Evaluation of TCP Performance in 3G Mobile Networks

Master ThesisElectrical EngineeringJune 2011

Evaluation of TCP Performance in 3G MobileNetworks

Mutyalu Jogi

School of ComputingBlekinge Institute of Technology37179 KarlskronaSweden

This thesis is submitted to the School of Computing at Blekinge Instituteof Technology in partial fulfillment of the requirements for the degree ofMaster of Science in Electrical Engineering. The thesis is equivalent to 20weeks of full time studies.

Contact Information

Author(s):Mutyalu JogiE-mail: [email protected]

University advisor(s):

Prof. Adrian PopescuCOM/BTHDr. Patrik ArlosCOM/BTH

School of ComputingBlekinge Institute of Technology371 79 KARLSKRONA SWEDEN

Internet: www.bth.se/comPhone: +46 455 385000SWEDEN

Acknowledgment

I would like to thank my supervisor, Prof. Adrian Popescu for spending hisvaluable time and the support that he has given to me. I would also like toshow my gratitude to Dr. Patrik Arlos for providing me guidance regardingthe experiment setup and measurements. Lastly, I offer my sincere thanksto Ravi Chandra Kommalapati, Kranthi kiran Kolachina and my friends fortheir support and making this work, a success.

Mutyalu JogiMadhu Vundavalli

3

Abstract

With the increase in mobile broadband services the operators are gainingprofits by providing high speed Internet access over the mobile network. Onthe other side they are also facing challenges to give QoS guarantee to thecustomers.

In this thesis we investigate the impact of data rate and payload sizeon One Way Delay (OWD) and packet loss over TCP performance in 3Gnetworks. Our goal is to evaluate the OWD and packet loss characteristicswith respect to payload size and data rate from the collected network leveltraces. To collect these traces an experimental testbed is setup with EndaceData Acquisition and Generation (DAG) cards, for accurate measurementsEndace DAG cards together with Global Positioning System (GPS) synchro-nization is implemented. The experiments are conducted for three differentSwedish mobile operator networks and further the statistics of OWD mea-surements and packet loss for different data rates and payload sizes areevaluated. Our results indicate that the minimal OWD occurred at higherdata rates and also shows a high delay variability. The packet loss has muchimpact on higher data rates and larger payload sizes, as the packet lossincreases with the increase in data rate and payload size.

Keywords: One Way Delay, Data rate, Packet loss, Payload.

i

Contents

Acknowledgment 3

Abstract i

Contents ii

List of Figures iv

List of Tables v

List of Abbreviations 1

Introduction 2

1 Problem Identification 41.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.2 UMTS Network Overview . . . . . . . . . . . . . . . . . . . . 51.3 Transmission Control Protocol . . . . . . . . . . . . . . . . . 61.4 Related Works . . . . . . . . . . . . . . . . . . . . . . . . . . 71.5 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.6 Research Questions . . . . . . . . . . . . . . . . . . . . . . . . 91.7 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91.8 Thesis layout . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

Implementation 11

2 Implementation 122.1 Experimental testbed . . . . . . . . . . . . . . . . . . . . . . . 13

2.1.1 DPMI . . . . . . . . . . . . . . . . . . . . . . . . . . . 132.1.2 Gateway Implementation . . . . . . . . . . . . . . . . 13

ii

2.1.3 Measurement Point . . . . . . . . . . . . . . . . . . . . 132.1.4 Sender/receiver systems . . . . . . . . . . . . . . . . . 14

2.2 Calculation of OWD . . . . . . . . . . . . . . . . . . . . . . . 142.3 Delay components . . . . . . . . . . . . . . . . . . . . . . . . 152.4 Experiment settings . . . . . . . . . . . . . . . . . . . . . . . 162.5 Analysis method . . . . . . . . . . . . . . . . . . . . . . . . . 20

Results 21

3 Results 223.1 Evaluation of Gateway . . . . . . . . . . . . . . . . . . . . . . 22

3.1.1 Packet loss . . . . . . . . . . . . . . . . . . . . . . . . 253.1.2 Identification of sawtooth pattern . . . . . . . . . . . . 263.1.3 Acknowledgment packets . . . . . . . . . . . . . . . . 27

