PLR) : This is the ratio of the number of MA C frames lost to the number of MA C frames transmitted. This is especially important for the data-oriented, best ef fort lo w priority traf c. 20

PAGE 30

3.1.2 Classes of T rafc F or the purpose of e v aluating the le v el of service dif ferentiation or QoS support that EDCA can pro vide, netw ork traf c has been di vided into three dif ferent classes:Gold class : This class of traf c is assigned to the A C that has the highest priority High quality streaming v oice and video o ws are assigned to this class.Silver class : This class is assigned to an A C that has less priority than that of the gold class. Lo wer quality streaming video and best-ef fort services are assigned to this class.Br onze class : Only best-ef fort is assigned to this class. T raf c assigned to bronze class has less priority than that of traf c assigned to gold or silv er classes. Since priorities cannot be assigned to traf c o ws in DCF this document uses the labels Class1, Class2 and Class3 to dif ferentiate between dif ferent types of o ws. 3.2 Implementation in ns-2.26 The performance e v aluation of the IEEE 802.11e EDCA and its comparison with the IEEE 802.11 DCF has been done using the netw ork simulator -2 (ns-2). ns-2 pro vides a comprehensi v e simulation test-bed for wired and wireless netw orks and has modeling of protocols at all the netw ork layers. ns-2 comes with support DCF b ut not EDCA. An e xtension from the T elecommunications Netw orks Group, T echnical Uni v ersity Berlin [35 ] pro vides the required implementation of the EDCA. This w as inte grated into the MA C layer of the ns-2 for the purpose of the performance e v aluation. 3.2.1 Ph ysical Lay er The wireless interf ace emulates the 914 MHz Lucent W a v e-LAN Direct-Sequence SpreadSpectrum (DSSS) radio interf ace [37 ]. W a v e-LAN is modeled as a shared-media radio with the follo wing properties:A nominal bit rate of 2 Mb/s. 21

PAGE 31

A nominal radio range of 250 meters.Each mobile possesses an unidirectional antenna with unity gain.The coordinates of the antenna on the mobile node are (X = 0.0, Y = 0.0, Z = 1.5), i.e. the antenna is located at the center of the mobile node and at a height of 1.5 meters.The recei v er and transmitter gain (i.e. Gr and Gt) are both unity Each mobile node has characteristics lik e v elocity and position information associated to it. Recie v ed signal po wer is estimated by using the position of a node which is calculated as a function of time. When a pack et is recei v ed at a mobile node, its recei v ed po wer is estimated by the physical layer interf ace. If the recei v ed po wer is lesser than carrier sense threshold, the pack et is discarded and is mark ed as error This occurs before the pack et is passed to the MA C layer In all other cases the pack et is handed o v er to the MA C layer 3.2.2 Medium Access Contr ol Lay er The Medium Access Control (MA C) layer in ns-2.26 used the IEEE 802.11 DCF by def ault. DCF has been briey e xplained in Chapter 2. F or a complete documentation of the standard, the reader is referred to [38 ]. F or details about its implementation in ns-2.26, the reader is referred to [36 ]. Extension from the T elecommunications Netw orks Group, T echnical Uni v ersity Berlin [35 ] ha v e been used to add the functionality of IEEE 802.11e EDCA to the MA C layer in ns-2.26. 3.3 Simulation Scenario In order to mak e a f air comparison, we ha v e to use tw o slightly dif ferent scenarios for EDCA and DCF The scenarios are illustrated in Figure 3.1 and Figure 3.2 respecti v ely Since the DCF does not support prioritization of traf c classes, i.e. it just uses a single scheduling queue for the transmission of the frames of all the traf c classes, multiple stations ha v e to be used to emulate multiple queues. This is in contrast to what EDCA does, that is, it uses a separate scheduling queue for e v ery traf c class and so only a single station can be used. Therefore, when EDCA is being used, it does not mak e a dif ference if we ha v e multiple sources of traf c in a single wireless station, since 22

PAGE 32

Figure 3.1 EDCA Simulation Scenario each source of traf c w ould use a dif ferent scheduling queue. In the scenario that has been used for EDCA (Figure 3.1), Station 1 acts as the source for 3 UDP (User Datagram Protocol) CBR(Constant Bit Rate) o ws and Station 2 acts as the sink. Each UDP source uses a dif ferent scheduling queue in Station 1. In the scenario that has been used for the DCF (Figure 3.2), station 1, 2 and 3 all act as a source for one UDP CBR o w each. Station 4 acts as the sink for the o ws. In case of the EDCA, each of the CBR o ws is assigned to Gold, Silv er and Bronze class respecti v ely while, in case of the DCF each of them is assigned to Class 1, 2 and 3 respecti v ely T able 3.1 Simulation P arameters Used in the AIFS, CW and PF Schemes P arameter AIFS CW PF Gold Silv er Bronze Gold Silv er Bronze Gold Silv er Bronze PF 2 2 2 2 2 2 1 2 3 AIFS 1 2 3 1 1 1 1 1 1 nC
PAGE 33

Figure 3.2 DCF Simulation Scenario The tw o stations are located 5 meters apart and are in the range of each other The physical characteristics of the mobile nodes netw ork interf ace are modeled to emulate the Lucent W a v eLAN DSSS radio interf ace. The physical channel emulates the tw o ray ground propagation model. Metrics of interest ha v e been described in Section 3.1.1. T able 3.1 and 3.2 sho w the parameters that ha v e been used for each of the schemes. The v alues for the parameters ha v e been chosen so as to represent the three seperate le v els of priority i.e. Gold, Silv er and Bronze. The nodes do not e xhibit an y kind of mobility and are assumed to be static. 3.3.1 Communication Model The three UDP CBR o ws mentioned in the pre vious section are set at dif ferent transmission rates ranging from 0.1 Mb/s to 3.0 Mb/s to e v aluate the schemes at dif ferent netw ork loads i.e. from v ery lo w netw ork loads to v ery high netw ork loads. All three of the traf c o ws transmit at the same rate at a gi v en point of time and the y persist throughout the duration of the o w The pack et size in all cases is 1000 bytes. CBR o ws are chosen for the simulations so the the netw ork load can be set to a particular v alue and the netw ork can be e v aluated for that particular traf c load. Also, CBR 24

PAGE 34

o ws ha v e stringent requirements in terms of delay and jitter This gi v es the opportunity to e v aluate the performance of the netw ork when dealing with applications lik e v oice. The bandwidth of fered by the channel is 2 Mb/s. The classication that has been used to describe dif ferent netw ork loads and their representati v e v alues is as follo ws:High Network Load : All the CBR traf c o ws transmit at 0.6 Mb/s, making the total netw ork load 1.8 Mb/s.Medium Network Load : All the CBR o ws transmit at 0.35 Mb/s, making the total netw ork load 1.05 Mb/s.Low Network Load : All the CBR o ws transmit at 0.2 Mb/s, making the total netw ork load 0.6 Mb/s. 3.4 Results All the schemes mentioned ha v e been e v aluated on the basis of three dif ferent criteria: service dif ferentiation, service v ariability and pack et loss rates and starv ation. 3.4.1 Ser vice Differ entiation The rst goal of the study is to e v aluate the v e dif ferent schemes in terms of their capability to pro vide service dif ferentiation. In order to sho w this, graphs with the a v erage delay and throughput are included in Figures 3.3 and 3.4. Se v eral observ ations can be made from these plots. The rst observ ation is that it is clear that with the e x emption of DCF all proposed schemes pro vide service dif ferentiation in terms of a v erage delay The a v erage delay observ ed by bronze pack ets is al w ays bigger than silv er pack ets, and silv er pack ets obtaining higher a v erage delays than gold pack ets. The second observ ation is that at lo w loads, i.e. at loads less than 0.8 Mb/s, all schemes, including DCF perform pretty much the same w ay and good. This is v ery important because it says that to pro vide good service dif ferentiation, the proposed scheme need to be ef fecti v e only during medium to high loads, i.e. at loads higher than 0.8 Mb/s. 25

PAGE 35

0 0.5 1 1.5 2 2.5 3 0 2 4 6 8 10 12 Network Load (Mb/s) Average Delay (Seconds) Gold Silver Bronze (a) AIFS 0 0.5 1 1.5 2 2.5 3 0 2 4 6 8 10 12 Network Load (Mb/s) Average Delay (Seconds) Gold Silver Bronze (b) CW 0 0.5 1 1.5 2 2.5 3 0 2 4 6 8 10 12 Network Load (Mb/s) Average Delay (Seconds) Gold Silver Bronze (c) PF 0 0.5 1 1.5 2 2.5 3 0 2 4 6 8 10 12 Network Load (Mb/s) Average Delay (Seconds) Gold Silver Bronze (d) AIFS+CW 0 0.5 1 1.5 2 2.5 3 0 2 4 6 8 10 12 Network Load (Mb/s) Average Delay (Seconds) Gold Silver Bronze (e) AIFS+CW+PF 0 0.5 1 1.5 2 2.5 3 0 2 4 6 8 10 12 Network Load (Mb/s) Average Delay (Seconds) Gold Silver Bronze (f) DCF Figure 3.3 A v erage Delay v ersus Load of AIFS, CW PF AIFS+CW AIFS+CW+PF and DCF Schemes 26

