USF Libraries
USF Digital Collections

Software testing testbed for MPEG-4 video traffic over IEEE 802.11b wireless lans

MISSING IMAGE

Material Information

Title:
Software testing testbed for MPEG-4 video traffic over IEEE 802.11b wireless lans
Physical Description:
Book
Language:
English
Creator:
Ikkurthy, Praveen Chiranjeevi
Publisher:
University of South Florida
Place of Publication:
Tampa, Fla.
Publication Date:

Subjects

Subjects / Keywords:
autocorrelation
arta
error-free length
error length
traffic characterization
Dissertations, Academic -- Computer Science -- Masters -- USF   ( lcsh )
Genre:
government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Abstract:
ABSTRACT: Several traffic characterization studies have been performed on wireless LANs with the main objective of realizing good and accurate models of the errors in the wireless channel. These models have been extended to model the effect of errors on higher layer protocols, mainly at the data link layer. However, no prior work has been done to study the application level characteristics of MPEG-4 video traffic over 802.11b wireless networks. In this thesis a traffic characterization study of MPEG-4 video traffic over IEEE 802.11b wireless LANs with the main goal of building a tool for software testing is performed. Using two freely available tools to send and receive real-time streams and collect and analyze traces, MPEG-4 encoded video frames are sent over a 11 Mbps, 802.11b wireless LAN to characterize the errors in the channel and the effect of those errors on the quality of the movie. The results of this traffic characterization were modeled using ARTA (Auto Regressive-To-Anything) software. These modeled characteristics were then used to build a tool that generates synthetic traffic emulating real wireless network scenario. The tool emulates the error length and error free length characteristics of the wireless network for the MPEG-4 video traffic using the corresponding modeled characteristics generated by ARTA. The tool can be used by software developers to test their MPEG-4 streaming media applications without the need of the real infrastructure. The tool can also be trained and extended to support testing of any streaming media applications.
Thesis:
Thesis (M.S.C.S.)--University of South Florida, 2003.
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 Praveen Chiranjeevi Ikkurthy.
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 - 001416920
notis - AJJ4772
usfldc doi - E14-SFE0000058
usfldc handle - e14.58
System ID:
SFS0024754:00001


This item is only available as the following downloads:


Full Text

PAGE 1

SOFTWARE TESTING TESTBED FOR MPEG 4 VIDEO TRAFFIC OVER IEEE 802.11B WIRELESS LANS by PRAVEEN CHIRANJEEVI IKKURTHY A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Computer Science Depart ment of C omputer Science and Engineering College of Engineering University of South Florida Major Professor: Miguel A. Labrador Ph.D. Kenneth J. Christensen, Ph.D. Srinivas Katkoori, Ph.D. Date of Approval: Ju ly 1 1 2003 Keywords: traffic characterization, error length, error free length, arta, autocorrelation Copyright 2003 Praveen Chiranjeevi Ikkurthy

PAGE 2

DEDICATION To my parents and my lovely sister

PAGE 3

ACKNOWLEDGMENTS I would like to express my gratitude to my major professor Dr. Miguel Labrador for giving the opportunity to work under his able guidance He had been my leading light all the while during my Masters and this work wouldnt have been possibl e without Dr. Labradors continuing support and tr ust in me. He helped me take my crucial decisions during my stay at USF and I owe some of my wonderful moments here to him. I would also like to thank Dr. Christensen and Dr. Katkoori for guiding me as com mittee members. They helped me shape this thesis better with their invaluable suggestions. I really appreciate the confidence my parents had in me They were always there for me encouraging and guiding me at tough times.

PAGE 4

i TABLE OF CONTENTS LIST OF TABLES i ii LI ST OF FIGURES i v ABSTRACT vii CHAPTER 1 INTRODUCTION 1 1.1 Background 1 1.2 Motivation 2 1. 3 Contributions of the Thesis 3 1. 4 Organization of this Document 3 CHAPTER 2 BACKGROUND AND LITERATURE REVIEW 4 2.1 IEEE 802.11b Wireless LANs 4 2. 2 MPEG 4 Video 8 2.3 Characterization of N etwork T raffic 9 CHAPTER 3 EXPERIMENTAL SETUP AND METHODOLOGY 1 1 3 .1 Experiment Test Bed 1 1 3.2 RUDE and CRUDE T ools 1 2 3.3 Experimental M ethodology 1 4 CHAPTER 4 EXPERIMENTAL RESULTS 1 6 4.1 Network Level Characte ristics 1 6 4.1.1 Effect of Distance from Access Point 1 6 4.1.2 Effect of MTU Packet Size 1 9 4.1.3 Effect of Network Load 2 4 4.1.4 Effect of Video Quality 3 1

PAGE 5

ii 4.2 Application Lev el Characteristics 3 3 4.2.1 Frame Level Characteristics 3 5 4.2.2 Byte Level Characteristics 37 CHAPTER 5 EXPERIMENTAL MODELING 4 2 5. 1 Modeling of EL and EFL Distributions 4 2 5. 2 M odeling of Characteristics with Autocorrelation 4 4 5. 3 Network Software Testing Test bed 4 5 5. 4 Verification of Emulated Characteristics 48 CHAPTER 6 CONCLUSIONS AND F UTURE WORK 5 2 REFERENCES 5 3

PAGE 6

iii LIST OF TABLES Table 3.1 Description of Hosts Us ed 1 2 Table 4.1 Best and Worst MTU Packet Sizes for Transmission of Jurassic Park Video under Various Network Load Conditions when Mobile Unit is 50ft from Access Point 29 Table 4.2 Best and Worst MTU Packet Sizes for Transmission of Aladdin Video under Various Network Load Conditions when Mobile Unit is 50ft from Access Point 29 Table 4.3 Best and Worst MTU Packet Sizes for Transmission of Star Wars Video under Various Network Load Conditions when Mobile Unit is 50ft from Access Point 30 Table 4. 4 Frames Affected Rates of MPEG 4 Frames by Error Bursts of Length L ess than F ive Packets for the Aladdin Video Trace under No Load Condition at 75ft from Access Point 3 6 Table 4. 5 Average Perc entage of Bytes Affected in I, B and P Frames by Error Lengths of One, Two, Three and Four Packets U sing A ll Four Packet Sizes for the Aladdin Video Trace under No Load Condition at 75ft from Access Point 39 Table 4. 6 Average Packet Size for Different Frames for Jurassic Park Video 39 Table 4. 7 Average Packet Size for Different Frames for Aladdin Video 40 Table 4. 8 Average Packet Size for Different Frames for Star Wars Video 4 0 Table 4. 9 Average Frame S ize for All MTU Packet Size Transmissions fo r All Video Qualities 4 0 Table 4.1 0 Average Percentage of Bytes Affected in the Combinations of I, B and P Frames by Error Lengths of One, Two, Three and Four Packets U sing A ll Four Packet Sizes for Aladdin Video under No Load at 75ft from Access Point 4 1

PAGE 7

iv LIST OF FIGURES Figure 2.1 802.11b S tandard in the OSI (Open Systems I nterconnection) Reference Model 5 Figure 2.2 Basic Service Set (BSS) Mode of Operati on of IEEE 802.11b Wireless LAN 6 Figure 2.3 Extended Service Set (ESS) Mode of O peration of IEEE 802.11b Wireless LAN 7 Figure 2.4 Independent Basic Service Set (IBSS) Mode of Operatio n of IEEE 802.11b Wireless LAN 7 Figure 2.5 GoP S tructure of an MPEG 4 S tream 9 Figure 3.1 Experimental Test Bed 1 1 Figure 4.1 Inverse CDF of Error Length for 1000 Byte Packets for Three Different Videos at 50 and 75ft from Ac cess Point with No Network Load 1 7 Figure 4.2 Inverse CDF of Error Free Length for 1000 Byte Packets for Three Different Videos at 50 and 75ft from Access point with No Network Load 18 Figure 4.3 Inverse CDF of Error Length for Jurassic Park Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 75ft from Access Point with No Network Load 19 Figure 4.4 Inverse CDF of Error Length for Aladdin Video for Packet S izes of 500, 750, 1000 and 1500 Bytes at 50ft from Ac cess Point with No Network Load 2 0 Figure 4.5 Inverse CDF of Error Free Length for Jurassic Park Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 50ft from Access Point with No Network L oad 2 1 Figure 4.6 Inverse CDF of Error Free Length for Jurassic Park Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 75ft from Ac cess Point with No Network Load 2 2 Figure 4.7 Inverse CDF of Error Free Length for Aladdin Video for Packet Siz es of 500, 750, 1000 and 1500 Bytes at 50ft from Access Point with No Network Load 2 2

PAGE 8

v Figure 4.8 Inverse CDF of Error Free Length for Aladdin Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 75ft from Access Point with No Network Load 2 3 Figure 4.9 Inverse CDF of Error Free Length for Star Wars Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 50 ft from Access Point with No Network Load 2 3 Figure 4.10 Inverse CDF of Error Free Length for Star Wars Video for Packet Sizes of 500 750, 1000 and 1500 Bytes at 75ft from Access Point with No Network Load 2 4 Figure 4.11 Inverse CDF of Error Free Length for Jurassic Park Video for Different Network Load Conditions of 750 Byte Packet Transmissions when Mobile Unit is at 50ft Away fr om Access Point 2 5 Figure 4.12 Inverse CDF of Error Free Length for Aladdin Video for Different Network Load Conditions of 750 Byte Packet Transmissions when Mobile Unit is at 50ft Away from Access Point 2 6 Figure 4.13 Inverse CDF of Error Length f or Star Wars Video for Different Network Load Conditions of 500 Byte Packet Transmissions when Mobile Unit is at 50ft Away from Access Point 27 Figure 4.14 Inverse CDF of Error Length for Jurassic Park Video for Different Network Load Conditions of 75 0 Byte Packet Transmissions when Mobile Unit is at 50ft Away from Access Point 27 Figure 4.15 Inverse CDF of Error Length for Aladdin Video for Different MTU Packet Transmissions for 4 Mbps Network Load when Mobile Unit is at 50ft Away from Access Poin t 28 Figure 4.16 Inverse CDF of Error Free Length for 1000 Byte MTU Transmissions of All the Video Qualities for 2 Mbps Network Load when Mobile Unit is at 50ft Away from Access Point 3 2 Figure 4.17 Inverse CDF of Error Free Length for 1000 Byte MT U Transmissions of All the Video Qualities for 4 Mbps Network Load when Mobile Unit is at 50ft Away from Access Point 3 2 Figure 4.18 Inverse CDF of Error Free Length for 1500 Byte MTU Transmissions of All the Video Qualities for 6 Mbps Network Load whe n Mobile Unit is at 50ft Away from Access Point 3 3 Figure 4.19 Error Length Distribution Using Jurassic Park Video with a Packet Size of 1000 Bytes being 50ft Away from Access Point under No Load Condition 3 4 Figure 4.20 Percentage of MPEG 4 Frame s of Aladdin Video Affected by Error Bursts Made of One Packet, when the Access Point is 75ft Away and Packets are 1 0 00 Bytes Long under No Network Load Condition 35