3.2 Evaluation of 3G UMTS Networks . . . . . . . . . . . . . . . 283.2.1 OWD in Upload . . . . . . . . . . . . . . . . . . . . . 283.2.2 OWD in Download . . . . . . . . . . . . . . . . . . . . 323.2.3 Acknowledgment packets . . . . . . . . . . . . . . . . 363.2.4 Segment Distribution . . . . . . . . . . . . . . . . . . 363.2.5 Packet loss . . . . . . . . . . . . . . . . . . . . . . . . 38

4 Conclusions 404.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404.2 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

Bibliography 41

Appendix 44

A Experimental Results 45A.1 Gateway . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

A.1.1 Upload . . . . . . . . . . . . . . . . . . . . . . . . . . 45A.1.2 Download . . . . . . . . . . . . . . . . . . . . . . . . . 47

A.2 Mobile operators . . . . . . . . . . . . . . . . . . . . . . . . . 50A.2.1 Download . . . . . . . . . . . . . . . . . . . . . . . . . 50A.2.2 Packet loss . . . . . . . . . . . . . . . . . . . . . . . . 54

iii

List of Figures

1.1 Format of TCP header . . . . . . . . . . . . . . . . . . . . . . 6

2.1 Detailed experiment setup . . . . . . . . . . . . . . . . . . . . 122.2 Wiring method . . . . . . . . . . . . . . . . . . . . . . . . . . 152.3 Experiment setup for GW evaluation . . . . . . . . . . . . . . 16

3.1 Minimum OWD distribution in GW . . . . . . . . . . . . . . 233.2 Identification of delay spike at 726 bytes in Upload, GW . . . 253.3 Identification of delay spike at 1218 bytes in Download, GW . 253.4 Percentage of packet loss, GW . . . . . . . . . . . . . . . . . 263.5 Identification of sawtooth pattern in GW, Upload . . . . . . . 273.6 Minimum OWD distribution in acknowledgment packets, GW 273.7 Minimum OWD distribution in mobile operators, Upload . . 283.8 OWD Distribution for 5 minute duration . . . . . . . . . . . 323.9 Minimum OWD distribution in mobile operators, Download . 333.10 Identification of delay spike in OP2, Download . . . . . . . . 333.11 Acknowledgment packets in mobile operators, Upload . . . . 363.12 Segment distribution in OP3, Upload . . . . . . . . . . . . . . 373.13 CCDF for payload size 726 bytes in OP3, Upload . . . . . . . 383.14 Percentage of packet loss in mobile operators, Upload . . . . 393.15 Percentage of packet loss in mobile operators, Download . . . 39

iv

List of Tables

2.1 Experiment settings from 40 bytes to 232 bytes . . . . . . . . 172.2 Experiment settings from 488 bytes to 726 bytes . . . . . . . 182.3 Experiment settings from 1000 bytes to 1416 bytes . . . . . . 19

3.1 Statistics of GW from 256kbps to 360kbps, Upload . . . . . . 233.2 Statistics of GW from 512kbps to 1024kbps, Download . . . 243.3 Statistics of mobile operators from 8kbps to 16kbps, Upload . 293.4 Statistics of mobile operators from 32kbps to 64kbps, Upload 303.5 Statistics of mobile operators from 128kbps to 360kbps, Upload 313.6 Statistics of mobile operators from 360kbps to 512kbps, Down-

load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343.7 Statistics of mobile operators from 1024kbps to 3072kbps,

Download . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

A.1 Statistics of GW from 8kbps to 16kbps, Upload . . . . . . . . 46A.2 Statistics of GW from 32kbps to 128kbps, Upload . . . . . . . 46A.3 Statistics of GW from 8kbps to 16kbps, Download . . . . . . 48A.4 Statistics of GW from 32kbps to 128kbps, Download . . . . . 48A.5 Statistics of GW from 256kbps to 360kbps, Download . . . . 49A.6 Statistics of GW from 2048kbps to 3072kbps, Download . . . 49A.7 Statistics of mobile operators from 8kbps to 16kbps, Download 51A.8 Statistics of mobile operators from 32kbps to 64kbps, Download 52A.9 Statistics of mobile operators from 128kbps to 256kbps, Down-