PAGE 36

0 0.5 1 1.5 2 2.5 3 0 0.2 0.4 0.6 0.8 1 Network Load (Mb/s) Average Throughput (Mb/s) Gold Silver Bronze (a) AIFS 0 0.5 1 1.5 2 2.5 3 0 0.2 0.4 0.6 0.8 1 Network Load (Mb/s) Average Throughput (Mb/s) Gold Silver Bronze (b) CW 0 0.5 1 1.5 2 2.5 3 0 0.2 0.4 0.6 0.8 1 Network Load (Mb/s) Average Throughput (Mb/s) Gold Silver Bronze (c) PF 0 0.5 1 1.5 2 2.5 3 0 0.2 0.4 0.6 0.8 1 Network Load (Mb/s) Average Throughput (Mb/s) Gold Silver Bronze (d) AIFS+CW 0 0.5 1 1.5 2 2.5 3 0 0.2 0.4 0.6 0.8 1 Network Load (Mb/s) Average Throughput (Mb/s) Gold Silver Bronze (e) AIFS+CW+PF 0 0.5 1 1.5 2 2.5 3 0 0.2 0.4 0.6 0.8 1 Network Load (Mb/s) Average Throughput (Mb/s) Gold Silver Bronze (f) DCF Figure 3.4 A v erage Throughput v ersus Load of AIFS, CW and PF AIFS+CW AIFS+CW+PF and DCF Schemes 27

PAGE 37

When the AIFS, CW and PF parameters are utilized alone, the schemes perform f airly similar with a small adv antage of the simpler PF scheme, which pro vides lo wer a v erage delay for the silv er and bronze classes while pro viding a similar delay for the gold class. AIFS and CW perform v ery similar with CW pro viding a little better dif ferentiation between the silv er and gold classes while pro viding better delay for the bronze class. Finally when the parameters are used in combination, the main observ ation is that the AIFS+CW and AIFS+CW+PF schemes pro vide a considerably better service for the gold class b ut at the e xpense of the other classes, in particular the bronze class, which no w recei v es a considerably longer a v erage delay The high delay v alues for the bronze class are because of the f act that at medium to high netw ork loads, the frequenc y at which the bronze class gets access to the medium is v ery lo w Retransmissions at the MA C layer further add to the delay High delay v alues for the bronze class means that the rate at which the frames are being sent out of the bronze class MA C queue is less than the rate at which the y are entering the MA C queue, thereby leady leading to o v ero w and loss of pack ets. This is further supported by Figure 3.7 which sho ws the P ack et Loss Rates (PLR) for all the classes of traf c at dif ferent netw ork loads. It is important to mention that the a v erage delay for the gold class traf c is almost constant at all load v alues, which is v ery good for streaming applications. As e xpected, the DCF scheme pro vides similar a v erage delay v alues to all classes. The a v erage delay is rather lo w when compared to the delays e xperienced by the other schemes because of the higher corresponding pack et loss rates e xperienced by the o ws in the DCF scheme. A v erage throughput v alues were also considered. Figure 3.4 sho ws the a v erage throughput obtained by each traf c class for all six schemes. As it can be seen, there is a good match between the a v erage delay and throughput results. F or e xample, the PF scheme pro vides better throughput for the silv er and bronze classes than the AIFS and CW schemes, which penalizes the bronze class hea vier Also, as a general observ ation, at medium to high loads, the throughput reduces with the netw ork load for the silv er and bronze classes while remains f airly constant for the gold class. All classes of traf c recie v e almost the same amount of throughput at lo w loads. This is because at lo w loads, the frequenc y at which the dif ferent classes of traf c contend for the channel is lo w thereby resulting in all the classes of traf c being able to access the channel at a f air rate. As the 28

PAGE 38

netw ork load increases, the frequenc y at which the dif ferent classes of traf c contend for the channel increases. The gold and silv er traf c classes ha ving more priority get access to the channel f aster and more frequenty compared to the bronze class, thereby leading the bronze class to achie v e lo w throughput at high netw ork traf c loads. Throughput for the bronze class is futher reduced because of frames being dropped at the queue in the MA C layer This happens because, the rate the which frames are transmitted out of the queue is less than the rate at which frames arri v e at the queue, thereby leading to o v ero w This f act is corroborated by the graphs sho wing the P ack et Loss Rates (PLR) for all the classes at dif ferent netw ork loads in Figure 3.7. The AIFS+CW and AIFS+CW+PF schemes pro vide the best throughput for the gold class at the e xpense of the silv er and bronze classes. In particular the y both penalize the bronze class v ery hea vily reducing its throughput to v ery lo w v alues. Considering these results and the f act that the PF scheme is the simplest to implement on top of the DCF scheme, looking at a v erage v alues, it may be a v ery good option. As e xpected, the throughput achie v ed by the three classes using the DCF scheme recei v e f air throughput v alues [39 ]. 3.4.2 Ser vice V ariability Service v ariability is also v ery important, in particular for streaming applications that need a rather steady end-to-end delay (no jitter) and throughput. Figures 3.5 and 3.6 include the instantaneous delay and throughput obtained by the gold and silv er classes at high loads, and the gold class at medium loads. Although results were obtained for the delay and throughput v ariations in all cases, all the graphs were not included because at medium to lo w loads all the schemes recei v ed similar good service, presenting almost unnoticeable v ariability As it can be seen from the graphs, there are important dif ferences among the schemes. F or instance, the schemes utilizing only one parameter (i.e., AIFS, CW or PF only) pro vide considerably higher v ariability than the combined schemes (AIFS+CW and AIFS+CW+PF), specially at high loads for the gold class. As no scheme pro vides clear better results in terms of v ariability for the silv er class at high loads, it might be used for those applications that need better a v erage service b ut ha v e better mechanisms to deal with v ariability such as compressed video applications with good playout b uf fers to minimize jitter 29

PAGE 39

10 20 30 40 50 60 70 80 90 0 0.2 0.4 0.6 0.8 1 1.2 Time (Seconds) Delay (Seconds) PF AIFS AIFS+CW AIFS+CW+PF CW (a) Delay V ariation for Gold Class at High Loads 10 20 30 40 50 60 70 80 90 0 0.5 1 1.5 2 2.5 3 Time (Seconds) Delay (Seconds) PF AIFS AIFS+CW AIFS+CW+PF CW (b) Delay V ariation for Silv er Class at High Loads 10 20 30 40 50 60 70 80 90 0 0.05 0.1 0.15 0.2 Time (Seconds) Delay (Seconds) PF AIFS AIFS+CW AIFS+CW+PF CW (c) Delay V ariation for Gold Class at Medium Loads Figure 3.5 Delay V ariation 30

PAGE 40

0 20 40 60 80 0 0.2 0.4 0.6 0.8 1 Time (Seconds) Throughput (Mb/s) PF AIFS AIFS+CW AIFS+CW+PF CW (a) Throughput v ariation for gold class at high loads 0 20 40 60 80 0 0.2 0.4 0.6 0.8 1 Time (Seconds) Throughput (Mb/s) PF AIFS AIFS+CW AIFS+CW+PF CW (b) Throughput v ariation for silv er class at high loads 0 20 40 60 80 0 0.2 0.4 0.6 0.8 1 Time (Seconds) Throughput (Mb/s) PF AIFS AIFS+CW AIFS+CW+PF CW (c) Throughput v ariation for gold class at medium loads Figure 3.6 Throughput V ariation 31