PAGE 9

v i Figure 4.21 Error Length Distribution Using Aladdin Video with a Packet Size of 1000 By tes being 75ft Away from Access Point under No Load Condition 37 Figure 4.22 Mean Amount of Bytes Affected in I, B and P Frames by Error Bursts Made of One Packet, when the Access Point is 75ft Away and 1000 Bytes MTU for Aladdin Video under No Network Load 38 Figure 5.1 Inverse CDF of Error Free Length for 500 Byte MTU Packets for Aladdin Video at 50ft from Access Point under No Network Load 4 3 Figure 5.2 Autocorrelation Function for First 100 Lags for the Error Free Length Characteristic Given in Figure 5.1 43 Figure 5.3 Modeling of Inverse CDF of Error Free Length Using ARTA Process for 500 Byte Packets for Aladdin Video at 50ft from Access Point and No Network Load 4 4 Figure 5.4 Matching of Autocorrelation Function for First 100 Lags fo r the Characteristic Given in Figure 5.3 by the ARTA Process 45 Figure 5.5 Working of the Network Software Testing Test Bed 46 Figure 5.6 Network Software Testing Test Bed S creenshot 46 Figure 5.7 Initialization Process of Video Traffic 47 Figure 5.8 Algorithm for Emulating Wireless Network by NSTT Using Modeled Characteristics Generated by ARTAGEN 48 Figure 5.9 Matching of Inverse CDF of the Actual Error Length Characteristic by the Emulated Characteristic for 1000 Byte Packets for Jurassic Pa rk Video at 75ft and No Network Load 49 Figure 5.10 Matching of Autocorrelation Function for First 100 Lags for the Actual Characteristic Given in Figure 5.9 by the Emulated Characteristic 49 Figure 5.11 Matching of Inverse CDF of the Actual Error L ength Characteristic by the Emulated Characteristic for 1000 Byte Packets for Aladdin Video at 75ft from Access Point and No Network Load 50 Figure 5.12 Matching of Autocorrelation Function for First 100 Lags for the Actual Characteristic Given in Figu re 5.11 by the Emulated Characterist ic 5 0 Figure 5.13 Matching of Inverse CDF of the Actual Error Length Characteristic by the Emulated Characteristic for 1000 Byte Packets for Star Wars Video at 75ft and No Network Load 5 1 Figure 5.14 Matching of A utocorrelation Function for First 100 Lags for the Actual Characteristic Given in Figure 5.13 by the Emulated Characteristic 5 1

PAGE 10

vii SOFTWARE TESTING TES TBED FOR MPEG 4 VIDEO TRAFFIC OVER IEEE 802.11B WIRELESS LAN S Praveen C. Ikkurthy ABSTRACT Severa l traffic characterization studies have been performed on wireless LANs with the main objective of realizing good and accurate models of the errors in the wireless channel. These models have been extended to model the effect of errors on higher layer proto cols, mainly at the data link layer. However, no prior work has been done to study the application level characteristics of MPEG 4 video traffic over 802.11b wireless networks. In this thesis a traffic characterization study of MPEG 4 video traffic over IE EE 802.11b wireless LANs with the main goal of building a tool for software testing is performed. Using two freely available tools to send and receive real time streams and collect and analyze traces, MPEG 4 encoded video frames are sent over a 11 Mbps, 802.11b wireless LAN to characterize the errors in the channel and the effect of those errors on the quality of the movie. The results of this traffic characterization were modeled using ARTA (Auto Regressive To Anything) software. These modeled characteri stics were then used to build a tool that generates synthetic traffic emulating real wireless network scenario. The tool emulates the error length and error free length characteristics of the wireless network for the MPEG 4 video traffic using the correspo nding modeled characteristics generated by ARTA. The tool can be used by software developers to test their MPEG 4 streaming media applications without the need of the real infrastructure. The tool can also be trained and extended to support testing of any streaming media applications.

PAGE 11

1 CHAPTER 1 INTRODUCTION 1.1 Background Wireless LANs have become very popular in many scenarios because they are very simple, convenient and cheap. Nowadays, we find wireless LANs in businesses, museums, libraries, and factories and in many other places where this technology is the only solution for cabling problems, cost, or flexibility reasons. The cost of wireless LANs have decreased so drastically that many households, hotels, apartment buildings, airports and the like have already installed such net works. An example of a common configuration at home is the connection of the wireless access point to a high speed cable modem or ADSL connection. Normally, wireless networks are attached to the wired infrastructure so that wireless users are also connect ed to the corporate network and/or the Internet. Wireless users will use the same business mission critical applications and all other types of a pplications they are used for running over wired networks. Furthermore, wireless users expect their application s to run as if they were wired connected. However, wireless networks are different, and several factors make them more challenging to satisfy those users' demands. Wireless networks are lossy in nature and challenged by r adio propagation and mobility fact o rs, therefore, applications do no t usually perform as well as in wired environments. As a result, it is important to have a deeper knowledge about the issues limitations, and the performance characteristics of these networks Network applications are dif ferent in nature and affected differently depending on the type of communication problems. For instance, data applications are very sensitive to packet losses but not that sensitive to packet delays. On the other hand, delay is very important for voice and video applications but they can tolerate some losses [ 1 ]. In the current work, traffic characterization study of video traffic over 802.11b wireless LANs at the packet and application levels with the aim of building a tool for software testing is presente d Even though other studies have characterized the errors in wireless LANs before, they have all used older wireless LANs technology such as the known Wave LAN [ 25 ] at 2 Mbps. Investigation and characterization of the effect of the wireless channel errors on video applications is performed to give a good idea of how much harm those errors cause to applications, in particular on MPEG 4 video encoded

PAGE 12

2 applications. Focus is on video applications because of the stringent Quality of Service ( QoS ) requirements th at they impose on the network, in terms of bandwidth, delay and losses. MPEG 4 video was used in the experiments as streaming applications are used in many domains in Department of Defense (DoD). A few of the applications of streaming video in DoD are: mis sion critical video, training video, IP telephony and video conference to name a few. 1.2 Motivation The ever increasing usage of wireless networks in everyday life is easily perceivable. Only last year, more that 14.4 million Wireless LAN (WVLAN) devic es, mostly IEEE 802.11b, were sold [ 2 ]. Furthermore, plans to rol l out wireless Internet access points within 5 minutes of anyone in the largest 50 US metropolitan areas by 2004 were announced recently by AT& T and IBM [ 3 ]. M PEG 4 video has become popular for transmission of real time v ideo traffic over the Internet. Several traffic characterization studies have been performed on wireless LANs with the main objective of realizing good and accurate models, and to model the errors in the wireless channel. Th ese models have been extended to model the effect on higher layer protocols, mainly at the data link layer. However, no prior work has been done to study the application level characteristics of MPEG 4 video traffic over 802.11b wireless networks. In thi s thesis a traffic characterization study of MPEG 4 video traffic over IEEE 802.11b wireless LANs is performed The main motivation of this thesis is to use this traffic characterization study to build a tool for software testing of streaming MPEG 4 videos This tool generates synthetic traffic emulating real wireless network scenario. The tool can be used by software developers to test their MPEG 4 streaming media applications without the need of the real infrastructure. T he tool eliminates the requirement of actual wireless network to test the MPEG 4 streaming media application over wireless network. This is done by the emulation of the wireless network loss characteristics by the tool. 1.3 Contributions of the T hesis The main contributions of the thesis work are listed below. A thorough traffic characterization study of MPEG 4 video traffic over IEEE 802.11b w ireless LANs. A t ool for generating synthetic MPEG 4 video traffic over IEEE 802.11b wireless LAN is developed. This tool takes the different param eters such as the load condition of the wireless network, distance from access point etc., and the actual video to be subject the above

PAGE 13

3 conditions. The tool generates the video output trace of the given video as though it was subjected to the network condi tions as specified in the input parameters. 1. 4 Organization of this D ocument This chapter provide d a brief outline and the motivation of this thesis work. Chapter 2 gives a broad overview of the past and current work related to this thesis. This is us eful in the perspective to have a very good understanding of the utility of the current work. Chapter 3 details the experimental setup and t he set of experiments performed Chapter 4 evaluates the results obtained during the various experiments. Chapter 5 discusses the Modeling of the characteristics analyzed in the thesis. It also presents the implementation details of the Network Software Testing Test Bed developed as a part of the thesis work. Chapter 7 concludes the thesis and gives the future directio ns in which research can be pursued in this topic.

PAGE 14

4 CHAPTER 2 BACKGROUND AND LITERATURE REVIEW In this Chapter, a brief background on the IEEE 802.11b Wireless LAN standard and the MPEG 4 video coding standard is given. This sets the stage for the choi ce made for the network and the video compression standard used in the experiments performed in this thesis. Finally, the literature review of the traffic characterization study and traffic modeling techniques employed is provided. 2.1 IEEE 802.11b Wirel ess LANs The IEEE 802.11b Wireless Local Area Network (WLAN) operates in the 2.4GHz license free Radio frequency (RF) band providing a data rate up to 11Mbps. The IEEE 802.11b standard is devoted to wireless LANs, developing a MAC (Medium Access Control Su b layer of the data link layer) layer protocol and physical medium specification (Figure 2.1). The IEEE 802.11 Physical Layer uses any one of the following three modulation techniques: Direct Sequence Spread Spectrum (DSSS) (operating in 2.4 GHz band at 1 Mbps and 2 Mbps), Frequency Hopping Spread Spectrum (FHSS) (operating 2.4 GHz band at 1 Mbps and 2 Mbps) and Infrared (operating at 1 Mbps and 2 Mbps). The IEEE 802.11b Wireless LANs operate in DSSS mode. A good description of IEEE 802.11 standard is give n in [29].