load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53A.10 Packet loss from 40 bytes to 1000 bytes, Upload . . . . . . . . 55A.11 Packet loss from 1218 bytes to 1416 bytes, Upload . . . . . . 56A.12 Packet loss from 40 bytes to 488 bytes, Download . . . . . . . 57A.13 Packet loss from 726 bytes to 1000 bytes, Download . . . . . 58A.14 Packet loss from 1218 bytes to 1416 bytes, Download . . . . . 59

v

List of Abbreviations

1. OWD: One Way Delay

2. GPS: Global Positioning System

3. DAG: Data Acquisition and Generation

4. QoS: Quality of Service

5. UMTS: Universal Mobile Telecommunication Systems

6. TCP: Transmission Control protocol

7. ECN: Explicit Congestion Notification

8. ACK: Acknowledgment

9. DPMI: Distributed Passive Measurement Infrastructure

10. SLA: Service Level Agreement

11. 3G: Third generation

12. USB: Universal Serial Bus

13. WCDMA: Wideband Code Division Multiple Access

14. CN: Core Network

15. UTRAN: Universal Terrestrial Radio Access Network

16. UE: User Equipment

17. Uu: Radio Interface

18. TDD: Time Division Duplex

19. FDD: Frequency Division Duplex

20. MAC: Medium Access Control

21. RLC: Radio Link Control

1

2

22. TTI: Transmission Time Interval

23. PDU: Protocol Data Unit

24. SDU: Service Data Unit

25. UM: Unacknowledged Mode

26. AM: Acknowledged Mode

27. IP: Internet Protocol

28. ISN: Initial Sequence Number

29. VoIP: Voice over IP

30. GPRS: General Packet Radio Service

31. UDP: User Datagram Protocol

32. ARQ: Automatic Repeat Request

33. GW: Gateway

34. MArC: Measurement Area Controller

35. CI: Capture Interface

36. FL: Frame Length

37. MPID: Measurement Point ID

38. TS: Time Stamp

39. SRC: Source

40. DST: Destination

41. NTP: Network Time Protocol

42. WAN: Wide Area Network

43. LAN: Local Area Network

44. RTT: Round Trip Time

45. CCDF: Complementary Cumulative Distribution Function

Introduction

3

Chapter 1

Problem Identification

1.1 Background

Now-a-days, mobile broadband services are becoming very popular due totheir ease of use and able to access the Internet virtually anywhere. The in-troduction of 3G technologies has made up a way for high speed data accessthrough mobile handsets, portable modems and laptops with USB donglesover the mobile network. With respect to the bandwidth, the mobile broad-band speeds are currently varying up to 21Mbps and, in the near futureit may be 50-100Mbps [1]. Universal Mobile Telecommunication Systems(UMTS) is one of the third generation mobile communication technologies,which are specially designed to provide high performance for mobile net-works. Here, the pre-user capacity allocation depends on the mobility pat-tern, the user density, the offered traffic and the radio conditions, etc. Theoperators and the service providers are gaining more profits by providingthese services. But on the other side they are also facing challenges to copeup with the market competition to provide QoS guarantee to the customers.

QoS is described in terms of guaranteed network parameters, applicationbehavior and Service Level Agreement (SLA) with class differentiation. Itwas divided in to Subjective QoS and Objective QoS. Objective QoS refersto network parameters such as delay, packet loss and jitter. Whereas, Sub-jective QoS refers to user’s experience of the network applications. Thecapacity distribution mechanism on mobile access links is considerably lesstransparent to the users, but still its characteristics should be known in or-der to reduce the affect of network on application performance. Networkmeasurements can provide useful information about the performance andcharacteristics of some specific network [2]. Passive and active measure-ments are performed to collect network traffic traces, which are useful inresearch and development of traffic models [3, 4].

In this thesis we investigate the impact of data rate and payload sizeover TCP performance in 3G networks. In order to evaluate this, a test

4

CHAPTER 1. PROBLEM IDENTIFICATION 5

bed is created. The test bed setup is created in such a way that the 3GUSB modem is used as a communicating device between the sender andthe receiver. The USB device is used by three Swedish mobile operatornetworks. To calculate OWD the network traces collected with respect topayload size and data rate are copied and later analyzed for OWD. Basedon these results we can analyze the TCP performance of 3G data servicesover mobile networks.