PAGE 41

T able 3.3 Mean and Standard De viation of Delay and Throughput of the AIFS, CW and PF Schemes in High Netw ork Load Conditions P arameter AIFS CW PF Gold Silv er Bronze Gold Silv er Bronze Gold Silv er Bronze Mean Delay 0.6393 1.6931 4.7719 0.3348 1.8519 4.4917 0.8486 1.3670 2.9579 StdDe v Delay 0.1179 0.3007 2.3066 0.2755 0.2158 1.2793 0.2091 0.2224 1.3649 Mean Tput 0.6940 0.2877 0.1223 0.7236 0.2581 0.1083 0.5614 0.3562 0.1846 StdDe v Tput 0.0568 0.0577 0.0616 0.0463 0.0368 0.0435 0.1028 0.0643 0.0853 T able 3.4 Mean and Standard De viation of Delay and Throughput of the AIFS+CW and AIFS+CW+PF and DCF Schemes in High Netw ork Load Conditions P arameter AIFS+CW AIFS+CW+PF DCF Gold Silv er Bronze Gold Silv er Bronze Class 1 Class 2 Class 3 Mean Delay 0.0203 1.8049 5.6174 0.0195 1.7072 7.4128 1.4134 1.4351 1.4470 StdDe v Delay 0.0042 0.1828 1.2825 0.0037 0.2323 4.2595 0.1621 0.1554 0.1752 Mean Tput 0.7399 0.2631 0.0849 0.7398 0.2814 0.0790 0.3352 0.3290 0.3269 StdDe v Tput 0.0100 0.0261 0.0497 0.0092 0.3980 0.0545 0.0443 0.0499 0.0463 It is important to notice that although the PF scheme w as considered a good choice in terms of a v erage v alues, it is the scheme pro viding the w orst service in terms of v ariability This sho ws the importance of looking at both metrics while assessing the performance of schemes of this nature. At medium loads, it can be seen that all schemes pro vide better service v ariability for the gold class. Ho we v er it can be seen that in terms of delay and throughput, while most schemes pro vide almost no v ariability the PF scheme beha v es poorly Ho we v er it is important to mention that violations in the service dif ferentiation were not observ ed from the a v erage results, and were not observ ed in the v ariability results either T able 3.3 and 3.4 summarize all the results obtained including the mean and standard de viation for the delay and throughput of all traf c classes for the priority-based schemes and DCF under high load conditions. From these tables, it can be concluded that:All priority based schemes pro vide good service dif ferentiation in terms of a v erage v alues.The schemes using more than one parameter pro vide v ery good service to gold users in terms of both, delay and throughput.Compared to the standard DCF scheme, the v ariability in throughput and delay for the silv er and bronze classes is rather w orse. 32

PAGE 42

The bronze class in the multiparameter schemes is highly penalized, recei ving v ery high delay and lo w throughput, clear indications of starv ation. 3.4.3 P ack et Loss Rates and Star v ation Finally the pack et loss rate e xperienced by the dif ferent traf c classes is also studied for all the schemes. Se v eral observ ations can be made looking at the results sho wn in Figure 3.7. First, the pack et loss rates (PLR) also reect the service dif ferentiation observ ed in the delay and throughput metrics. Second, matching pre vious results, it can be seen that PLRs at lo w load are almost zero in all cases. It can also be observ ed that the schemes using only one parameter start dropping pack ets from the gold class earlier than the multiparameter schemes. Finally it can be seen that at medium to high loads, the PLR is f airly high for silv er and bronze classes. This result may impact applications using the silv er class more than those using the bronze class since silv er classes may still be used for real-time applications. In this case, the PLR may be high enough to deteriorate the user percei v ed quality of the video more than e xpected. F or applications using bronze classes, this result means that a v ery lo w throughput will be achie v ed and man y retransmission will tak e place due to higher layer protocols lik e TCP It is important to mention that none of the e v aluated schemes can protect lo w priority o ws from starv ation. In the cases studied here, all traf c sources get at least a small portion of the bandwidth because all o ws transmit the same amount of information and there are only three o ws competing. This w as done this w ay in order to be able to see and analyze the ef fect of the mechanisms more clearly Ho we v er if this mix of traf c and o ws is changed and for e xample more gold traf c o ws are present in the netw ork, it may happen that bronze traf c o ws will ha v e access to the netw ork in v ery rare occasions. 3.5 Conclusion This chapter presents a performance e v aluation of the IEEE 802.11e EDCA standard. The IEEE 802.11e EDCA uses the parameters Aribitrary Inter Frame Space (AIFS), Contention W indo w (CW) and Priority F actor (PF) to pro vide support for service dif ferentiation. 33

PAGE 43

0 0.5 1 1.5 2 2.5 3 0 0.2 0.4 0.6 0.8 1 Network Load (Mb/s) Packet Loss Rate Gold Silver Bronze (a) AIFS 0 0.5 1 1.5 2 2.5 3 0 0.2 0.4 0.6 0.8 1 Network Load (Mb/s) Packet Loss Rate Gold Silver Bronze (b) CW 0 0.5 1 1.5 2 2.5 3 0 0.2 0.4 0.6 0.8 1 Network Load (Mb/s) Packet Loss Rate Gold Silver Bronze (c) PF 0 0.5 1 1.5 2 2.5 3 0 0.2 0.4 0.6 0.8 1 Network Load (Mb/s) Packet Loss Rate Gold Silver Bronze (d) AIFS+CW 0 0.5 1 1.5 2 2.5 3 0 0.2 0.4 0.6 0.8 1 Network Load (Mb/s) Packet Loss Rate Gold Silver Bronze (e) AIFS+CW+PF 0 0.5 1 1.5 2 2.5 3 0 0.2 0.4 0.6 0.8 1 Network Load (Mb/s) Packet Loss Rate Gold Silver Bronze (f) DCF Figure 3.7 P ack et Loss Rate v ersus Load of AIFS, CW PF AIFS+CW AIFS+CW+PF and DCF Schemes 34

PAGE 44

Fi v e dif ferent schemes using these parameters are included in the e v aluation. The rst scheme assigns dif ferent Arbitration Inter Frame Space (AIFS) interv als to dif ferent classes to pro vide differentiation. The second scheme assigns dif ferent v alues of the Contention W indo w (CW), and the third scheme assigns a dif ferent Priority F actor (PF) to each class, which only multiplies the Backof f Interv al (BI) included in the current DCF standard by a dif ferent number The fourth and fth schemes utilize the AIFS and CW and AIFS, CW and PF together respecti v ely as a w ay to pro vide better dif ferentiation. The results sho w that in terms of a v erage delay and throughput, the single parameter schemes pro vides good dif ferentiation. Ho we v er the same single parameter schemes are the w orst performing scheme in terms of v ariability in particular for the highest priority class under medium to high loads. In order to obtain good throughput and lo w v ariability for streaming applications, the scheme cannot rely on just one parameter and a multiparameter scheme is needed. All schemes perform f airly similar and well under lo w to medium loads. None of these schemes can actually a v oid starv ation. Finally tw o important conclusions are that:Single parameter schemes pro vide good service dif ferentiation, b ut poor service v ariability .Muti-parameter schemes pro vide good service v ariability b ut lead the lo w priority traf c to starv ation. 35

PAGE 45

CHAPTER 4 MEASUREMENT -B ASED ADMISSION CONTR OL MECHANISM FOR IEEE 802.11E This chapter describes the measurement-based admission control mechanism proposed, elaborates on its functionality describes it with a sample scenario and then presents the results pro ving its functionality 4.1 Pr oblem Statement and Design Considerations Chapter 3 concluded that starv ation and jitter are the tw o important problems that the IEEE 802.11e EDCA f aces. These tw o problems are in v ersely proportional to each other The single parameter schemes (PF CW and AIFS) impro v ed on the starv ation of the bronze class to an e xtent, b ut resulted in the gold class e xhibiting a high amount of jitter In contrast, when the multiparameter schemes (AIFS+CW and AIFS+CW+PF) were used, the gold class e xhibited v ery lo w jitter b ut e v entually led the bronze class traf c to starv ation. In order to protect e xisting traf c o ws from either of the abo v e mentioned conditions, it is necessary to ha v e an ef fecti v e admission control mechanism. In an admission control mechanism, a ne w traf c o w is admitted based on current netw ork conditions. If the introduction of the ne w o w were to cause intolerable deterioration of the e xisting o ws, then the o w w ould not be admitted, else it w ould be. The admission control mechanism introduced in this document is measur ementbased i.e. it relies metrics tak en from the netw ork, in particular bandwidth currently being utilized (in terms of Mb/s) and jitter The v alues for these parameters are maintained at all times. T o de v elop an ef fecti v e and feasible admission control mechanism for the IEEE 802.11e EDCA, the follo wing design considerations ha v e to be k ept in mind: 36

PAGE 46

The mechanism should be completely distrib uted in nature. All admission control decisions ha v e to be tak en locally and should not be based on information present at other wireless stations within the netw ork.The mechanism should try to maximize netw ork utilization, b ut not at the cost of af fecting the performance of the e xisting traf c o ws.Changes to the e xisting protocol should not be drastic. This helps in establishing backw ard compatibility .No preemption should be used. The decision to admit or to reject a traf c o w should be tak en at the time when the traf c o w requests admission to the channel and not after it has been admitted into the netw ork. 4.2 Ov er view and Design of the Admission Contr ol Mechanism F or simplicity this document from hereon refers to only tw o types of traf c: High Priority and Lo w Priority High Priority traf c refers to traf c o ws with real-time requirements lik e v oice and video (UDP CBR traf c), whereas Lo w Priority traf c refers to best ef fort traf c (mostly TCP traf c). Chapter 3 strongly concluded that the multiparameter scheme (AIFS+CW+PF) greatly reduces jitter b ut in turn leads to starv ation. If this scheme is used along with a bandwidth allocation mechanism, then a minimum amount of bandwidth can be assured for the lo w priority traf c. Figure 4.1 illustrates a situation where the total bandwidth is 2 Mb/s and 80% of the netw ork bandwidth, i.e. 1600 Kb/s is the minimum assured to the lo w priority traf c. Minimum assured bandwidth for the lo w priority traf c can be interpreted as to say that high priority traf c will not infringe on that part of the netw ork bandwidth. The same does not hold true for the lo w priority traf c. When the bandwidth allocated to the high priority traf c is being under utilized, the lo w priority traf c, o wing to the nature of TCP w ould try to use that under utilized part of the bandwidth. The logic behind allo wing this is that when a ne w high priority traf c o w requesting admission to the medium is admitted, it w ould automatically reduce the amount of bandwidth being used by the 37