PAGE 15

5 Figure 2.1 IEEE 802.11b S tandard in t he OSI (Open Systems Interconnection) Reference M odel The IEEE 802.11 MAC layer addresses three areas: reliable data recovery, access control and security [28]. The IEEE 802.11 standard supports two MAC schemes: Distributed Coordination Function (DCF) supporting best effort delivery of data and Point Coordination Function (PCF) designed for delay sensitive traffic. The DCF makes use of CSMA/CA (Carrier Sense Multiple Access/Collision Avoidance) algorithm to control the access of the MAC layer for contending traffic. If a station needs to transmit a MAC frame, it listens to the channel and waits for it to be idle. Once it sees the channel idle, it can transmit its frames; otherwise, it needs to wait until the current transmission ends. To avoid the collisions and the known hidden terminal problem in wireless networks IEEE 802.11 standard improves channel reliability by including RTS and CTS signals. For complete details of the CSMA/CA algorithm, refer to [2 7]. PCF is an additional access method along with DCF. It assigns the access point as the Point Coordinator (PC) or poll master. The function of the PC is polling the stations and enabling them to transmit without contending for the channel, which is the c ase in DCF scheme. A basic wireless LAN consists of a wireless node, typically a laptop with wireless card and an access point (AP), which acts as the bridge between the wireless node and the external network (called Distribution System) which typically i s a wired network. The IEEE 802.11b defines two modes of operation:

PAGE 16

6 Infrastructure Mode: In this mode, there is at least one access point connected to the Distribution System. Here the Wireless Stations (STAs) communicate directly with the access point. Th is mode of operation involving only one access point is called Basic Service Set (BSS) as shown in Figure 2.2. When two or more access points communicate with each other through the Distribution System, we have the case of Extended Service Set (ESS) as sho wn in Figure 2.3. Figure 2.2 Basic Service Set (BSS) Mode of Operatio n of IEEE 802.11b Wireless LAN Ad Hoc Mode: In this mode of operation, the wireless stations communicate with each other directly. The Ad Hoc mode of operation is also called Independ ent Basic Service Set (IBSS) and the corresponding scenario is shown in Figure 2.4.

PAGE 17

7 Figure 2.3 Extended Service Set (ESS) Mode of Operatio n of IEEE 802.11b Wireless LAN Figure 2.4 Independent Basic Service Set (IBSS) Mode of Operatio n of IEEE 802.1 1b Wireless LAN

PAGE 18

8 2.2 MPEG 4 Video MPEG stands for Moving Pictures Experts Group, an International Standards Organization group formed to standardize audio and video compression. Three MPEG standards exist, MPEG 1, MPEG 2 and MPEG 4. MPEG 1 was designed w ith the idea of storing moving pictures and audio on Compact Disks at a low bit rate. MPEG 2 has become very popular because it has been used as the compression mechanism for Digital Video Broadcasting (DVB) and Digital Video Disc (DVD). [5] gives a good s tatistical analysis of MPEG video traffic over ATM networks. In contrast to MPEG 1 and MPEG 2, MPEG 4 is object based. MPEG 4 objects are part of a scene which can be accessed or manipulated independently. MPEG 4 achieves higher compression ratios than MP EG 2 and has better coding tools, making it more suitable for the Internet and wireless delivery of applications [4]. MPEG 4 is chosen as the video standard for transmission of video traffic as it is more and more being deployed as the standard to transmit real time video and multimedia applications over the Internet. The MPEG standard defines four distinct pictures encoding: Intra coded Picture (I frame), Predictive Coded Picture (P frame), Bidirectional Predictive Coded Picture (B frame) and DC Coded P icture (D frame). A brief description of the individual frames is given below. I frames (intra coded frames) are self contained and coded using a Discrete Cosine Transform (DCT) based technique similar to JPEG. I frames are used as random access points in MPEG streams, and they give the lowest compression ratios within MPEG. P frames (predicted frames) are forward predictive coded using motion compensation prediction in reference to a previous I frame or another P frame. The compression ratio of the P fram es is significantly higher than of I frames. B fram es (bi directional or interpolated frames) are bi directionally coded using two reference frames, a past and a future frame (which can be I or P frames). Bidirectional or interpolated coding provides the h ighest amount of compression. D frames (DC Coded frame) store the DC component of each DCT block. The I, B and P frames are arranged in a periodic pattern known as a Group of Pictures (GOP). Figure 2.5 shows the GOP of MPEG 4 video used in the experiments performed, and relationships among frames. The MPEG 4 Group of Pictures is made of 12 frames in the following order: IBBPBBPBBPBB. The first two B frames (2 and 3) are bi directionally coded using the past frame (I frame 1) and the future frame P (frame 4 ). Therefore, each B picture is encoded based on the previous and following I and/or P frames. P frames on the other hand are dependent on previous I or P frames. Due to these dependencies the decoding order will be

PAGE 19

9 Figure 2.5 GoP S tructure of an MPEG 4 S tream [4] different from the encoding order. The P frame 4 must be decoded before B frames 2 and 3, and I frame 1 (the last I frame) before B frames 11 and 12. If the MPEG 4 sequence is transmitted over the network, the actual transmission order should be {1, 4, 2, 3, 7, 5, 6, 10, 8, 9, 1, 11, and 12}. From the graph and the explanation above about the dependencies, it is easy to conclude that I frames are the most important ones since they contain the actual video content and all other pictures are err or coded based on the I frames. 2.3 Characterization of N etwork T raffic Traffic characterization is a well studied field, in particular over wired networks. [6] gives a good analysis of packet dynamics such as out of order delivery, packet corruption, pat tern of packet losses and the distribution of the duration of the losses and packet transit delays. The effects of route asymmetries in Internet routing are studied, taking end to end routing behaviors into consideration. In [8], the authors studied traces to show that the distribution of the packet inter arrival times of Ethernet traffic is not exponentially distributed and that it exhibits self similarity. Prior to this work, network traffic was modeled as a Poisson process. In [7], the authors evaluated packet traces in wide area networks to investigate whether the arrival process of FTP connections and TELNET packet arrivals corresponded to the commonly used Poisson arrival process. They find that Poisson modeling is applicable only for user session arri vals such as rlogin, and not for FTP and TELNET packet arrivals. New models are required for modeling the burstiness and the possible self similarity of wide area traffic. A good survey of widely used traffic characterization techniques is given in [9]. V ideo traffic has in built self similarity as it can be clearly understood that there is only a slight variation in consecutive video frames. In other words, Variable Bit Rate (VBR) video has

PAGE 20

10 high correlation on large time scales. Various approaches to mode ling video traffic have been done. [16 gives a statistical analysis of a VBR video sample. The paper finds that a video sequence exhibits autocorrelation which is long range dependent and hence can be modeled using a self similar process. The paper present s a source model for generating VBR video traffic synthetically. TES (Transform Expand Sample) models have also been applied to modeling VBR video traffic. A couple of good examples of these works can be seen in [21] and [23]. A good overview of TES and it s modeling technique is given in [22]. Video traffic has been studied over 10 and 100 Mbps Ethernet LANs characterizing the quality of the video in term of glitches [18]. In wireless networks, [11] characterizes the throughput, average delay, jitter, fram e error rate, IP loss rate and other parameters for UDP traffic over IEEE 802.11b wireless LANs. In [10], the error length and error free length distributions were collected, analyzed and validated, characterizing the loss behavior of the WaveLAN. The stud y shows average packet error rates of 2 to 3% and correlated errors. However, neither study relates the effects of the errors to the applications in current wireless networks. Here, in the current thesis work, insights about how those errors affect the qua lity of the MPEG 4 encoded video are provided. In addition, a trace based approach to characterize the errors is given at the packet level to verify those results on the new generation of wireless LANs, the 11 Mbps IEEE 802.11b standard. In [24], character ization of the error length and error free length characteristics for fixed packet size transmissions of MPEG 4 video traffic over wireless LANs is performed. In the thesis, this prior work is repeated but with Variable size MTU packet transmissions. This is a more reliable assumption as this approach utilizes lesser network bandwidth than the fixed packet transmissions. Therefore, results from fixed size packet transmissions will not be included in this thesis. In the current work, the observed network cha racteristics are modeled using the ARTA (Auto Regressive To Anything) [19] modeling process to generate synthetic traffic with the same characteristics of the observed real traffic

PAGE 21

11 CHAPTER 3 EXPERIMENTAL SETUP AND METHODOLOGY 3.1 Experiment Test Bed The test bed used in the experiments is shown in Figure 3.1. Three hosts are employed for the experimental setup. The first host ( Techdb ) was connected to a Fast Ethernet switched network and the mobile host ( Ap manager2 ) is a laptop in a wireless LAN. The third host ( Giga2 ) was connected to the Gigabit Ethernet and was used for load generation. The wireless LAN utilized corresponds to the IEEE 802.11b standard. It runs at up to 11 Mbps in the license free 2.4 GHz band and implements the Carrier Sense Multi ple Access with Collision Avoidance (CSMA/CA) Medium Access Control (MAC) protocol [20]. In a closed office environment, like the one used here, the radio characteristics limit the transmission rate of 11 Mbps to a range of up to 80 ft. F igure 3.1 Expe rimental Test B ed

PAGE 22

12 An isolated Ethernet segment was used to make sure no packet losses occur in the wired portion of the network. Table 3.1 shows the names and characteristics of each host used for the experimental setup. Table 3.1 Description of H osts U s ed Name Processor RAM Interface Type Techdb Pentium 200 Mhz 64 MB ISA 100Base T Giga2 Pentium III 500 Mhz 128 MB ISA Gigabit Ap manager2 Pentium III m 866 Mhz 256 MB PCMCIA 2.4 GHz DSSS All the hosts ran the Linux OS, kernel version 2.4.3. One hour frame length MPEG 4 video traces of three different videos of Star Wars, Aladdin, and Jurassic Park movies were used for all our tests [13]. This is similar to the approach used in [30], [17], and [12] for performing measurement and evaluation of wireless network characteristics. RUDE (Real time UDP Data Emitter) and CRUDE (Collector for RUDE) applications [14] were installed in the video traffic generator and mobile computer, respectively. The mode of operation and the modifications done to these tools ar e explained in more detail next. 3.2 RUDE and CRUDE T ools RUDE transmits UDP data packets based on a script file. CRUDE is the receiver and logging utility of the traffic flows generated by RUDE [15]. RUDE supports two different types of network traffic flows: Constant bit rate (CBR) traffic and trace based traffic. RUDE takes packet length traces of the following type: . The trace based flow was modified for suiting the purpose of MPEG 4 transmission. This traffic type supports the transmission of frame length traces of MPEG video traffic. The rate of transmission of the UDP packets was controlled based on the frame generation rate of MPEG 4 video (25 frames/sec). The ra te of transmission field can be modified by the user. Hence this flow type can be used for trace based transmission of any MPEG traffic. The different video characteristics supported by the modified RUDE are:

PAGE 23

13 Frame Type: The type of MPEG 4 frame transmi tted, either I, B or P. End of frame indicator: Indicator for whether the current packet is the last packet of the current frame. Packet size: size of transmitted MPEG packet Start of Macro block indicator End of macro block indicator The last two characte ristics are for future use. They have been added to study the characteristics of the video at macro block level of MPEG frame. The input trace file given to the modified RUDE has the following format: . In addition to the MPEG 4 video characteristics, RUDE also transmits the following characteristics: Packet sequence number Transmission time (In seconds, micro seconds) Source IP address The logging capability of CRUDE was also enhanced to collect the characteristics of the wireless channel. They are: Wireless Link Quality Signal Level Noise Level These characteristics give a good idea of the wireless network conditions under whi ch the experiments were performed. In addition to collecting the characteristics transmitted by RUDE, CRUDE collects the following characteristics: Destination IP address Receive Time (In microseconds) With the transmission and receive time, the delay and jitter characteristics of the application are collected. Based on the sequence numbers of the packets, the error characteristics of the MPEG video traffic are analyzed.