1.2 UMTS Network Overview

UMTS is a third generation cellular network, designed to provide high per-formance mobile access to the Internet. It is based on Wide Code DivisionMultiple Access (WCDMA) technique and further divided in to Core Net-work (CN), Universal Terrestrial Radio Access Network (UTRAN) and UserEquipment (UE). The additional mechanisms are added in order to enhancefrom 2G to 3G networks by allowing different existence access techniquesand core network technologies. The UE in the UMTS network is connectedto UTRAN through the radio interface (Uu) and more over the radio inter-face supports both TDD and FDD mode operations. Both modes use thesame network architecture and protocols [5]. But currently the FDD modeis usually used in the deployed 3G networks. The radio interface protocolarchitecture is divided in to three layers namely physical (L1), data link(L2), and network layers (L3). Data link (L2) layer is further divided in toMedium Access Control (MAC) layer, Radio Link Control (RLC) layer andthe packet convergence protocol.

The WCDMA mechanism divides the bandwidth in to physical channelsin the physical layer operation of Uu radio interface, where the transportchannels information are mapped. Here in the physical layer process, thetemporization is set by establishing 10 ms frames. The transport channelsat this time are clocked at multiples known as Transmission Time Inter-val (TTI). Every TTI blocks in physical layer are received from the upperprotocols to further transmit these over the wireless link. The TTI blocksusually can be 1, 2 or 4 frames depending on the type of service. Size of thedata transmitted over TTI is varied according to the configured data rate.

The RLC sub layer divides all the mechanisms in to smaller data blocks(Packet Data Units, PDUs) needed for reassembly of upper layer data pack-ets (Service Data Units, SDUs) and fragmentation. So that they can betransmitted through radio interface by the lower layers. While, the MACsub layer maps the logical channels to transport channels, which are offeredby the physical layer. The RLC sub layer provides two transparent modesto transfer the data to upper layers. The modes are Unacknowledged Mode(UM) and Acknowledged Mode (AM). RLC AM mode is used for the In-ternet data transport. The AM uses the sliding window protocol with a


selective repeat mechanism in order to avoid data errors and losses. In thismechanism the receiving RLC entity sends back an acknowledgment mes-sage to the sending RLC entity, to check that the PDUs are received witherror free and without losses. Depending on the errors and losses the receiverdecides whether to retransmit the PDUs or send the new ones. To avoidstalling and further delays the RLC transmitting entity uses a polling mech-anism, as it polls the receiving entity and requests a status report. The RLCAM mode reduces the radio packet corruption, but introduces high packetdelay variability.

1.3 Transmission Control Protocol

TCP is a conventional and reliable stream oriented protocol, which providesconnection oriented service. It comes under transport layer and providesguaranteed delivery of the data with no loss. TCP reduces the overloadingin network by adapting to which it is connected. The adaptation processis done by slow start method, in which it sends a packet and waits for anacknowledgment. Depending upon the received acknowledgment it changesits window size and packet size. There might be a packet loss if the windowsize exceeds beyond the network capacity. To avoid data loss RTT’s areused to calculate the retransmission time outs. It reduces the data loss withthe help of its internal timer. RTT comes under the connection orientedservices in one of the performance metrics of QoS. TCP is like as remainingprotocols, where it contains a TCP header. In the transport layer the datareceived from the upper layers is attached with a TCP header and sent tothe lower layers to reach its destination address. While in the receiving sidethe data received from the lower layers is checked and then removing theheaders it is sent to the upper layers. Figure 1.1 shows the TCP header.

Figure 1.1: Format of TCP header

Each field included in the TCP header are explained as follows. A TCPheader contains of a source port number and destination port number, to


identify the sending and receiving data. Along with these two port numbersthe source address and destination address in the IP header uniquely iden-tify each connection. The combination of source IP address, port numberand destination IP address, port number refer to as socket. Both Sourceand destination sockets when combined are defined as a stream. Sequencenumber is a 32 bit number, where it places the packet sequence, in order toensure that the data is transferred with reliability. When a new connectionis being initiated in TCP it turns on the SYN flag and further it enters intoconnection established state. The sequence number field which is chosen bythe host of that particular connection contains an Initial Sequence Number.The first byte of data sent in the sequence number field will be the ISN plusone. The ACK flag is always kept ON when the connection is established.acknowledgment number is a 32-bit number, which contains the next se-quence number that the receiver expects to receive from the sender. Headerlength is the number of 32 bit words in the TCP header. It is always amultiple of 32 bits. ECN is an optional field which is intended to protectagainst accidental concealment of packets from the TCP sender. Controlbits refer to as SYN, ACK, and FIN. Window size is a 16-bit unsigned num-ber, where the number of bytes beginning with the one indicated in theacknowledgment field.