PAGE 47

Figure 4.1 Bandwidth Allocation for High Priority and Lo w Priority T raf c lo w priority traf c. This is due to the f act that when the AIFS+CW+PF scheme is used for service dif ferentiation, the probability of the high priority traf c getting access to the medium is greatly increased. The v alue for the maximum a v ailable bandwidth of the netw ork can be set the netw ork administrator using his kno wledge of the netw ork and types of applications. The abo v e mentioned mechanism impro v es upon netw ork utilization while protecting the high priority traf c from performance deterioration and also guarantees a certain minimum amount of bandwidth for the lo w priority traf c. This is well supported by the results that are presented later in this chapter The mechanism has been implemented by modifying the e xisting IEEE 802.11e EDCA standard. No w the IEEE 802.11e EDCA is not completely backw ard compatible with the DCF This is because the v alues for AIFS for all classes of traf c in the IEEE 802.11e EDCA are greater than the v alue of DIFS in IEEE 802.11 DCF according to the current form of the standard [12 ]. Thus, the IEEE 802.11e EDCA forces change in WLAN infrastructure. The proposed admission control mechanism can be implemented as part of the IEEE 802.11e EDCA standard when it is nalized. Making it a part of the standard will eliminate an y more infrastructure changes and backw ard compatibility issues. Applications are e xpected to interact with the MA C layer to kno w if traf c generated by them can be admitted into the netw ork. This is because the MA C cannot 38

PAGE 48

dif ferentiate between frames belonging to dif ferent traf c o ws, so o ws ha v e to be rejected at the application layer 4.2.1 Metrics Used f or Admission Contr ol In addition to the metrics discussed in Section 3.1.1, the follo wing metrics are used to implement the admission control mechanism:Curr ently utilized bandwidth : This metric is calculated at e v ery wireless station within the netw ork that is in listening range. When a wireless station recei v es a frame, it checks whether the frame is addressed to it. If it is not, it is discarded. In this admission control mechanism, before a frame is discarded, the wireless station notes its priority and increments a local counter depending on the priority A separate counter is maintained for high priority and lo w priority The counters are used to calculate the amount of bandwidth in the netw ork being utilized by both the lo w priority and high priority traf c. All bandwidth measurements are done in terms of Mb/s or Kb/s in this document. F or simplicity it is assumed that the wireless netw ork does not suf fer from issues lik e the hidden terminal problem or the e xposed terminal problem and all stations hear all transmissions.Jitter : This metric is used only for the frames belonging to the high priority traf c cate gory By denition, jitter is the v ariation in the inter -arri v al times of the pack ets belonging to a particular o w Since the MA C cannot dif ferentiate between frames belonging to dif ferent o ws, it is impossible to calculate jitter as mentioned abo v e. An alternati v e method that is used in this study to calculate jitter is to nd the v ariation in the delays of the frames belonging to the high priority traf c. Ev ery high priority traf c frame is time-stamped before it goes into the queue to contend for the channel. The size of the timestamp is 4 bytes and is included in the body of the MA C frame. The size of the body of a MA C frame ranges from 0 to 2312 bytes depending on the size of the pack et recie v ed from the netw ork layer If the size of the pack et is lar ger than 2312 bytes, then the pack et is split into multiple MA C frames. The timestamp is just added to the pack et recie v ed from the netw ork layer and the MA C layer tak es care of the 39

PAGE 49

se gmentation if required. Thus, the o v erhead induced due to the timestamp is ne gligible. As e xplained before, e v ery wireless station within the netw ork can note the delay of e v ery high priority frame it recei v es. The wireless station notes the timestamp in it before discarding the frame if it is not addressed to it. The v ariance of the delay v alues of the last 50 high priority frames recei v ed is calculated and maintained locally in e v ery wireless station. This v alue is used to determine whether to admit a ne w high priority o w or not. Only the last 50 frames are considered because the decision to admit a ne w traf c o w should depend on current v alues and not on the o v erall v alues. There may be dif ferences in delay calculated for a particular frame in dif ferent stations depending on the propogation delay Since, the propagation delay in a WLAN of small to medium size is ne gligible, we ignore the dif ference. It is assumed that all stations ha v e synchronized clocks.Network Utilization : This metric measures the total amount of bandwidth in the medium that is being utilized by all the types of traf c in the netw ork. This metric is v ery useful in determining the success of the admission control mechanism. 4.2.2 Algorithm Used f or the Admission Contr ol Mechanism A o wchart for the algorithm being used is gi v en in Figure 4.2. As mentioned in the pre vious section, e v ery wireless station in the netw ork within listening range locally maintains the metrics, curr ently utilized bandwidth and jitter F or simplicity the algorithm sho wn considers only tw o classes of traf c, high priority and lo w priority If needed, more classes of traf c can be dened to support dif ferent types of applications. In the proposed admission control mechanism, when an application w ants to send data o v er the netw ork, it rst communicates with the MA C layer passing parameters that tell the MA C layer its priority The MA C layer depending on the priority of traf c performs one or tw o tests. If the application requesting admission to the medium generates high priority traf c, then the MA C performs tw o tests, i.e. rstly it checks whether the jitter being e xperienced by high priority traf c in the netw ork is less than a specied limit and secondly it checks whether the bandwidth a v ailable in the netw ork is more than the amount of bandwidth required by the traf c o w This specied 40

PAGE 50

Figure 4.2 Flo wchart for the Admission Control Mechanism limit for jitter and the bandwidth allocation for the high priority and lo w priority traf c can be set by the administrator on the basis of his kno wledge of the netw ork and applications. The amount of bandwidth required by a particular class of traf c is set to a certain predened v alue depending on the application that is intended to used that traf c class. F or e xample, if the high priority class were to be used by v oice o ws, then the bandwidth requirement of the high priority traf c o ws is set at 64 Kb/s [4]. Only video o ws w ould be assigned to that traf c class in that case. Similarly if video traf c were to be considered, then a dif ferent class of traf c can be dened in the mechanism with the bandwidth requirements set at 1024 Kb/s [4]. Only video o ws w ould be assigned to that traf c class. The dif ferent traf c classes and bandwidth requirements for each traf c class can be made 41

PAGE 51

standard. All applications will ha v e to set their priority such that the y are classied into a particular traf c class. 4.2.3 P arameters Set by Netw ork Administrator The netw ork administrator is gi v en the ability to modify the follo wing parameters in the proposed admission control mechanism:Bandwidth allocation : The netw ork administrator has the ability to set the limits on the amount of traf c that dif ferent traf c classes can use. An e xample of the bandwidth allocation scheme is gi v en in Section 4.2. The best-ef fort traf c class is not set a limit on the amount of bandwidth it can use, b ut is promised a minimum amount of bandwidth. The limits can be set based on the netw ork administrator' s kno wledge of the netw ork and its applications. An automated mechanism can be implemented such that the bandwidth allocation is done automatically based on the traf c characteristics of the wireless netw ork at a gi v en point of time.Maximum allowable jitter : The netw ork administrator also has the ability to set a limit on the amount of jitter being e xperienced by higher priority classes of traf c. When a ne w o w is being admitted, if the amount of jitter is more than the specied limit, then the ne w o w will be denied admission to the medium. The maximum allo w able jitter ho we v er is not the maximum amount of jitter that the netw ork can e xperience at an y point of time. It only species the maximum limit be yond which a ne w o w will not be allo wed. 4.2.4 Assumptions The admission control mechanism assumes that the netw ork does not suf fer from conditions lik e ”Hidden T erminal Problem” and the ”Exposed T erminal Problem”. If the netw ork suf fers from the ”Hidden T erminal Problem”, then it w ould not be able to mak e accurate measurements for the metrics Curr ently utilized bandwidth and jitter Future e xtensions can be designed and implemented to o v ercome these problems in the proposed mechanism, much lik e the R TS/CTS (Request 42