PAGE 24

14 3.3 Experimental Methodology The experimental methodology involved sessions of transmi ssion and collection of frame length traces of one hour MPEG 4 movie video. The transmission was done by the traffic generator (RUDE). The collection of this video transmission was done by the traffic collector (CRUDE). The required network load in the wir eless channel was generated by the second host in the Fast Ethernet. With RUDE, the MPEG 4 video trace was loaded and transmitted over the network. On the mobile computer CRUDE receives the packets transmitted by RUDE and stores a trace for each transmissi on. No modifications to the wireless interface of the mobile hosts were made. From the trace data, the traffic was analyze by network load, packet size, distance from access point and video quality to get a quantitative understanding of the breakdown of t his video traffic for these key parameters of interest. These traces can be used by other applications such as gnu plot to process and plot the results. A session was initiated when the traffic generator starts a new transmission or when the traffic colle ctor signals the end of previous transmission for the traffic generator to initiate a new session. Care was taken to ensure that the network conditions during the transmission are unchanged. At the end of the one hour session, the traffic generator sends a termination request to the collector, thereby successfully completing the session. At termination, the traffic collector gives summary statistics of the session. The fields of each entry of the trace file collected include: Packet sequence number, transmi ssion time, receive time, source IP address, destination IP address, actual packet size, size of video data in the packet, type of video frame, end of frame indicator and signal and noise levels in the wireless channel. The whole process of multiple sessio ns has been automated by having synchronized cron jobs at the traffic generator and traffic collector. The experiments performed involved thorough analysis of MPEG 4 video traces with different video qualities subjected to different MTU packet sizes, at d ifferent distances from access point and under different load conditions. The characteristics of MPEG 4 video over wireless network with different scenarios are analyzed using the following parameters: Video Quality : Three different types of videos were us ed. They are high network intensive, medium network intensive and low network intensive. That is they represent the amount of bandwidth utilization of the network by the different videos. The videos used are Jurassic Park (Highest bandwidth utilizing), Ala ddin (Medium bandwidth) and Star wars (Low bandwidth utilizing). A good description of the statistics of the MPEG 4 video trace files used in the current work can be obtained from [31].

PAGE 25

15 Packet Size: Four different packet MTU (Maximum Transmission Unit) s izes of 500, 750, 1000 and 1500 bytes were used; i.e. the maximum size packets transmitted was varied. Four different packet sizes (MTU sizes) were used to assess the best packet size for transmission of these videos of different qualities over wireless ne tworks. Distance from Access point : Two different distances (50 ft and 75 ft) from the access point were chosen. The first case can be taken as the average case of distance of a mobile host from an access point whereas the 75 ft case represents the interes ting border region case of the access point [25]. Network load condition : The experiments were run at different load conditions to analyze the effect of network load on the different video characteristics. The different network load conditions considered f or the experiments are: 2 Mbps, 4 Mbps, 6 Mbps, 8 Mbps load and No Load. The synthetic load generation was done by a simple C program which makes use of multiple threads to generate the required load. Here, each thread generates a load of 0.4 Mbps. i. e., packets are generated by each thread with a constant bit rate (one 1000 byte packet is generated every .01 seconds). So, to generate 6 Mbps load, 15 load threads need to be initiated at the load generator towards the traffic collector. The experiments were performed with the different load conditions only when the mobile unit was at a distance of 50 ft from the access point because of interest in the average usage case.

PAGE 26

16 CHAPTER 4 EXPERIMENTAL RESULTS In this chapter the experimental results and the analysis of the observed characteristics are presented. The network level and application level characteristics were evaluated. At network level, the error characteristics of the MPEG 4 video transmitted were evaluated. The Error Length (EL) and Error Fre e Length EFL were evaluated and characterized at the network level. Error Length represents the burst size of consecutive packets in error (or consecutive packets lost). On similar terms, Error Free Length represents burst of consecutive packets which are received without errors. The inverse CDF (Cumulative Distribution Function) of these characteristics were evaluated. The inverse CDF was used for the analysis as it can be used for modeling these characteristics. At the application level, the effect of pa cket loss on individual frames of MPEG 4 video given by the amount of data lost from each frame type is shown. 4.1 Network Level Characteristics In this section, the network level results for the video traffic generated at different distances from access point, for different MTU packet size transmissions, for different video qualities and network load conditions are presented. The distribution of EL and Error Free Length at network level for these scenarios were plotted and evaluated. 4.1.1 Effect of Di stance from Access Point Two different distances for were considered for the experiments. Distances of 50ft and 75ft from the access point were chosen, representing the average usage case and the interesting border condition respectively. It is clearly un derstood that the nearer the mobile host is to the access point, the better the signal strength and better the error characteristics of the video. This can be clearly seen from the Figures 4.1 and 4.2. In Figure 4.1 the inverse of the Cumulative Density F unction (1 CDF) of the Error Length (P [EL] > x) in number of consecutive packets (x) when the mobile unit is 50ft and 75ft away from the access point for 1000 byte packet transmissions of the three different quality videos when there is no external load in the wireless

PAGE 27

17 channel are presented. Figure 4.2 presents the Inverse CDF of the Error Free Length in number of packets for the same scenarios. Figure 4.1 Inverse CDF of Error Length for 1000 Byte Packets for Three Different Videos at 50 and 75ft fro m Ac cess Point with No Network Load In the case of the Error Length distribution, these curves say that as the distance from the access point decreases, the probability of having a larger error bursts decreases. For instance, the probability of having a b urst greater than 5 packets for Jurassic video is around 0.19 for 75ft case while it is reduced to 0.07 for the 50ft case. Star Wars gives the best case scenario here, as all the packet losses are less than or equal to 3 for the 50ft case while the probabi lity of having an error burst greater than 5 errors is only 0.001 for the 75ft case. This kind of varied behavior for different videos can be explained by the number of packets comprising the frames in these different cases. In the high quality video, Jura ssic Park, the average number of packets making up each frame is 4.6 whereas, in case of low quality video Star Wars, the average number of packets in each frame is 1.1. Hence, due to fixed frame generation rate of MPEG 4, the number of packets in the chan nel in case of Star Wars is much smaller than that of Jurassic Park and hence the smaller the largest error burst. With the similar reasoning it can be seen that Aladdin (2.8 packets/ frame), being medium quality video, has characteristics in the middle re gion.

PAGE 28

18 The Error Free Length distribution curves have the following trend: as the distance from the access point increases, the probability of having a higher size error free burst decreases. The probability of having an error free burst greater than 100 p ackets for Jurassic video is around 35% for 50ft case while it is reduced to 5% for 75ft case. On the other hand, the corresponding characteristics are 10% and 2% for 50ft and 75ft cases of Star Wars video. This shows that as the video quality decreases, t he Error Free Length characteristics of the video traffic worsen. This can be explained as follows: we know that the number of video traffic packets in the network decreases with the decrease in video quality (due to MPEG video transmission rate of 25frame s/ sec). Hence, given a window of time, the probability of having a large error free burst decreases with decrease in video quality. Hence, it can be seen that as the video quality decreases, the Error Free Length characteristics worsen. All our Error Free Length and Error Length curves for Jurassic Park video look very similar to the ones presented in [10] in two aspects. First, they also present three different segments to be fitted individually, two straight lines and one concave curve, saying that a sim ple geometric model will not capture the real loss behavior. Second, the values pretty much match, in other words, the probability of finding an error burst or error free burst of 4 packets is quite similar. Figure 4.2 Inverse CDF of Error Free Length fo r 1000 B yte P ackets for T hree D ifferent V ideos at 50 and 75ft from Access Point with No Network L oad

PAGE 29

19 4.1.2 Effect of Packet Size Four different packet sizes were selected for the transmission of MPEG 4 video to assess an optimal packet size for each video quality in a general scenario. In Figures 4.3 the Inverse CDF the Error Length for Jurassic Park video when the mobile unit is 75ft away from the access point under No Load condition using packet sizes of 500, 750, 1000 and 1500 bytes, respectively is pres ented. In the case of the Error Length distribution, these curves say that as the packet size is decreased the probability of having a burst of errors of the same size decreases. For instance, the probability of having a burst greater than 10 errors with a packet size of 500 bytes is around 0.15 while it is reduced to 0.009 if we used packets of 1500 bytes. In the worst case of using packets of 500 bytes, around 50% of the bursts are equal to or less than 4 packets. This number increases to around 90% when using 1500 bytes packets. Similar behavior is observed with Aladdin videos for both the distances at No Load condition. In general it is found that 1500 byte packets give the best Error Length characteristics followed by 1000 bytes and 750 bytes with 500 b ytes being the worst. Figure 4.4 clearly illustrates this stratification. Figure 4.3 Inverse CDF of Error Length for Jurassic Park Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 75ft from Access Point with No Net work Load