1.4 Related Works

There are several proposals being made to improve the performance in 3Gmobile networks.

One such proposal is to investigate the impact of packet size on minimalOWD for the uplink in third generation networks [6]. This is an importantperformance factor for real time scenarios such as Voice over IP (VoIP) andmultimedia applications [7]. Based on this, there are methods suggested tofind the affect of packet size on OWD. The results from these measurementsshow that, for increased packet size the performance increases, whereas forsmaller packet sizes the performance is decreased [6]. These works are basedon passive measurements, which characterize the delay behavior in mobilenetworks.

In the study [8], the authors conducted real time experiments on TCP [9]to calculate the RTT’s for UMTS and General Packet Radio Service (GPRS)networks in Austria. The experiments were conducted during the periods(18:00-24:00). The RTT was 476 ms in GPRS and 127ms in UMTS networks,respectively. From the experiments they found that there is no significantchange in the RTT distribution during different time loads and large changesin the traffic loads. So, they suggested the choice to conduct the experimentswith less duration to get better results in the traffic behavior. Consideringthese measures we have selected the observation period of 5 minutes for each


experiment and all experiments are conducted between 8:00-18:00.In [10], the author performed a set of experiments with respect to payload

size and data rate in order to evaluate the OWD in 3G networks. Theexperiments are done on considering User Datagram Protocol (UDP) tocalculate the OWD. From the results they found an interesting pattern at128kbps in upload scenario for all the mobile operators. While uploadingthe data with combination of different payload sizes and data rates, theminimum delay is decreased at 128 kbps and again increased to further datarates. Also it is has been noticed that in all the trace files, the initial sequenceof packets experience higher delays for some fraction of seconds. So, theyconcluded that the problem is due to UMTS signaling, while attaining achannel after an idle period.

In [11], the authors characterized the TCP behavior with respect to dif-ferent data rates in dedicated channels over UMTS networks. These resultsare based on simulations and from the conclusions they came to understandthat TCP mechanisms offer reliability, but they cannot provide QoS guar-antee to the end users in terms of delay and throughput.

Another study [12] discusses the implementation details of Real TimeLive RTT Analyzer. It is used to calculate the RTT for each TCP stream,which is capable of reading both offline as well as online streams and helpsto analyze the statistics which are obtained from the collected RTT streams.

In study [5], the research work was done to observe the end to end be-havior of Internet access through UMTS networks. This work was mainlyfocussed on delay and proposed a testbed which replicates the UMTS Inter-net access environment. The results show that the packet loss has negligibleeffect when compared to delay. This is because the UMTS link level retrans-mission mechanisms are efficient in retransmitting the unsuccessful packetsto the receiver. But the Automatic Repeat Request (ARQ) [13] loss recoverymechanisms failed to meet the QoS in 3G networks.

In study [11], the experimentation was done with the mobile testing soft-ware MOSET on commercial WCDMA/3G network. From those findings,the authors investigated that the initial sequence of packets are affected bythe UMTS signaling while acquiring the channel. They also noticed thatgood throughput is an important factor when the data is transmitted af-ter an inactivity state. There are several popular active measurement toolswhich can characterize the end to end performance of data services on mo-bile network. These measures can be taken by injecting the sample packetsinto the network. The tools H.323 Beacon [14] and Iperf [15] are especiallydesigned to monitor the measurement infrastructure and calculate OWD,RTT, jitter and packet loss.