PAGE 52

to Send/Clear to Send) mechanism w as introduced in the IEEE 802.11 DCF to eliminate the ”Hidden T erminal Problem”. The admission control mechanism also assumes that all the stations ha v e synchronized clocks. If the clocks are not synchronized, it w ould be lead to calculation of incorrect delay v alues for the frames, ultimately resulting in f alse jitter v alues. [40 ] proposes a protocol for synchronizing clocks in the wireless stations of a WLAN. Results ha v e sho wn that the protocol can achie v e an accurac y of 150 microseconds. The accurac y required by the admission control mechanism is in the order of milliseconds. So this protocol can be used along with the admission control mechanism to achie v e clock synchronization. 4.2.5 Sample Scenario In order to better understand the admission control mechanism, this subsection e xplains a sample scenario with dummy data. Consider the scenario sho wn in Figure 4.3. T w o wireless stations are within range of each other and form a netw ork. Assume that the medium the y share has a bandwidth of 1 Mb/s. Both the stations acts as sources and sinks for high and lo w priority traf c. The algorithm for the admission control mechanism is set up such that the 80% of the bandwidth is allocated to lo w priority traf c and the remaining 20% is allocated to high priority traf c. Thus, high priority traf c can use a maximum of 200 Kb/s of the a v ailable bandwidth. The maximum allo w able jitter i.e. v ariance in delay is set as 2 ms. The v alue set here for jitter is only for the purpose of understanding the mechanism. Real end-to-end jitter v alues for v oice can be 50 ms at most [41 ]. The bandwidth required by o ws belonging to the high priority class is set at 64 Kb/s as this traf c class is only to be used by v oice o ws. Suppose that v e high priority v oice o ws request admission to the medium, each after a gap of 10 seconds alternately from Station 1 and Station 2 starting at time 10 seconds. Also, assume that 5 best-ef fort lo w priority o ws request for admission into the netw ork at time 0 seconds. F or each of these lo w priority o ws, the admission control mechanism checks if the current jitter v alue for high priority traf c in the netw ork is less than 2 ms. Since, there is no high priority traf c in the netw ork, the v alue for jitter is zero and so all the lo w priority o ws are admitted. When the rst high priority 43

PAGE 53

Figure 4.3 Sample Scenario o w requests admission to the medium from Station 1, the admission control mechanism again rst v eries if the amount of jitter for high priority traf c in the netw ork is less than 2 ms. As stated abo v e, the lack of high priority o ws in the netw ork causes the v alue for jitter to be zero. Ne xt the mechanism v eries if the amount of bandwidth required by the high priority v oice o w is less than amount of bandwidth a v ailable for it, i.e. the amount of bandwidth remaining out of the allocated quota. Again, since there are no high priority o ws in the netw ork, i.e. 200 Kb/s is the a v ailable bandwidth, and the bandwidth required by v oice o ws is 64 Kb/s, the high priority o w is admitted into the netw ork. In the manner described abo v e, assume that the second v oice o w is also admitted into the netw ork. Suppose this increases the jitter in the netw ork to around 3 ms at time 40 seconds, then the third v oice o w is rejected because the amount of jitter is greater than the specied limit. Assuming that the jitter in the netw ork decreases to belo w 2 ms and the total bandwidth being occupied by high priority traf c is 115 Mb/s at time 50 seconds, the fourth o w clears both the admission control conditions and is also admitted into the netw ork. If at time 60 seconds, the amount of bandwidth being occupied by the high priority traf c is 172 Mb/s and the jitter is less than 2 ms, then the fth high priority v oice o w is rejected because it f ails the second condition, i.e. there is less than 64 Mb/s bandwidth left out of the allocated bandwidth for the high priority traf c. 44

PAGE 54

Figure 4.4 Simulation Scenario for the Ev aluation of the Admission Control Mechanism 4.3 Implementation in ns-2 The measurement-based admission control mechanism described in the pre vious sections has been implemented in ns-2. Details about ns-2 and the e xtensions used ha v e been described in Section 3.2. The admission control mechanism has been implemented by modifying the IEEE 802.11e EDCA e xtension obtained from [24 ]. Details about the Physical and MA C layer are mentioned in Section 3.2.1 and 3.2.2 respecti v ely 4.4 Simulation Scenario The simulation scenario that has been used to e v aluate the admission control mechanism is sho wn in Figure 4.4. The three stations are located 5 meters apart from each other and are within range. Since most of the Internet traf c can be The physical characteristics of the mobile node netw ork interf aces are modeled to emulate the Lucent W a v eLAN DSSS radio interf ace. The physical channel emulates the tw o ray ground propagation model. 45

PAGE 55

T able 4.1 P arameters Used for the AIFS+CW+PF Scheme Used in the Admission Control Mechanism P arameter AIFS+CW+PF scheme High Priority Lo w Priority AIFS 1 3 PF 1 2 CW 7-15 15-1023 The simulation scenario only considers tw o classes of traf c, i.e. high priority and lo w priority V oice is cate gorized as high priority traf c and best-ef fort traf c is cate gorized as lo w priority The parameters for the high priority class are set to accomadate v oice o ws, i.e. bandwidth required by each o w of the high priority class is set to 64 Kb/s. All three stations act as sources and sinks for high priority UDP CBR o ws and lo w priority TCP FTP (File T ransfer Protocol) o ws. The metrics of interest are described in Section 4.2.1. The AIFS+CW+PF scheme is used for all the simulations. The reasons for choosing this scheme are described in Section 4.2. The parameters that ha v e been used for this scheme are sho wn in T able 4.4. It is assumed that the wireless stations do not e xhibit an y kind of mobility When the IEEE 802.11e EDCA standard is used for channel access, A Cs or classes of traf c contend for the channel and not the stations. Therefore for a scenario to be realistic, the number of A Cs or classes of traf c is what matters. Since most of the Internet traf c can be di vided into tw o main classes of traf c, high priority real-time traf c and lo w priority bestef fort traf c, the simulation scenario discussed abo v e is a good representati v e of a real wireless netw ork. 4.4.1 Communication Model A common simulation model is used for all the scenarios. T en lo w priority TCP FTP o ws request admission to the medium alternately from Station 1, 2 and 3 at time 10 seconds. Also, starting from time 10 seconds, ten UDP CBR o ws request admission to the medium at a gap of 10 seconds each, alternately from Station 1, 2 and 3. There is no acti vity in the netw ork for the rst 10 seconds. This has been considered due to implementation issues. The size of the FTP pack ets is set 46

PAGE 56

at 1000 bytes. The CBR o ws are transmitted at a rate of 64 Kb/s with a pack et size of 1000 bytes to emulate a v oice o w 4.5 Results T w o scenarios are presented in this section to e v aluate the performance of the admission control mechanism. Both the scenarios are simulated on a v ersion of the IEEE 802.11e EDCA without the proposed admission control mechanism and a v ersion with the proposed mechanism. As stated before both the v ersions use the AIFS+CW+PF scheme with the parameters specied in T able 4.4. Additionally for the IEEE 802.11e EDCA with the admission control mechanism, for both the scenarios, the bandwidth of the netw ork is partitioned such that 20% of it is allocated to high priority traf c and 80% of it is allocated to lo w priority traf c. These v alues are chosen because this is approximately the partitioning of traf c on the Internet [42 ]. Since the bandwidth of fered by the channel is 2 Mb/s, this translates to 400 Kb/s for the high priority traf c and 1600 Kb/s for the lo w priority traf c. The parameter that changes for each scenario is the amount of allo w able jitter F or scenario one, it is set to 3 ms and for Scenario tw o, it is set to 0.4 ms. The abbre viation A C used in the graphs stands for admission control. 4.5.1 Scenario One F or the IEEE 802.11e EDCA v ersion with the admission control mechanism the limit of allo wable jitter is set at 3 ms for this scenario. The v alue set here for jitter is only for the purpose of e v aluating the mechanism. Real end-to-end jitter v alues for v oice can be 50 ms at most [41 ]. In the IEEE 802.11e EDCA v ersion without the admission control mechanism, 10 lo w priority FTP o ws are admitted into the netw ork at time 10 seconds. Also starting at time 10 seconds, 10 UDP CBR o ws, each at a gap of 10 seconds are admitted into the netw ork alternately from Station 1, 2 and 3, i.e. the rst CBR o w is admitted into the netw ork from Station 1 with sink as Station 2 at time 10 seconds, the second CBR o w is admitted into the netw ork from Station 2 with sink as Station 3 at time 20 seconds and so on. While these o ws are directly admitted into the netw ork in the IEEE 802.11e EDCA without the admission control, in the IEEE 802.11e EDCA v ersion with 47