PAGE 30

20 Figure 4.4 Inverse CDF of Error Length for Aladdin Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 50ft from Access Point with No Network L oad The error characteristics of Star Wars for different packet sizes were observed to be alm ost the same. This can be attributed to the fact that the average packet size for Star Wars for different packet size traces is almost the same. Figure 4.5 gives the Error Free Length distribution for the Jurassic Park Video at 50ft from access point and at no network load condition. For EFL < 100 packets, there was not much difference if packets are 1000 bytes or less. For EFL > 100 packets, it can be clearly seen that as packet length increases, Error Free Length characteristic worsens. All through the p lot, it was also observed that 1500 byte packets give the worst Error Free Length characteristics. Hence, as expected, as the packet size increases the Error Free Length characteristic worsens. But in the 75ft case (Figure 4.6), it was seen that all the pa cket sizes gave almost similar Error Free Length characteristics. As the wireless network conditions worsen the video traffic experiences losses more or less in the same manner irrespective of the packet sizes. Hence the clear pattern in Error Free Length characteristics in this case as was observed in the 50ft case was not seen. The Error Free Length characteristics of Aladdin video are given in Figures 4.7 and 4.8. For the 50ft case, 500 byte packets give best Error Free Length characteristics followed by 750 byte, 1000 byte and 1500 byte packets respectively. A similar Error Free Length characteristic was observed for the

PAGE 31

21 75ft case. The difference in the Error Free Length characteristics of Aladdin and Jurassic Park for the 75 fl case can be attributed to the difference in bandwidth utilization by these two different videos. The bandwidth utilization of Aladdin is much smaller than that of Jurassic Park. Hence, even in worse network conditions, we can expect to see the expected trend in Error Free Length c haracteristics. The Error Free Length characteristics of Star Wars are given in Figures 4.9 and 4.10. In Figure 4.9, a clear trend is not seen. This is because the average packet size in all these transmissions is almost the same. With the worsening of ne twork conditions (Figure 4.10), slightly different characteristics by different packet size transmissions can be expected. Figure 4.5 Inverse CDF of Error Free Length for Jurassic Park Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 50ft fro m Access Point with No Network Load

PAGE 32

22 Figure 4.6 Inverse CDF of Error Free Length for Jurassic Park Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 75ft from Access Point with No Network Load Figure 4.7 Inverse CDF of Error Free Length for A laddin Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 50ft from Access Point with No Network Load

PAGE 33

23 Figure 4.8 Inverse CDF of Error Free Length for Aladdin Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 75ft from Access Point with N o Network Load Figure 4.9 Inverse CDF of Error Free Length for Star Wars Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 50ft from Access Point with No Net work Load

PAGE 34

24 Figure 4.10 Inverse CDF of Error Free Length for Star Wars Video for Packet Sizes of 500, 750, 1000 and 1500 Bytes at 75ft from Ac cess Point with No Network Load 4.1.3 Effect of Network Load Network loads of 2 Mbps, 4 Mbps, 6 Mbps and 8 Mbps were introduced in the wireless channel to study the effect of network load on the vari ous network and application level characteristics. It is observed earlier that the inverse CDFs of Error Free Length characteristics under no network load follow have three stage curves as seen in [10]. Similar behavior was not seen when the network is loa ded. This can be clearly observed from the Figure 4.11. Here, it can be seen that the curve gradually becomes a straight line as the network load is increased. This trend was also observed for the Error Free Length characteristics of the other videos as w ell for all packet size transmissions. As the network load increases, the slope of the characteristic curve is expected to decrease. In other words, as the network load is increased the probability of having a large error free burst should decrease. But f rom Figure 4.11, it is seen that the slope of the Error Free Length characteristic starts to decrease only from a network load of 4 Mbps. The network utilization by the network load is minimal (only 18%) in 2 Mbps condition. Moreover, the probability of ha ving a video traffic packet being lost in the wireless channel in 2 Mbps condition is less than that can be seen in No Load condition. This is because the packets generated by the load generator have

PAGE 35

25 higher probability to be lost as compared to the MPEG 4 video traffic packets. i.e., there are more load packets in the network at any given window of time as compared to the video traffic packets. But as the network load increases, the bandwidth available to the MPEG 4 video traffic decreases and hence the obs erved Error Free Length characteristics worsen. This phenomenon was also observed with the other packet size transmissions of Jurassic Park video. Same phenomenon was observed with the transmission of packets of other two video qualities for all the packet sizes. An additional interesting phenomenon was observed with the Aladdin video packet transmissions. In Figure 4.12, for the 4 Mbps case of Aladdin video, the Error Free Length characteristic was comparable and in some cases better than the No Load case. Interestingly, this was not seen in the transmissions of packets of the other two video qualities. Figure 4.11 Inverse CDF of Error Free Length for Jurassic Park Video for Different Network Load Conditions of 750 Byte Packet Transmissions when Mobile Unit is at 50ft A way from Access Point

PAGE 36

26 Figure 4.12 Inverse CDF of Error Free Length for Aladdin Video for Different Network Load Conditions of 750 Byte Packet Tran smissions when Mobile Unit is 50ft A way from Access Point In general, best performance of Error Free Length characteristics was given by 2 Mbps case followed by No Load (and even 4 Mbps in few Aladdin video transmissions) and worst performance is given in 8 Mbps network load followed by 6 Mbps case. Higher amount of losses in the network and h ence worse Error Free Length characteristics are expected as the network load is increased. This clearly explains the observed trend in Error Free Length characteristics. The effect of network load condition on the Error Length characteristic was also eva luated. For most of the scenarios, Error Length characteristic was best under 2 Mbps load condition followed by the No Load condition. Figure 4.13 shows this observation. This phenomenon was observed in all the packet transmissions of Star Wars and Aladdin videos. This is completely expected behavior based on the explanation for the observed behavior of Error Free Length characteristic with network load. Hence 2 Mbps case was followed by the No Load case, then by the 4 Mbps, 6 Mbps and 8 Mbps cases, respec tively. Interestingly, the same phenomenon was not seen in the case of Jurassic Park video transmissions (Figure 4.14). A complete reversal of the effect of network load on the Error Length characteristic can be seen here. The Error Length characteristic h as the best results under the 6 Mbps load followed by 8 Mbps condition. Also, 2 Mbps and No Load condition cases gave the worst Error Length characteristics.

PAGE 37

27 Figure 4.13 Inverse CDF of Error Length for Star Wars Video for Different Network Load Conditio ns of 500 Byte Packet Transmissions when Mobile Unit is 50ft A way from Access Point Figure 4.14 Inverse CDF of Error Length for Jurassic Park Video for Different Network Load Conditions of 750 Byte Packet Transmissions when Mobile Unit is 50ft A way from Access Point

PAGE 38

28 The effect of each individual network load condition on optimal packet size for transmission of all the different video qualities was also evaluated. Interestingly, it was found that the error characteristics vary for each video quality. Fig ure 4.15 presents the error characteristics of the Aladdin video for a particular network load. Figure 4.15 Inverse CDF of Error Length for Aladdin Video for Different MTU Packet Transmissions for 4 Mbps Network Load when Mobile Unit is at 50ft A way fro m Access Point Here, it can see that all the different packet sizes give almost similar Error Length characteristics. As the network load in this scenario is high, almost all packet size transmissions are equally affected. It can be clearly seen that 500 byte MTU packets give the worst Error Length characteristics. But no conclusion can be reached about the best packet sizes of 750, 1000 and 1500 bytes give an almost equal Error Length characteristics. Hence no packet size gives the best Error Length char acteristics in this particular case.

PAGE 39

29 Tables 4.1 4.3 give the best and worst MTU packet sizes under different network conditions for each video quality for the Error Length characteristics. Table 4.1 Best and Worst MTU Packet Sizes for Transmissio n of Jurassic Park Video under Various Network Load Conditions when Mobile Unit is 50ft from Access Point Load Best Packet Size Worst Packet Size Observations 8 Mbps None None Almost all packet sizes gave same performance here. 6 Mbps None None Almo st all packet sizes gave same performance here. 4 Mbps None None Almost all packet sizes gave same performance here. 2 Mbps 1000 Bytes 750 Bytes 1500 byte packets gave second best Error Length performance. No Load 1500 Bytes 500 Bytes 1000 byte packe ts gave the second best performance close to 1500 byte packets. Table 4.2 Best and Worst MTU Packet Sizes for Transmission of Aladdin Video under Various Network Load Conditions when Mobile Unit is 50ft from Access Point Load Best Packet Size Worst Pac ket Size Observations 8 Mbps None None Almost all packet sizes gave same performance here. 6 Mbps None None All the packets gave very close Error Length performance. 4 Mbps None 500 Bytes Here 500 byte packets are clearly worst case. 2 Mbps 1000 Byt es 750 Bytes 750 Bytes clearly is the worst case here. No Load 1500 Bytes 500 Bytes 1000 byte packets are second best performance and close to 1500 byte packets. Figure 4.4

PAGE 40

30 Table 4.3 Best and Worst MTU Packet Sizes for Transmission of Star Wa rs Video under Various Network Load Conditions when M obile U nit is 50ft from A ccess P oint Load Best Packet Size Worst Packet Size Observations 8 Mbps None None All the packets gave same performance here. 6 Mbps None 750 Bytes Here 750 byte packets gav e worst Error Length characteristics. All other packet sizes gave similar Error Length performance. 4 Mbps None 750 Bytes Here 750 byte packets are clearly worst case. 2 Mbps 1000 Bytes 1500 Bytes 1000 byte packet transmission has errors of lengths 1 an d 2 only. 750 Byte transmissions are second with max Error Length of 3. Also, 1500 Bytes clearly is the worst case here. No Load 1500 Bytes 500 Bytes 1000 byte packets gave second best performance. From the observations made in Table 4.1, it can be see n that under heavy network load conditions no particular packet size gives the best Error Length characteristics. If the low network load condition scenarios are considered (No Load and 2 Mbps), it can be clearly seen that 1000 and 1500 byte MTU packet tra nsmissions were optimal for best Error Length characteristics. Similarly, from Tables 4.2 and 4.3 it can be seen that these two packet sizes gave optimal performance in Aladdin and Star Wars videos as well. Under higher loads, all the packet transmissions gave same Error Length performance. Likewise, from Tables 4.1, 500 and 750 byte packets gave worst Error Length characteristics for Jurassic Park video. For Aladdin video, 500 and 750 byte packet transmissions gave the worst Error Length characteristics. Table 4.3 shows that 750 and 500 byte MTU packet transmissions gave worst error characteristics (with the exception of 2 Mbps case) for Star Wars video. Hence, under low load scenarios, 500 and 750 byte MTU packets gave worst Error Length characteristics for all videos. Also, 1000 and 1500 byte MTU packet transmissions gave best Error Length characteristics for all video transmissions. Both 1000 bytes and 1500 byte packet transmissions gave equally best Error Length characteristics because the average pack et sizes were less than 1000 bytes (Tables 4.11 4.13) in most of the scenarios. As 1000 byte packet transmissions incurred lesser overhead than 1500 byte transmissions, 1000 byte packets were optimal for transmission of MPEG 4 video traffic over IEEE 802 .11b Wireless LAN.