1.5 Motivation

Apart from the past research there is a wide range of work done on OWDto improve the performance of data services in UMTS networks. OWDplays a important role to measure network performance and for applicationdesign. Generally, RTT is used as an approximation of delay and in TCP theperformance can be improved in asymmetric networks in which the OWDis much more accurate than RTT [16]. It is also important that QoS iswidely adopted with the increase of multimedia applications such as VoIP,online gaming and video conferencing. These applications are required toprovide real time service with reliability. Considering both directions maylead to errors, since many communication networks are asymmetric, whichproduce major traffic flows in one direction only [17]. Thus the choice oftaking One Way measurements are especially important in wireless networks,as the link performance affects on TCP and UDP [18]. In the previousresearch the investigation of OWD is done on UDP and there is less attentionpaid towards TCP [10]. So, considering these factors it motivates us toinvestigate the OWD for TCP performance. Such measurements can beuseful for application developers and mobile operators to choose optimalpacket size and data rate for UMTS network conditions.

1.6 Research Questions

1. What are the OWD and packet loss characteristics in 3G mobile net-works?

2. How does the OWD and packet loss characteristics vary for three dif-ferent Swedish mobile operator networks?

3. What are the OWD and packet loss characteristics for both uplink anddownlink applications in 3G mobile networks?

1.7 Objectives

One of the sub goal of this thesis is to provide comparative study betweenthree Swedish mobile operator networks. The comparative study is basedon OWD and packet loss characteristics of TCP. Moreover, the three mobileoperator networks are namely Tele2, Tre and Telia [19, 20, 21]. The firstframe of this thesis is focussed on implementing measurement infrastructureand collecting network level traces, which are captured from a real networkat sending and receiving ends. To evaluate the 3G UMTS networks anexperimental testbed was created. The main contribution of this thesisis to provide the information about experimental configuration and thenlater carried out the experiments. After the experiments, the results are


explained with the comparison of mobile operator networks for both uploadand download scenarios.

1.8 Thesis layout

The document is organized as follows. In the next chapter the experimentsetup is explained, followed by the results and conclusions which are obtainedfrom experiments.

Implementation

11

Chapter 2

Implementation

Evaluation of OWD and packet loss characteristics is the main goal of thisthesis. For this evaluation, we conducted a set of experiments by usingvarious scenarios. This chapter discusses and describes the experimentalsetup that we used for the evaluation.

Figure 2.1: Complete experiment test bed

An existing traffic generator tool was used, which sends TCP packetsperiodically from Client (sender) to the Server (receiver). The experimentsetup is shown in Figure 2.1. The sender is connected to the Gateway(GW), which uses the mobile operator 3G/UMTS network over radio linkvia Huawei E220 USB modem [22]. Moreover, the GW is connected to3G network using option WCDMA preferred network. The receiver is con-nected to the Internet by using public IP address. Using this setup, thesender was able to send TCP packets to the receiver over a mobile opera-tor network and the receiver is able to receive it through the core network.The Measurement Point (MP) presented in between these two systems actsas a capturing device. The packets generated from sender and arrived atthe receiver get accurate time stamps from this MP. Using this setup, it ispossible to evaluate the OWD.

12

CHAPTER 2. IMPLEMENTATION 13

2.1 Experimental testbed

This section discusses the hardware and software tools used in the experi-ments. In the experiment setup, the Distributed Passive Measurement In-frastructure (DPMI) configuration is implemented [23] and the remainingelements of the experiments are discussed below.

2.1.1 DPMI

This infrastructure was used to analyze the behavior of networks, as it pro-vides modularity, security, ease of use and moreover it reduces cost reduc-tion. The components included in this infrastructure are Measurement AreaController (MArC), Measurement Point and a Consumer. MARC controlsthe network in the MP and the dag interfaces of the MP captures the datausing wiretaps. The captured packet is then encapsulated into DPMI packetheader format. This header consists of Capture Interface (CI), Frame Length(FL), Measurement Point ID (MPID), Time Stamp (TS), Capture Lengthand actual payload. We get the accurate precision time stamp and actualcapture time of the packet from the TS field in the header. The frame de-capsulation is done for each packet and they are in big-endian or little-endianformat.

2.1.2 Gateway Implementation

Between sender and receiver, the GW is implemented. It is a Intel Core 2Duo CPU P8600 2.40GHZ with 4GBytes RAM (Windows Vista) connectedto the sender through a built in Broadcom Ethernet Card. The GW is alsoconnected with the UMTS network via Huawei E220 USB modem. Thereason behind connecting this GW through USB modem is that it is notconceivable to connect MP through USB interface. The sender is connectedto the GW via wiretap A and the receiver is connected to the Internetvia wiretap B through Ethernet link. With this experiment setup the trafficgenerator is able to generate TCP packets from sender over operator’s UMTSnetwork and the receiver collects them through Internet.