PAGE 57

T able 4.2 V alues for Jitter at T ime Instances when CBR Flo ws are Admitted in the Mechanism with Admission Control (Scenario 1) T ime Jitter v alue(ms) 10 0 20 0.05 30 0.08 40 0.2 50 0.4 60 0.7 the admission control, each high priority has to clear tw o tests and each lo w priority has to clear one test as described in Section 4.2 before being admitted into the netw ork. All the ten lo w priority FTP o ws are admitted into the netw ork at time 10 seconds in the IEEE 802.11e EDCA with the admission control mechanism because before time 10 seconds, there are no high priority o ws in the netw ork and so the v alue of jitter for the high priority traf c is zero. In this scenario, the rst six CBR o ws requesting admission into the medium, the rst at time 10 seconds and the sixth at time 60 seconds are all admitted into the netw ork because the amount of jitter in the netw ork at the time the y were admitted w as less then 3 ms and also the amount of bandwidth required by each of the o ws, i.e. 64Kb/s is less than the amount of remaining bandwidth for the high priority traf c. The v alue for jitter for the high priority traf c when each of these six o ws is admitted is sho wn in T able 4.2. CBR o ws 7-10, the se v enth requesting admission at time 70 seconds and the tenth requesting admission at time 100 seconds are all rejected because the amount of bandwidth a v ailable for the high priority traf c is less than 64 Kb/s. Figure 4.5(a) sho ws the instantaneous delay and throughput with and without admission control for the high priority class as a whole. There is clearly less v ariability in terms of delay when the admission control mechanism is used. The e xcessi v e amount of delay v ariability when the admission control mechanism is not used is caused because of high amount of traf c. This is pre v ented when the admission control mechanism is used. The amount of throughput achie v ed when the admission control mechanism is not used is higher than the throughput achie v ed when it is used after time 60 seconds. This is because when the admission control mechanism is used, after that time 60 seconds, no more CBR o ws are admitted into the netw ork. The amount of throughput 48

PAGE 58

achie v ed by a single v oice o w with or without admission control is almost the same, thus pro ving that the AIFS+CW+PF scheme pro vides v ery good service dif ferentiation. 49

PAGE 59

20 40 60 80 100 0 0.05 0.1 0.15 0.2 Time(Seconds) Delay(Seconds) Without AC With AC Mean-Without AC Mean-With AC 20 40 60 80 100 0 0.5 1 1.5 2 Time(Seconds) Throughput(Mb/s) Without AC With AC Mean-Without AC Mean-With AC (a) Instantaneous Delay and Throughput with and without Admission Control for High Priority Class (left to right) 30 40 50 60 70 80 90 100 110 0 0.05 0.1 0.15 0.2 Time(Seconds) Delay(Seconds) Without AC With AC Mean-Without AC Mean-With AC 30 40 50 60 70 80 90 100 110 0 0.05 0.1 0.15 0.2 Time(Seconds) Throughput(Mb/s) Without AC With AC Mean-Without AC Mean-With AC (b) Instantaneous Delay and Throughput with and without Admission Control for one V oice Flo w (left to right) Figure 4.5 Instantaneous Delay and Throughput with and without Admission Control for High Priority Class and one V oice Flo w in Scenario 1 Ho we v er the delays e xperienced in both the cases are quite dif ferent again after time 10 seconds when the admission control mechanism comes into play Thus, it is clearly e vident that the admission control mechanism pre v ents e xisting o ws from performance deterioration. Figure 4.5(b) sho ws the instantaneous delay and throughput with and without admission control for one v oice o w in particular Delay observ ed for one v oice o w is similar to the delay to the delay observ ed for the whole high priority class. 50

PAGE 60

Figures 4.6(a) and 4.6(b) display the Probability Density Function (PDF) with and without admission control for the delay v alues of the high priority traf c as a whole and for one single v oice o w respecti v ely It is clearly indicated by the graphs that when admission control mechanism is used, v ariability in jitter is considerably lo w 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 Delay(Seconds) ProbabilityVariance = 0.0005 Seconds Mean = 0.0437 Seconds 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 Delay(Seconds) ProbabilityVariance = 0.003 Seconds Mean = 0.0743 Seconds (a) Probability Density Function for the Delay of High Priority T raf c with and without Admission Control (left to right) 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 Delay(Seconds) ProbabilityVariance = 0.0005 Seconds Mean = 0.0520 Seconds 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 Delay(Seconds) ProbabilityVariance = 0.0028 Seconds Mean = 0.0812 Seconds (b) Probability Density Function for the Delay of one V oice Flo w with and without Admission Control (left to right) Figure 4.6 Probability Density Function for the Delay of High Priority T raf c and one V oice Flo w with and without Admission Control in Scenario 1 51

PAGE 61

20 40 60 80 100 0 2 4 6 8 10 Time(Seconds) Delay(Seconds) Without AC With AC Mean-Without AC Mean-With AC 20 40 60 80 100 0 0.5 1 1.5 2 Time(Seconds) Throughput(Mb/s) Without AC With AC Mean-Without AC Mean-With AC (a) Instantaneous Delay and Throughput with and without Admission Control for Lo w Priority Class (left to right) 20 40 60 80 100 0 2 4 6 8 10 Time(Seconds) Delay(Seconds) Without AC With AC Mean-Without AC Mean-With AC 10 20 30 40 50 60 70 0 0.1 0.2 0.3 0.4 0.5 Time(Seconds) Throughput(Mb/s) Without AC With AC Mean-Without AC Mean-With AC (b) Instantaneous Delay and Throughput with and without Admission Control for one FTP Flo w (left to right) Figure 4.7 Instantaneous Delay and Throughput with and without Admission Control for Lo w Priority Class and one FTP Flo w in Scenario 1 Figures 4.7(a) and 4.7(b) sho w the instantaneous delay and throughput with and without admission control for the lo w priority class as a whole and for one particular FTP o w respecti v ely The graphs clearly sho w that the lack of an admission control mechanism leads the lo w priority traf c to starv ation conditions, while the presence of an admission control mechanism pre v ents it by not allo wing more o ws than what the netw ork can handle. V ery high delays, i.e. in the order of seconds are observ ed when admission control is not used. When admission control is not used, ne w high priority o ws k eep getting admitted to the netw ork, thereby increasing the rate at which the y 52

PAGE 62

contend for the channel. This in turn, increases the number of times the lo w priority o ws ha v e to contend for the channel to gain access successfully A point is reached when the rate at which the lo w priority frames are transmitted from the MA C queue is less than the rate at which the y arri v e at the MA C queue leading to o v ero w and loss of frames. Upon not recie ving ackno wledgements for transmitted pack ets, the lo w priority TCP o ws retransmit the lost pack ets. TCP also decreases the transmission rate of pack ets, which is clearly observ ed by the lack of throughput being achie v ed by the lo w priority o ws when admission control is not used. The number of retransmissions and the time it tak es to contend for the channel successfully results in the high delay v alues for the lo w priority class when admission control is not used. 20 40 60 80 100 0 0.5 1 1.5 2 Time(Seconds) Throughput(Mb/s) Low Priority High Priority Mean-Low Priority Mean-High Priority High Priority + Low Priority 20 40 60 80 100 0 0.5 1 1.5 2 Time(Seconds) Throughput(Mb/s) Low Priority High Priority Mean-Low Priority Mean-High Priority High Priority + Low Priority Figure 4.8 Netw ork Utilization with and without Admission Control in Scenario 1 (left to right) Figure 4.8 sho ws the netw ork utilization with and without admission control in terms of throughput. The graphs clearly sho w that the admission control mechanism does not sacrice netw ork utilization at the e xpense of protecting e xisting o ws. In f act, netw ork utilization impro v es when the admission control mechanism is used. This is because, without admission control, the amount of high priority traf c contending for the channel k eeps on increasing causing starv ation for the lo w priority traf c. Also, since the lo w priority traf c is TCP the rate at which at which it transmits automatically reduces. This decreases the amount of throughput achie v ed by the lo w priority class. When admission control is used, the amount of high priority traf c contending for the channel and gaining access to it is restricted and so the amount of bandwidth a v ailable to the lo w priority traf c is higher when compared to the bandwidth a v ailable when admission control is not used. Higher 53

PAGE 63

amount of throughput being achie v ed by the lo w priority class when admission control is used leads to higher netw ork utilization. T able 4.3 P ack et Loss Rates for Scenario 1 T raf c class PLR-with A C PLR-without A C High Priority 0.0001 0.0001 Lo w Priority 0.0035 0.003 V oice o w 0.0 0.0 FTP o w 0.0032 0.0038 The pack et loss ratio(PLR) for the high priority and lo w priority traf c as a whole and for one particular v oice and one particular ftp o w with and without admission control mechanism are sho wn in T able 4.3. 4.5.2 Scenario T w o The traf c model used for this scenario is the same as the one used in Scenario 1. The only dif ference is the amount of allo w able jitter in the admission control mechanism is set to 0.0004 seconds. The v alue set here for jitter is only for the purpose of e v aluating the mechanism. Real end-to-end jitter v alues for v oice can be 50 ms at most [41 ]. This results in the CBR o w that requests admission to the medium at time 50 seconds getting rejected because the amount of jitter at that point of time is more than 0.0004 seconds. The CBR o ws that request admission at time 60 seconds and time 70 seconds are admitted because the amount of jitter at those points of time is less then 0.0004 and also the amount of bandwidth a v ailable for high priority traf c w as more than the amount required by the o ws. T able 4.4 sho ws the jitter v alues for the instances of time at which the o ws ha v e been accepted or rejected. Figures 4.9(a) and 4.9(b) sho w the instantaneous delay and throughput with and without admission control for the high priority class as a whole and for one particular v oice o w respecti v ely At time 50 seconds, the admission control mechanism starts beha ving dif ferently when compared to the one without admission control because a o w is rejected at that point of time to protect the e xisting o ws from increasing jitter v alues. 54