PAGE 41

31 4.1.4 Effect of Video Quality In this subsection comparison and evaluation of the Error Length and Error Free Length characteristics of the three different videos is presented. In subsection 4.1.1 it was seen that the Error Length and Error Free Length characteristics of each of the video traffic improved with decrease in distance of the mobile unit from the access point. From Figure 4.1 it can be seen that the best performance of error characteristics was given by Star Wars video follo wed by Aladdin and Jurassic Park in the order for both the 50ft and 75ft cases. This can be attributed to level of bandwidth utilization by these three videos, i.e., worsening of errors in the channel is expected with increase in bandwidth utilization of t he video traffic. Hence, Star Wars being least bandwidth utilizing, gave best Error Length characteristics and Jurassic park gave worst Error Length. Similar behavior was observed with other MTU packet size transmissions for the No Load condition. But when the network load was introduced, the error characteristics differ. Under load, Error Free Length characteristic has two steps. For low network loads (2 Mbps and No Load), the Aladdin video gave best performance followed by Jurassic Park and Star Wars. T his is evident from the Error Free Length characteristic for the three different videos for the optimal MTU packet size (1000 bytes) case in Figure 4.16. With a further increase in network load, Star Wars gradually gave best Error Free Length characteristi cs with increase in load. In the 4 Mbps case, Star Wars videos Error Free Length characteristic was the second best as compared to worst case earlier (Figure 4.17). Also, for higher network loads (6 and 8 Mbps), Star Wars video gave best Error Free Length characteristics followed by Aladdin and Jurassic Park. This can be clearly seen from Figure 4.18. As for the low network load scenarios, Aladdin and Jurassic Park videos gave better Error Free Length characteristics as compared to Star Wars. This can be attributed to the fact that as the network is very lightly loaded and the video traffic of Aladdin and Jurassic Park contains more packets in any given window of time as compared to Star Wars, hence an upper hand in giving better Error Free Length characte ristics. As the network load increases, the video traffic data is subjected to increased losses in the wireless channel. Also, due to larger number of packets in any given window of time, Jurassic Park and Aladdin lost more packets (hence worse Error Free Length characteristics) as compared to Star Wars. Also, if Aladdin and Jurassic Park are compared, Aladdin is expected to lose lesser packets in a given window of time than Jurassic Park, hence better Error Free Length characteristics for Aladdin as compa red to that of Jurassic Park. This explains the observed Error Free Length characteristics at higher network load.

PAGE 42

32 Figure 4.16 Inverse CDF of Error Free Length for 1000 Byte MTU Transmissions of A ll the Video Qualities for 2 Mbps Network Load when Mob ile Unit is at 50ft away from Access Point Figure 4.17 Inverse CDF of Error Free Length for 1000 Byte MTU Transmissions of A ll the Video Qualities for 4 Mbps Network Load when Mobile Unit is 50ft away from Access Point

PAGE 43

33 Figure 4.18 Inverse CDF of Erro r Free Length for 1500 Byte MTU Transmissions of A ll the Video Qualities for 6 Mbps Network Load when Mobile Unit is at 50ft away from Access Point As discussed before, 1000 byte MTU packets gave best error characteristics for all the video qualities and 500 byte packets gave the worst error characteristics. Hence, given all the earlier observations, Aladdin and Jurassic Park videos gave optimal performance for network loads less than or equal to 2 Mbps (actually 2 Mbps was better as it was the best netwo rk load condition for most of the videos) with MTU packet size of 1000 bytes. On the other hand for Star Wars, compared to the other two cases, even though gave comparatively better error free characteristics at higher load, here as well 1000 byte packet t ransmissions under 2 Mbps load condition are optimal. 4.2 Application L evel Characteristics In this section application level results are included showing how packet losses affect individual MPEG 4 frames. To evaluate the effect of the channel errors on the MPEG 4 frames, the following approach is taken. First, the Probability Density Function of the Error Length in number of packets was plotted to evaluate the percentage of each Error Length. It was seen that most of the errors occur for Error Length les s than 5 packets. Figure 4.19 gives an example. Here, 89% of the errors occur for Error Lengths less than 5 packets. In the best case, for example, Star

PAGE 44

34 Wars 1000 byte packets for 50ft and No Load, all the errors were contained in Error Lengths < 5 packets In the worst case (Jurassic Park 500 byte packets for 75ft and No Load), the same probability was around 50%. Overall considering all the scenarios on average, it was found that 91% of the errors were contained in Error Lengths less than 5 packets. As a result, evaluation of the effect of error bursts made of one, two, three and four consecutive packets on MPEG 4 frames (Frame Level Characteristics) was pursued. These error bursts were evaluated individually to know what type of MPEG 4 frames they affect the most. Finally, those packets and frames were evaluated even further to know how much information on average each of the MPEG 4 frames loses (Byte Level Characteristics). This case is called Byte Level characteristics as the effect of error bursts on a ll the bytes of the given MPEG 4 frame is evaluated here. Evaluation of the effect of the parameters discussed in Chapter 3 on these two characteristics is presented here. Figure 4.19 Error Length Distribution U sing Jurassic Park Video with a Packet Siz e of 1000 Bytes being 50ft away from Access Point u nder No Load Condition

PAGE 45

35 4.2.1 Frame Level Characteristics In this subsection, the effect of error bursts on individual I, B and P frames (Frames Affected Rates, FARs) is evaluated. Frames Affected Rate is the percentage of frames lost due to the given error burst. Figure 4.20 shows an example of Frames Affected Rates for I, B and P frames by error bursts of length one packet. It is clear that the when only one packet is in error, I frames are the most a ffected followed by P and B frames. This is completely expected behavior as there are least number of I frames followed by P and B frames in the video stream. Even if the number of I frames affected is small, the overall percentage of I frames affected wil l still be large as there are very few I frames in the video traffic. With similar reasoning it can be seen that B frames are the least affected followed by P frames. Similar trend was observed in all the network scenarios. Figure 4.20 Percentage of MP EG 4 Frames of Aladdin Video Affected by Error Bursts Made of One Packet, when the Access Point is 75ft away and Packets are 1 0 00 Bytes Long u nder No Network Load Condition Table 4.4 gives an example of the effect of error bursts with EL < 5 packets on Fr ames Affected Rate. The trend observed in Figure 4.2 is clearly seen in most of the cases (except when the frame affected rate is too low). Here, it can be seen that as the error burst size increases, the

PAGE 46

36 percentage of frames being affected decreases for all frame types. This is because of two reasons: firstly, as the error burst size increases, the probability of having that error burst decreases accordingly. Secondly, an error burst affects packets of a single frame or adjacent frames. There is a strong possibility for packets of a single frame to be affected hence decreasing the percentage loss to each frame type even further. Table 4. 4 Frames Affected Rates of MPEG 4 Frames by Error Bursts of Length Less than Five Packets for Aladdin Video Trace u nder No Load Condition at 75ft from Access Point Burst Size Pkt Size = 500 Bytes Pkt Size = 750 Bytes I P B I P B One Pkt 6.0 4.3 3.8 2.4 1.8 1.5 Two Pkts 2.6 2.0 1.7 1.0 0.8 0.7 Three Pkt 0.6 0.5 0.5 0.4 0.18 0.16 Four Pkt 0.3 0.3 0.4 0.04 0.06 0.05 To tal 9.5 7.1 6.4 3.84 2.84 2.41 Burst Size Pkt Size = 1000 Bytes Pkt Size = 1500 Bytes I P B I P B One Pkt 7.0 4.9 3.7 7.9 4.6 3.4 Two Pkts 1.2 1.17 1.0 1.4 1.0 0.8 Three Pkts 0.5 0.49 0.42 0.5 0.27 0.23 Four Pkts 0.3 0.25 0.2 0.08 0.12 0.08 Total 9 .0 6.81 5.32 9.88 5.99 4.51 Under low load (No Load and 2 Mbps) and medium load (4 Mbps) conditions, for 500 and 1000 byte MTU packets, Aladdin video gives the best Frame level characteristics for the I frame followed by Jurassic Park and Star Wars. But for the same 500 and 1000 byte MTU packets under higher loads, Jurassic Park gave the best Frame level characteristics for I frame followed by Aladdin and Star Wars videos. In case of 750 byte MTU packets, Jurassic Park video gave best frame level charact eristics under all network loads followed by Aladdin and Star Wars videos respectively. But in case of 1500 byte MTU packets, no specific trend was observed.

PAGE 47

37 4.2.2 Byte Level Characteristics Now that it is known which frames were affected the most one additional question remains. How much information is lost? In other words, when a frame is affected, is it lost completely or barely affected? This subsection addresses this question. We shall continue our analysis of the example of error bursts of one pa cket given in Figure 4.20 with the error burst distribution of this scenario given in Figure 4.21. Figure 4.21 Error Length Distribution U sing Aladdin Video with a Packet Size of 1000 Bytes being 75ft away from Access Point u nder No Load Condition

PAGE 48

38 F igure 4.22 Mean Amount of Bytes Affected in I, B and P Frames by Error Bursts Made of One Packet, when the Access Point is 75ft away and 1000 Bytes MTU for Aladdin Video u nder No Network Load Here, the average amount of bytes that are affected in I, B and P frames related to the frame size were evaluated. For the given example, the corresponding plot is given in Figure 4.22. As it can be seen, every time an I frame is affected by a burst of one packet, I frames lost about 18% of the data, P frames lost ab out 34%, and B frames lost around 48% of the data. This is again goods news. Not only I frames are the less affected by the bursts of errors but also they lost the least amount of bytes. Table 4. 5 shows all the cases, or the average percentage of bytes aff ected in I, B and P frames by Error Lengths of one, two, three and four packets using packet sizes of 500, 750, 1000 and 1500 bytes. Note that there are no packet losses in the Combination case (C) as an Error Length of one packet can affect a packet of a single frame type only.