2.1.3 Measurement Point

The use of MP is to measure the sending and receiving times of the packet.Wiretaps collect the data and sends that data to the MP. Wiretaps are usedat two places in the setup, one between sender and GW and other betweenthe external network and receiver. The other job of wiretap is to duplicatethe incoming traffic and send it to the MP. MP is a Linux computer set upwith two Endace DAG 3.5 E cards. These DAG cards are mainly used tocapture traffic without any loss of its interfaces. Using this arrangement,the MP is able to filter the packets and intercept the traffic according to


the rules given by the MArC. Then the collected network traces from theMP are stored in the consumer. To get accurate measurements, the DAGinterfaces are duly synchronized. This synchronization is done with the useof GPS.

2.1.4 Sender/receiver systems

The sender and the receiver systems are two Linux computers in which atraffic generator tool is installed on both the systems. One will be acting asa generator and other as a sink. Both systems (p2-400 MHz with Linux 2.4kernels) are connected to a 10 Mbps full duplex Ethernet cards. The trafficgenerator tool is a C++ program, which generates TCP packets at senderand collects them at the receiver. This tool consists of two applicationsnamely Client and the Server. The Server waits for the Client request toestablish a connection between the two entities. The Client then sends thedata according to the arguments given by the user entering into a connectionestablished state. Then the data is being exchanged between the Client andServer until the Client closes the connection. Client takes experiment ID,run ID, key ID, destination port, number of packets, destination IP address,packet length and inter frame gap in microseconds (µs). Server takes portnumber, packet length and direction.

2.2 Calculation of OWD

The main objective of this thesis is to calculate the OWD. The OWD is thetime taken for a packet to travel from source (SRC) to destination (DST).When a packet (p) leaves from the sender, a time stamp (SRC Tp) is made.Similarly at the receiver side, a time stamp (DST Tp) is made. Then theOWD can be calculated as follows.

OWD = DSTTp − SRCTp (2.1)

The time stamps taken from SRC and DST are obtained from two differentclocks and the basic problem when evaluating this OWD is the synchroniza-tion problem [24]. So, for accurate measurements the clocks should be dulysynchronized. Some of the well known Synchronization methods are Net-work Time Protocol (NTP) and GPS [25, 26]. NTP can synchronize withinthe deviation of 10-20 ms for WAN and less than 1ms for LAN [6]. In theseexperiments DAG cards are used in sync with the GPS, which outcomes insynchronization range of 60 ns. In practice it is difficult to obtain precisetime stamps if we use different DAG cards for both sender and the receiver.To overcome this, a special wiretap scheme is used, which enables to capturetraffic on the same DAG card and obtain time stamp from the same clock.The method uses wiretaps and special wiring in combination with the DAG


cards [27], in order to acquire time stamps from the generated data. Thereare two wiretaps, namely wiretap A and wiretap B. Figure 2.2 shows thediagram of wiretap scheme.

Figure 2.2: Wiretap scheme

The process carried out in the wiretap scheme is as follows. When apacket leaves from the sender, it passes through wiretap A. Here, a copyof packet is made and sent to the DAG 1 on interface 0. Then the DAG1 gives a time stamp as SRC Tp to the packet. The packet travels acrossthe network and passes through wiretap B. Here, wiretap B makes a copyof the packet and sends it to the DAG 1 on interface 1. Then the DAG 1gives a time stamp as DST Tp to the packet. Similarly, when the packet issent from the receiver, it passes through wiretap B and a copy is made andsend to the DAG 0 on interface 1. The DAG 0 gives a time stamp as DSTTp. Then the packet again travels across the network and passes throughwiretap A and again here a copy is made and sent to the DAG 0 on interface0. The DAG 0 gives a time stamp as SRC Tp. With this scheme it is easierto collect the time stamps on the same DAG card resulting in more accuratetime stamps.