PAGE 64

T able 4.4 V alues for Jitter at T ime Instances when CBR Flo ws are Admitted in the Mechanism with Admission Control (Scenario 2) T ime Jitter v alue(ms) 10 0 20 0.05 30 0.08 40 0.2 50 0.425 60 0.3 70 0.2 55

PAGE 65

20 40 60 80 100 0 0.05 0.1 0.15 0.2 Time(Seconds) Delay(Seconds) Without AC With AC Mean-Without AC Mean-With AC 20 40 60 80 100 0 0.5 1 1.5 2 Time(Seconds) Throughput(Mb/s) Without AC With AC Mean-Without AC Mean-With AC (a) Instantaneous Delay and Throughput with and without Admission Control for High Priority Class (left to right) 30 40 50 60 70 80 90 100 110 0 0.05 0.1 0.15 0.2 Time(Seconds) Delay(Seconds) Without AC With AC Mean-Without AC Mean-With AC 30 40 50 60 70 80 90 100 110 0 0.05 0.1 0.15 0.2 Time(Seconds) Throughput(Mb/s) Without AC With AC Mean-Without AC Mean-With AC (b) Instantaneous Delay and Throughput with and without Admission Control for one V oice Flo w (left to right) Figure 4.9 Instantaneous Delay and Throughput with and without Admission Control for High Priority Class and one V oice Flo w in Scenario 2 56

PAGE 66

Figures 4.10(a) and 4.10(b) display the Probability Density Function (PDF) with and without admission control for the delay v alues of the high priority traf c as a whole and for one single v oice o w respecti v ely The beha vior of the graphs is similar to that in Scenario 1. 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 Delay(Seconds) ProbabilityVariance = 0.0005 Seconds Mean = 0.0437 Seconds 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 Delay(Seconds) ProbabilityVariance = 0.003 Seconds Mean = 0.0743 Seconds (a) Probability Density Function for the Delay of High Priority T raf c with and without Admission Control (left to right) 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 Delay(Seconds) ProbabilityVariance = 0.0005 Seconds Mean = 0.0508 Seconds 0 0.1 0.2 0.3 0.4 0.5 0 0.1 0.2 0.3 0.4 0.5 Delay(Seconds) ProbabilityVariance = 0.0028 Seconds Mean = 0.0812 Seconds (b) Probability Density Function for the Delay of one V oice Flo w with and without Admission Control (left to right) Figure 4.10 Probability Density Function for the Delay of High Priority T raf c and one V oice Flo w with and without Admission Control in Scenario 2 57

PAGE 67

Figures 4.11(a) and 4.11(b) sho w the instantaneous delay and throughput with and without admission control for the lo w priority class as a whole and for one particular FTP o w respecti v ely The beha vior of the graphs is similar to that in Scenario 1. 20 40 60 80 100 0 2 4 6 8 10 Time(Seconds) Delay(Seconds) Without AC With AC Mean-Without AC Mean-With AC 20 40 60 80 100 0 0.5 1 1.5 2 Time(Seconds) Throughput(Mb/s) Without AC With AC Mean-Without AC Mean-With AC (a) Instantaneous Delay and Throughput with and without Admission Control for Lo w Priority Class (left to right) 20 40 60 80 100 0 2 4 6 8 10 Time(Seconds) Delay(Seconds) Without AC With AC Mean-Without AC Mean-With AC 10 20 30 40 50 60 70 0 0.1 0.2 0.3 0.4 0.5 Time(Seconds) Throughput(Mb/s) Without AC With AC Mean-Without AC Mean-With AC (b) Instantaneous Delay and Throughput with and without Admission Control for one FTP Flo w (left to right) Figure 4.11 Instantaneous Delay and Throughput with and without Admission Control for Lo w Priority Class and one FTP Flo w in Scenario 2 58

PAGE 68

Figure 4.12 sho ws the netw ork utilization with and without admission control in terms of throughput. Once again the netw ork utilization impro v es when admission control mechanism is used. The pack et loss ratio(PLR) for the high priority and lo w priority traf c as a whole and for one particular v oice and one particular ftp o w with and without admission control mechanism are sho wn in T able 4.5. 20 40 60 80 100 0 0.5 1 1.5 2 Time(Seconds) Throughput(Mb/s) Low Priority High Priority Mean-Low Priority Mean-High Priority High Priority + Low Priority 20 40 60 80 100 0 0.5 1 1.5 2 Time(Seconds) Throughput(Mb/s) Low Priority High Priority Mean-Low Priority Mean-High Priority High Priority + Low Priority Figure 4.12 Netw ork Utilization with and without Admission Control in Scenario 2 (left to right) T able 4.5 P ack et Loss Rates for Scenario 2 T raf c class PLR-with A C PLR-without A C High Priority 0.0001 0.0001 Lo w Priority 0.0037 0.003 V oice o w 0.0 0.0 FTP o w 0.0046 0.0038 4.6 Conclusion This section introduced a measurement-based admission control mechanism for the IEEE 802.11e EDCA. This mechanism tak es decisions to admit ne w o ws into the netw ork based on the jitter and currrently being utilized bandwidth metrics. Its performance w as e v aluated and compared with the performance of the IEEE 802.11e EDCA without an y admission control mechanism. The admission control mechanism clearly demonstrated that it can protect e xisting high priority traf c from jitter 59

PAGE 69

and lo w priority traf c from starv ation. It is also e vident that the mechanism does not trade of f netw ork utilization for the purpose of protecting e xisting traf c. 60

PAGE 70

CHAPTER 5 CONCLUSION AND FUTURE W ORK WLANs are being used e v erywhere and hold great promise in the future as points of access to the Internet. Simultaneously the use of streaming applications via the Internet is also increasing tremendously In order to support these applications adequately WLANs ha v e to pro vide support for QoS. The current standard for WLANs, i.e. the IEEE 802.11 consists of tw o main functions, Distrib uted Coordination Function (DCF) and Point Coordination Function (PCF). DCF is a f air protocol and all stations contend and gain access to the channel with equal priority Thus, it cannot pro vide an y le v el of service dif ferentiation or support for QoS. PCF w as designed k eeping in mind time-bounded applications, b ut due to its comple xity is not implemented. In order to o v ercome these shortcomings, the IEEE 802.11e standard is being proposed. The IEEE 802.11e standard consists of tw o functions, namely the Enhanced Distrib uted Channel Access (EDCA) and the Hybrid Coor dination Function (HCF) Controlled Channel Access (HCCA). The IEEE 802.11e EDCA utilizes the Arbitrary Inter Frame Space (AIFS), Contention W indo w (CW) and the Priority F actor (PF) to pro vide support for service dif ferentiation. This thesis presents a performance e v aluation of the IEEE 802.11e EDCA mechanism. Fi v e dif ferent schemes using these parameters are included in the e v aluation and compared to DCF The rst scheme assigns dif ferent Arbitration Inter Frame Space (AIFS) interv als to dif ferent classes to pro vide dif ferentiation. The second scheme assigns dif ferent v alues of the Contention W indo w (CW), and the third scheme assigns a dif ferent Priority F actor (PF) to each class, which only multiplies the Back of f Interv al (BI) included in the current DCF standard by a dif ferent number The fourth and fth schemes utilize the AIFS and CW and AIFS, CW and PF together respecti v ely as a w ay to pro vide better dif ferentiation. The results of the e v aluation indicate that in terms of 61

PAGE 71

a v erage delay and throughput, the single parameter schemes pro vide good service dif ferentiation. But, in terms of service v ariability single parameter schemes are the w orst performing. In order to obtain good throughput and lo w v ariability for streaming applications, multi-parameter schemes ha v e to be used. T o conclude, when single parameter schemes were used, applications suf fer from unacceptable jitter When multi-parameter schemes were used, the v alues for jitter were v ery good b ut the y cause starv ation for the lo w priority class. In order to o v ercome these problems, a measurement-based admission control mechanism is proposed. The mechanism mak es use of a bandwidth allocation scheme, wherein a percentage of the total netw ork bandwidth is dedicated to high priority traf c and the rest is used by lo w priority traf c. All stations k eep note of the amount of bandwidth being used by the high priority and lo w priority traf c. The y also k eep a note of the v alue of jitter for the high priority traf c. Based on these measurements, a decision is tak en as to admit a ne w traf c o w or not. An e v aluation of the admission control mechanism is also presented. The admission control mechanism clearly protects e xisting traf c o ws from the problems of jitter and starv ation and simultaneously impro v es upon the netw ork utilization. It also satises the objecti v es of being completely distrib uted, of not using preemption to implement admission control and does not introduce dramatic changes to the e xisting IEEE 802.11e EDCA mechanism. The proposed admission control mechanism assumes the wireless netw ork to be free from conditions such as ”The Hidden T erminal Problem” and ”The Exposed T erminal Problem”. An e xtention of this mechanism can use transfer of data between wireless stations to share netw ork information and thereby pre v ent the mechanism from being af fected from the abo v e stated problem. Currently the mechanism depends on the netw ork administrator to set the v alues for the bandwidth allocation depending on his kno wledge of the netw ork and applications. The process of bandwidth allocation can be automated, so the v alues for it can be automatically set based on netw ork traf c characteristics at a gi v en point of time. 62