PAGE 49

39 Table 4. 5 Average Percentage of Bytes Affected in I, B and P Frames by Error Lengths of One, Two, Three and Four Packets U sing All Four Packet Sizes for the Aladdin Video Trace under No Load Condition at 75ft from Access Poin t Burst Size Pkt Size = 500 Bytes Pkt Size = 750 Bytes I P B C I P B C One Pkt 9.7 20.4 31.5 0.0 13.8 30.8 45.9 0.0 Two Pkts 17.6 26.7 34.4 26.4 (IPB) 25.9 45.2 43.2 33.6 (IPB) Three Pkts 23.6 30.1 34.2 33.9 (IPB) 37.1 43.0 52.7 49.4 (IPB) Four Pkts 25.0 34.7 40.0 43.6 (IPB) 0.0 52.1 60.0 43.3 (IPB) Burst Size Pkt Size = 1000 Bytes Pkt Size = 1500 Bytes I P B C I P B C One Pkt 18.5 34.0 481. 0.0 26.4 44.4 60.1 0.0 Two Pkts 32.8 41.1 51.3 51.5 (IPB) 47.8 51.6 66.1 59.2 (IPB) Three Pkts 37.6 43.6 50.6 59.1 (IPB) 59.2 55.0 60.2 67.5 (IPB) Four Pkts 45.7 53.0 55.8 59.0 (IPB) 59.5 67.3 72.6 82.8 (IPB) Why are B frames the most affected and I frames the least? Tables 4. 6 4. 8 give the average packet size (in bytes) for each frame of the different video transmissions. Table 4.9 gives the average frame size (in bytes) in each video transmission. Let us consider the case of one packet loss for 1000 byte transmission of Aladdin video (Figure 4.22). From Tables 4. 7 and 4. 9, it can be seen on average an error burst of one packet affects: 17% of data in the affected I frames, 30% of data in affected P frames and 44% of data in affected B frames. These numbers closely match those observed in Figure 4.22. Similar reasoning applies to all the cases in. From t he tables, the similar reasoning holds for all the videos of all the MTU packets. This clearly explains why I frames were least affected and B frames the most even at the byte level. Table 4. 6 Average Packet Size for Different Frames for Jurassic Park Vi deo Packet size I Frame (in bytes) P Frame (in bytes) B Frame (in bytes) 1500 bytes 1315 1245 1179 1000 bytes 888 856 824 750 bytes 664 646 628 500 bytes 433 425 417

PAGE 50

40 Table 4. 7 Average Packet Size for Different Frames for Aladdin Video Packet siz e I Frame (in bytes) P Frame (in bytes) B Frame (in bytes) 1500 bytes 1257 1132 1006 1000 bytes 862 802 741 750 bytes 650 615 580 500 bytes 426 411 395 Table 4. 8 Average Packet Size for Different Frames for Star Wars Video Packet size I Frame (in by tes) P Frame (in bytes) B Frame (in bytes) 1500 bytes 987 289 108 1000 bytes 728 280 108 750 bytes 568 270 108 500 bytes 391 237 107 Table 4.9 Average Frame S ize for All MTU Packet Size Transmissions for All Video Qualities Frame Size I Frame (in by tes) P Frame (in bytes) B Frame (in bytes) Star wars 1552 290 98 Aladdin 4795 2592 1694 Jurassic 7180 4461 3177 From Table 4.5 it can be seen that as the length of the burst increases so does the amount of data affected; this is an expected behavior. The same behavior was observed when the packet size was increased. The Combination column shows the weighted average of the amount of data lost in all affected frames. However it does not show in what proportion are individual frames affected. Table 4. 10 answers that question, showing the amount of affected bytes in I, P and B frames every time a burst of one, two, three or four packets affect more than one type of frame. As it can be seen, I frames were barely affected most of the times, followed by P and B frames when more than one frame was involved in an error burst. Similar behavior for all the other cases involving different network load conditions and MTU packet sizes was observed.

PAGE 51

41 Table 4.1 0 Average Percentage of Bytes Affected in the Combi nations of I, B and P Frames by Error Lengths of One, Two, Three and Four Packets U sing A ll Four Packet Sizes for Aladdin Video under No Load at 75ft from Access Point Burst Size Pkt Size = 500 Bytes Pkt Size = 750 Bytes I P B I P B Two Pkts 25.7 19.7 37.2 3.7 24.8 46.2 Three Pkts 35.1 21.5 53.2 14.2 28.5 64.2 Four Pkts 6.8 30.8 54.7 5.6 27.0 58.8 Burst Size Pkt Size = 1000 Bytes Pkt Size = 1500 Bytes I P B I P B Two Pkts 5.4 39.8 67.1 6.9 45.7 75.2 Three Pkts 9.9 44.5 68.0 15.7 43.2 85.1 Four P kts 12.1 43.2 76.5 15.0 67.5 84.2 Looking at all the results presented in the Tables 4.5 and 4.10, it can be seen that the video traffic for this scenario is less affected when using small packet sizes (packets of 500 bytes MTU here). However, as the pac ket size is decreased, the probability of having bigger bursts of errors increases with the implication that bigger bursts affect I frames more. Similar behavior was observed with the Byte Level characteristics of Error Length for all other scenarios invol ving all possible combinations of the parameters. As a result, a better recommendation for packet size could be 750 bytes. But from section 4.1.3, more specifically referring to Tables 4.1 4.3, it can be seen that 750 bytes packets gave worst Error Lengt h characteristics for load conditions. Hence, a better alternative is to use packets of 1000 bytes as they offer the best compromise here and best network level characteristics as well. Also, in general, it was found that error bursts affect B frames the m ost followed b y P and I frames, respectively.

PAGE 52

42 CHAPTER 5 EXPERIMENTAL MODELING In Chapter 4, the results of the traffic characterization study of three different video movies over IEEE 802.11b wireless LANs are presented. The main goal here is to mode l these results to produce synthetic traffic generated by tool for software testing. In this chapter the modeling approach taken to handle this problem is also described. 5.1 Modeling EL and EFL D istributions Modeling Error Length and Error Free Length distributions can be very easy, for instance using a simple geometric distribution given the mean number of packets in error as the input parameters, which can be easily derived from the experimental data. However, this modeling technique applies to these cases where the random variable samples are independent. In Figure 5.2 we show that the data corresponding to the experiment in Figure 5.1 presents autocorrelation. This was somehow expected as it is widely know that errors in wireless networks occur in bu rsts. Modeling EL and EFL distribution with autocorrelation is a rather difficult endeavor, but studied and solved in the past.

PAGE 53

43 Figure 5.1 Inverse CDF of Error Free Length for 500 Byte MTU Packets for Aladdin Video at 50ft from Access Point under No N etwork Load Figure 5.2 Autocorrelation Function for First 100 Lags for the Error Free Length Characteristic Given in Figure 5.1

PAGE 54

44 5.2 Modeling of Characteristics with Autocorrelation The modeling of the characteristics having autocorrelation was done u sing the software called ARTAGEN which generates statistics for fitting and simulating random processes having autocorrelation [ 19 ]. This is done by ARTAGEN by taking the input process as an Auto Regressive to Anything (ARTA) process. ARTAGEN gives an almo st perfect fit for all the characteristics along with a good fit for the corresponding autocorrelation function. Originally ARTAGEN supports modeling a statistic along with its autocorrelation for the first 20 lags. Necessary changes were made to ARTAGEN to facilitate it to support modeling of a random process along with its autocorrelation for the first 100 lags to get a better model of the statistic. Figures 5.3 and Figure 5.4 give an example of modeling the characteristic given in Figure 5.1. Figure 5.3 Modeling of Inverse CDF of Error Free Length U sing ARTA Process for 500 Byte Packets for Aladdin Video at 50ft from Access Point and No Network Load Similar models of error length and error free length characteristics were obtained for all the differ ent scenarios discussed in Chapter 3.

PAGE 55

45 Figure 5.4 Matching of Autocorrelation Function for First 100 Lags for the Characteristic Given in Figure 5.3 by the ARTA Process 5.3 Networks Software Testing Test Bed Based on the traffic characterization study and the modeled characteristics obtained using ARTAGEN, the Network Software Testing Test Bed (NSTT) was designed and developed. The tool gives emulated model of the different wireless network scenarios for MPEG 4 video traffic considered in this thesis. The main components and the working of the tool are shown in Figure 5.5. The operations of the Network Software Testing Test Bed are as follows: Step 1: Using the user interface of the Test Bed, the user enters the wireless network scenario to be simulate d. The user interface of the tool is given in Figure 5.6. The tool takes in the following parameters from the user as the scenario to be simulated: Video Quality (Low, Medium or High Bandwidth utilizing). MTU packet size of packets to be transmitted (500, 750, 1000 or 1500 bytes). Network Load (No load, 2 Mbps, 4Mbps, 6Mbps or 8Mbps). Distance from the access point (50ft or 75ft).

PAGE 56

46 Figure 5.5 Working of the Ne twork Software Testing Test Bed Figure 5.6 Network Softwa re Test ing Test Bed Screenshot

PAGE 57

47 Step 2: Based on the input parameters specified by the user, the tool generates the name of the trace file for the video traffic generation. The initialization process for the video traffic generation is given in Figure 5.7. The Net work Software Testing Test Bed, residing in the traffic collector, sends the filename of the trace file to be used for the traffic generation to the video traffic collector. It also signals CRUDE to initiate traffic collection in the video traffic collecto r. The video traffic generator initiates transmission of packets from the corresponding input trace file using RUDE. Figure 5.7 Initialization Process of Video Traffic Step 3: The Network Software Testing Test Bed sits between RUDE and CRUDE (Figure 5 .7) to emulate the given wireless network scenario requested by the user. The Test bed maintains a database of the network characteristics modeled by ARTAGEN for all the different scenarios considered in Section 3.3. The test bed uses the modeled network c haracteristics generated by ARTAGEN, from this database, for the scenario input by the user to emulate the required wireless network environment. The basic algorithm employed by the test bed to emulate the wireless network environment for the MPEG 4 video traffic is given in Figure 5.8.

PAGE 58

48 Figure 5.8 Algorithm for Emulating Wireless Network by NSTT U sing Modeled Charac teristics Generated by ARTAGEN Step 4: The Network Software Testing Test Bed directs the video traffic subjected to emulated wireless netwo rk provided by ARTAGEN towards the traffic collector (i.e., CRUDE). Step 5: The video trace files collected by CRUDE were used to generate the error length and error free length characteristics of the emulated wireless network environment. The test bed ca n be implemented on a single computer with RUDE, CRUDE and NSTT installed, hence removing even the necessity of an Ethernet for the emulation process. 5.4 Verification of Emulated Characteristics It was verified that the Network Software Testing Test Bed effectively emulates the required wireless network scenario. We verified that all the emulated characteristics generated by Network Software Testing Test Bed match the actual characteristics as can be seen from the emulation of Error Length characteristic s for different videos as shown in Figures 5.9 5.14.