PAGE 72

REFERENCES [1] N. Ramos, D. P anigrahi and S. De y, “Quality of Service Pro visioning in 802.11e Netw orks: Challenges, Approaches, and Future Directions, ” IEEE Network pp. 14–20, July/August 2005. [2] B.P Cro w ,I. W idjaja, J.G. Kim and P .T Sakai, “IEEE 802.11 W ireless Local Area Netw orks, ” IEEE Communications Ma gazine v ol. 35, pp. 116–126, September 1997. [3] “Denition: Quality of Service, ” http://www .ldcircuit.com/libra ryglo ss ary .htm. [4] Q. Ni, “Performance Analysis and Enhancements for IEEE 802.11e W ireless Netw orks, ” IEEE Network pp. 21–27, July/August 2005. [5] A. Grilo and N. Nunes, “Performance Ev aluation of IEEE 802.11e, ” in Pr oceedings of 13th IEEE International Symposium on P er sonal, Indoor and Mobile Radio Communications pp. 511–517. [6] “Denition: SlotT ime, ” http://www .atis.or g/tg2K/slott ime.ht ml. [7] S. Choi et al., “IEEE 802.11e Contention-based Channel Access(EDCF) Performance Ev aluation, ” in Pr oc. IEEE ICC Anchorage, AK, May 2003. [8] S. Mangold et al., “IEEE 802.11e W ireless LAN for Quality of Service, ” in Pr oc. Eur o. W ir eless Florence, Italy February 2002. [9] Q. Ni, L. Romdhani and T T urletti, “A Surv e y of QoS enhancements for IEEE 802.11 W ireless LAN, ” W ile y J W ir eless and Mobile Comp. v ol. 4, pp. 547–66, August 2004. [10] W P attara-Atik om, P Krishnamurthy and S. Banerjee, “Distrib uted Mechanisms for Quality of Service in W ireless LANs, ” IEEE W ir eless Communications pp. 26–34, June 2003. [11] J. Deng and R.S. Chang, “A Priority Scheme for IEEE 802.11 DCF Access Method, ” in IEICE T r ans. Commun. v ol. E82-B, 1999, pp. 96–102. [12] G. Chesson et al., “EDCF Proposed Draft T e xt, ” T ec h. r ep. IEEE wkg doc. 82.11-01/131r1 2001. [13] M. Ben v eniste, “TCMA Proposed Draft T e xt, ” T ec h. r ep. IEEE wkg doc. 82.11-01/117r2 2001. [14] I. Aad and C. Castelluccia, “Dif ferentiation Mechanisms for IEEE 802.11, ” in Pr oc. IEEE INFOCOM April 2001. 63

PAGE 73

[15] M. Barry A. T Campbell and A. V eres, “Distrib uted Control Algorithms for Service Dif fer entiation in W ireless P ack et Netw orks, ” in Pr oc. IEEE INFOCOM 2001, pp. 582–90. [16] K. Kanodia et al., “Distrib uted Multi-hop Scheduling and Medium Access with Delay and Throughput Constraints, ” in Pr oc. A CM MOBICOM 2001 2001, pp. 200–09. [17] H. Zhang, “Service Disciplines for Guaranteed Performance Service in P ack et Switching Netw orks, ” in Pr oc. IEEE October 1995, pp. 1376–96. [18] A. Banchs, A. Perez and X. Perez, “Pro viding Throughput Guarantees in IEEE 802.11 W ireless LAN, ” in Pr oc. WCNC 2002. [19] A. Banchs, X. Perez and A. Perez, “Distrib uted W eighted F air Queueing in 802.11 W ireless LAN, ” in Pr oc. IEEE ICC April 2002, pp. 3121–27. [20] A. Banchs, A. Perez and X. Perez, “Distrib uted F air Scheduling W ireless LAN, ” in Pr oc. A CM MOBICOM 2000, pp. 167–78. [21] W P attara-atik om, S. Banerjee and P Krishnamurthy, “Starv ation Pre v ention and Quality of Service in W ireless LANs, ” in Pr oc. IEEE WPMC October 2002, pp. 1078–82. [22] M. Sridhar and G. V ar ghese, “Ef cient F air Queuing Using Decit Round-robin, ” in IEEE/A CM T r ans. Net. 1996, pp. 375–85. [23] R. Garroppo, S. Giordano, S. Lucetti and L. T a v anti, “ Admission Re gion of Multimedia Ser vices for EDCA in IEEE 802.11e Access Netw orks, ” in Pr oc. Networks 2004 W ien, 2004, pp. 411–416. [24] G. Dimitriadis and F P a vlidou, “Comparati v e Performance Ev aluation of EDCF and EY NPMA Protocols, ” in IEEE Communications Letter s 2004, pp. 42–44. [25] K. Xu, Q. W ang and H. Hassanein, “Performance Analysis of Dif ferentiated QoS Supported by IEEE 802.11e Enhanced Distrib uted Coordination Function (EDCF) in WLAN, ” in Pr oc. of GLOBECOM 2003, pp. 1048–1053. [26] De yun Gao,Jianfei Cai and King Ngi Ngan, “Admission Control in IEEE 802.11e W ireless LANs, ” IEEE Network pp. 6–13, July/August 2005. [27] Y Xiao and H. Li, “Ev aluation of Distrib uted Admission Control for he IEEE 802.11e EDCA, ” IEEE Communications v ol. 42, pp. S20–S24, 2004. [28] H. Li and Y Xiao, “V oice and V ideo T ransmissions with Global Data P arameter Control for the IEEE 802.11e Enhanced Distrib uted Channel Access, ” IEEE T r ans. P ar allel Distrib Sys. v ol. 15, pp. 1041–53, 2004. [29] Y Xiao, H. Li and S. Choi, “Protection and Guarantee for V oice and V ideo T raf c in IEEE 802.11e W ireless LANs, ” in Pr oc. IEEE INFOCOM Hong K ong, 2004, pp. 2152–62. [30] A. V eres et al., “Supporting Service Dif ferentiation in W ireless P ack et Netw orks Using Distrib uted Control, ” in IEEE JSA C 2001, pp. 2081–93. 64

PAGE 74

[31] D. Gu and J. Zhang, “A Ne w Measurement-based Admission Control Method for IEEE 802.11 W ireless Local Area Netwroks, ” in Mitsubishi Elec. Resear c h Lab ., T ec h. r ep. TR-2003-122 October 2003. [32] L. Zhang and S. Zeadally, “HARMONICA: Enhanced QoS Support with Admission Control for IEEE 802.11 Contention-based Access, ” in Pr oc. IEEE RT AS 2004, pp. 64–71. [33] D. Pong and T Moors, “Call Admission Conrol for IEEE 802.11 Conention Access Mechanism, ” in Pr oc. IEEE GLOBECOM 2003, pp. 174–78. [34] A. Banchs, X. Perez-Casta and D. Qiao, “Pro viding Throughput Guarantees in IEEE 802.11e W ireless LANs, ” in Pr oc. 18th International T eler af c Cong September 2003. [35] S. W ietholter and C. Hoene, “Design and V erication of an IEEE 802.11e EDCF Simulation Model in ns-2.26, ” in T ec hnical Report TKN-03-019(http://www .tkn.tuberlin.de/r esear c h/802 .11e ns2/) 2003. [36] K. F all and K. V aradhan, “The NS manual (formerly ns notes and documentation), ” the VINT Pr oject. A collabor ation between r eear c her s at UC Berk ele y LBL, USC/ISI, and Xer ox P ARC 2001. [37] Lucent T echologies, “W a v eLAN/PCMCIA card user' s guide, ” October 1996. [38] W ireless LAN Medium Access Control (MA C) and Physical Layer (PHY) Specications, “The Institute of Electrical and Electronics Engineers, ” July 1997. [39] S. Bab u, M. Labrador and D. Armitage, “Performance Ev aluation of Priority-based Schemes for Service Dif ferentiation in 802.11 WLANs, ” in Pr oc. CITSA 2005 2005. [40] M. Mock, R. Frings, E. Nett and S. T rikaliotis, “Clock Synchronization for W ireless Local Area Netw orks, ” in Pr oc. 12th Eur omicr o Conf Real-time Systems 2002. [41] M. Karam and F T obagi, “Analysis of Delay and Jitter of V oice T raf c o v er the Internet, ” in Pr oc. IEEE INFOCOM 2001. [42] C. Fraleigh, S. Moon, B. L yles, C. Cotton, M. Khan, R. Rock ell, D. Moll, T Seely and C. Diot, “P ack et-Le v el T raf c Measurements from the Sprint IP Backbone, ” IEEE Network Ma gazine pp. 6–16, No v ember 2003. 65