PAGE 59

49 Figure 5.9 Matching of Inverse CDF of the Actual Error Length Characteristic by the Emulated Characteristic for 1000 Byte Packets for Jurassic Park Vi deo at 75ft and No Network Load Figure 5.10 Matching of Autocorrelation Function for First 100 Lags for the Actual Characteristic Given in Figure 5.9 by the Emulated Characteristic

PAGE 60

50 Figure 5.11 Matching of Inverse CDF of the Actual Error Length Characteristic by the Emulated Characteristic for 1000 Byte Packets for Aladdin Video at 75f t and No Network Load Figure 5.12 Matching of Autocorrelation Function for First 100 Lags for the Actual Characteristic Given in Figure 5.11 by the Emulated Characteristic

PAGE 61

51 Figure 5.13 Matching of Inverse CDF of th e Actual Error Length Characteristic by the Emulated Characteristic for 1000 Byte Packets for Star Wars Vi deo at 75ft and No Network Load Figure 5.14 Matching of Autocorrelation Function for First 100 Lags for the Actual Characteristic Given in Figure 5 .13 by the Emulated Characteristic

PAGE 62

52 CHAPTER 6 CONCLUSIONS AND FUTURE WORK In this thesis, a thorough traffic characterization study of MPEG 4 video over IEEE 802.11b wireless LANs is performed based on the following parameters used for generating scena rios: distance of mobile unit from access point, packet MTU size, network load condition, and video quality (based on network bandwidth utilization). With the traffic characterization study of MPEG 4 video traffic over IEEE 802.11b wireless LANs the main c onclusions are: The MTU packet size of 1000 bytes is optimal for transmission of the three different quality videos when they are subjected to different possible network load conditions and distances from access points. Even though I frames are the most a ffected at the frame level, they are the least affected at the byte level followed by P and B frames for all the scenarios. Most of the video traffic gives optimal error length and error free length characteristics when the background network traffic is 2 Mbps for the average distance case (i.e., the mobile unit is 50ft away from access point). The Error Length and Error Free Length characteristics exhibit inherent autocorrelation. The results of the traffic characterization along with their autocorrelation were modeled using ARTA (Auto Regressive To Anything) software. These modeled characteristics were then used to build the Network Software Testing Test Bed to emulate the IEEE 802.11b wireless network environment for the MPEG 4 video traffic. Finally, it was verified that the traffic characteristics of the emulated wireless network effectively model the corresponding characteristics of the actual wireless network. Future work goes in several directions. Firstly, this tool can be used as a test bed to cha racterize anomalies, which can be used for software testing. Secondly, traffic characterization of the latest IEEE 802.11a wireless LANs can be performed to see the difference in loss behavior of IEEE 802.11a and IEEE 802.11b wireless LANs. Thirdly, the st udy can be extended to present results at the Macroblock level. Finally, investigation can be carried out to establish relationships between the MPEG layer and the wireless PHY Layer characteristics such as the effect of Signal/ Noise ratio and signal leve l on the video quality.

PAGE 63

53 REFERENCES [1] M. Garret. A Service Architecture for ATM: From Applications to Scheduling IEEE Network May June, 1996. [2] Steven M. Cherry. Whats right with Telecom In IEEE Spectrum January 2003. [3] In Cometa Inc. http://www.cometa.com [4] J. Watkinson. The MPEG Handbook Focal Press, 2001. [5] O. Rose. Simple and Efficient Models for Variable Bit Rate MPEG Video Traffic In Performance Evaluation 30 (1 2): 69 85 (1997) [6] V. Paxson, End to End Internet Packet Dynamics In IEEE/ACM Transaction on Networking, vol. 7, no. 3, pp. 277 -292, June 1999. [7] V. Paxton and S. Floyd. Wide area traffic: The failure of Poisson modeling In IEEE/ACM Transactions on Networking vol. 3, pp. 226 244, June 1995. [8] W. Leland, M. Taqqu, W. Willinger, and D. Wilson. On the self similar nature of Ethernet traffic In IEEE/ACM Transaction on Networking 2(1): 1 15, 1994. [9] A. Rueda and W. Kinsner. A Survey of Traffic Characterization Techn iques in Telecommunication Networks In Proceedings of IEEE Canadian Conference on Electrical and Computer Engineering pages 830 833, 1996. [10] G. Nguyen, R. Katz, B. Noble, and M. Satyanarayanan. A Trace Based Approach for Modeling Wireless Channel Beha vior In Proceedings of Winter Simulation Conference pages 597 604, 1996. [11] M. Arranz, R. Aguero, L. Munoz, and P. Mahonen. Behavior of UDP based Applications over IEEE 802.11 Wireless Networks In Proceedings of 12th IEEE International Symposium on P ersonal, Indoor and Mobile Radio Communications pages 72 77, 2001. [12] D. Eckhardt and P. Steenkiste. Measurement and Analysis of the Error Characteristics of an In building Wireless Network In Proceedings of SIGCOMM pages 243 254, 1996. [13] F. Fitzek and M. Reisslein. MPEG 4 and H.263 Video Traces for Network Performance Evaluation In IEEE Network Vol. 15, No. 4, pages 40 54, November/December 2001.

PAGE 64

54 [14] J. Laine, S. Saaristo, and R. Prior. RUDE and CRUDE Programs. In http://www.atm.tut.fi/rude/ [15] RUDE and CRUDE man pages. [16] M. Garrett and W. Willinger. Analysis, modeling and generation of self similar VBR video traffic In Proceedings of ACM SIGCOMM '94, volume 24 of Computer Communications Review pages 269 -280, London, October 1994. [17] B. Noble, M. Satyanarayanan, G. Nguyen, and R. Katz. Trace Based Mobile Network Emulation In Proceedings of SIGCOMM pages 51 61, 1997. [18] F. Tobagi and I. Dalgic. Performance Evaluation of 10Base T and 100B ase T Ethernets Carrying Multimedia Traffic In IEEE Journal on Selected Areas in Communications 14(7):1436 1454, 1996. [19] M. Cario and B. Nelson. Numerical Methods for Fitting and Simulating Autoregressive to Anything Processes In INFORMS Journal o n Computing (10): 72 81, 1998. [20] IEEE std 802.11 1997 information technology telecommunications and information exchange between sys In IEEE Std 802.11 1997 ,pages i 445, 1997. [21] S. Lee, B. Melamed, A. Reibman, and B. Sengupta. Analysis of a vide o multiplexer using TES as a modeling methodology In Proceedings of IEEE Global Telecommunications Conference pp. 131 135, 1991. [22] B. Melamed. An Overview of TES Processes and Modeling Methodology In Performance Evaluation of Computer and Communica tions Systems, (L. Donatiello and R. Nelson, Eds.), pp. 359 -393, Springer Verlag Lecture Notes in Computer Science. [23] B. Melamed and D. Pendarakis. A TES Based Model for Compressed ``Star Wars'' Video In Proceedings of IEEE GLOBECOM Communications M ini Conference San Francisco, California, 120 126, November, 1994. [24] P. Ikkurthy and M. Labrador. Characterization of MPEG 4 Video Traffic over IEEE 802.11b Wireless LANs In Proceedings of IEEE Local Computer Networks Conference pp. 421 430, Nov ember ,2002 [25] AT &T Corp, WaveLAN/ PCMCIA User's Guide [26] IEEE P802.11, The Working Group for Wireless LANs. http://grouper.ieee.org/groups/802/11/ [27] Breezecom, IEEE 802.11 Technical Tu torial Available at URL: http://www.breezecom.com/Materials/PDFFiles/802.11Tut.pdf [28] W. Stallings. Wireless Communications and Networks Pearson Education Inc. [29] B. Crow, I. Widjaja, J. Kim and P. Sakai. IEEE 802.11 Wireless Local Area Networks In Proceedings of IEEE Communications Magazine 1997, pp 116 126.

PAGE 65

55 [30] G Xylomenos and G Polyzos. TCP and UDP Performance Over a Wireless LAN In Proceedings of the IEEE INFOCOM` 99 pages 439 -446, March 1999. [31] Trace Statistics of MPEG 4 traces for network performance evaluation. http://www tkn.ee.tu berlin.de/~fitzek/TRACE/trace.html


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 001416920
003 fts
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 030729s2003 flua sbm s000|0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0000058
9 035
AJJ4772
b SE
SFE0000058
040
FHM
c FHM
049
FHME
090
QA76
1 100
Ikkurthy, Praveen Chiranjeevi.
0 245
Software testing testbed for MPEG-4 video traffic over IEEE 802.11b wireless lans
h [electronic resource] /
by Praveen Chiranjeevi Ikkurthy.
260
[Tampa, Fla.] :
University of South Florida,
2003.
502
Thesis (M.S.C.S.)--University of South Florida, 2003.
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.
3 520
ABSTRACT: Several traffic characterization studies have been performed on wireless LANs with the main objective of realizing good and accurate models of the errors in the wireless channel. These models have been extended to model the effect of errors on higher layer protocols, mainly at the data link layer. However, no prior work has been done to study the application level characteristics of MPEG-4 video traffic over 802.11b wireless networks. In this thesis a traffic characterization study of MPEG-4 video traffic over IEEE 802.11b wireless LANs with the main goal of building a tool for software testing is performed. Using two freely available tools to send and receive real-time streams and collect and analyze traces, MPEG-4 encoded video frames are sent over a 11 Mbps, 802.11b wireless LAN to characterize the errors in the channel and the effect of those errors on the quality of the movie. The results of this traffic characterization were modeled using ARTA (Auto Regressive-To-Anything) software. These modeled characteristics were then used to build a tool that generates synthetic traffic emulating real wireless network scenario. The tool emulates the error length and error free length characteristics of the wireless network for the MPEG-4 video traffic using the corresponding modeled characteristics generated by ARTA. The tool can be used by software developers to test their MPEG-4 streaming media applications without the need of the real infrastructure. The tool can also be trained and extended to support testing of any streaming media applications.
590
Adviser: Labrador, Miguel A.
653
autocorrelation.
arta.
error-free length.
error length.
traffic characterization.
690
Dissertations, Academic
z USF
x Computer Science
Masters.
773
t USF Electronic Theses and Dissertations.
4 856
u http://digital.lib.usf.edu/?e14.58