2.3 Delay components

From Figure 2.1, it is observed that the OWD is setup with two components,namely ∆1 and ∆2 . ∆1 refers to the OWD due to GW and ∆2 refers toOWD caused over the network. In order to evaluate the GW, we conductedanother set of experiments by replacing the 3G USB modem with a D-link DUB-E100 fast Ethernet USB adapter [28]. The main reason behindchoosing the D-link DUB-E100 fast Ethernet USB adapter is that the packetencounters the same behavior as that of the USB modem.


Figure 2.3: Experiment setup for GW evaluation

2.4 Experiment settings

The experiments are conducted for three different Swedish mobile opera-tor networks. Each experiment run is conducted for the duration of fiveminutes to collect the traces at network level for both upload and down-load. These experiments are performed from 8:00 to 18:00 hours during themonths of October, November, and December 2010. The experiment runsare generated at the application layer by taking payload sizes of 40, 232,488, 726, 1000, 1218 and 1416 bytes respectively. Each payload size is thentried with data rates of 8, 16, 32, 64, 128, 256 and 360 kbps for both uploadand download. We also used data rates of 512, 1024, 2048 and 3072 kbpsfor download. The main reason for choosing the data rates from 8 to 3072kbps is that, as we are considering the VoIP and video streaming applica-tions, the VoIP may utilize lower links i.e. 8kbps and for video streaming itmay utilize higher links i.e. around 3072 kbps in 3G mobile networks. So,we have chosen these data rates to check the variation from lower links tohigher links. The settings of the experiment configuration are shown in theupcoming tables. Here, in the settings, N represents number of packets andTs for interframe gap between each packet taken for each payload (PL) withrespect to data rates (R).


Table 2.1: Experiment configuration from 40 bytes to 232 bytesPL Rate Direction N Ts

[Bytes] [kbps] [µ s]

40 8 Upload/Download 7500 4000040 16 Upload/Download 15000 2000040 32 Upload/Download 30000 1000040 64 Upload/Download 60000 500040 128 Upload/Download 120000 250040 256 Upload/Download 240000 125040 360 Upload/Download 337500 88940 512 Download 480000 62540 1024 Download 960000 31340 2048 Download 1920000 15640 3072 Download 2880000 104














2.5 Analysis method

The network level traces which are collected by MP are stored in binary formi.e. in cap files. These cap files are further processed with trace analyzertool to convert it in to human readable form and stored in a file. The tracefile contains a pkt num (sequence of packet numbers at the DAG interface),DI(DAG capture interface), MPID, time stamp, PDU (Protocol Data Unit)length, IP header, source IP, destination IP, TCP sequence number andIP ID number. Based on this information, we developed an analysis tool,which takes the combination of TCP sequence number and IP ID number,for each packet and generates a unique ID for these two numbers. This isdone for both sending and receiving DAG interfaces. The unmatched uniqueID number is treated as packet loss and the matched ID number is furtherprocessed for OWD.

The percentage of packet loss (ρ) can be calculated by using the equa-tion 2.2. Here δ represents number of packets lost and N represents totalnumber of packets. With this, we were able to evaluate OWD and packetloss characteristics.

ρ =δ

N× 100 (2.2)

Results

21

Chapter 3

Results

This chapter reports the measurements obtained from experiments. To avoidqueuing or processing delays, we used minimum delay as base for calculatingtotal OWD. In the first section the results from the experiments conductedon GW are explained and further it is followed by the experiments conductedon 3G commercial mobile networks. The tables and graphs explained in thischapter are the main findings which are derived from the experiments. Theremaining experimental data is presented in the Appendix section.

3.1 Evaluation of Gateway

Regarding the experiments of 3G mobile operators, additional experimentswere also conducted on GW to find the affect of GW before performing ex-periments on 3G UMTS networks. The experiments on GW were conductedby using the same experiment settings, which were used for the 3G mobilenetworks.

The graphical representations of the minimum (min) OWD are presentedin Figure 3.1 as a three dimensional plot. The X-axis represents R in kbps,Y-axis represents PL in bytes and Z-axis as OWD in ms.

From Figure 3.1, it is evident that the min OWD increases as the pay-load size increases for data rates 8, 16, 32 kbps respectively. Whereas, forremaining data rates the min OWD exhibits an irregular pattern. This isobserved in both upload and download scenarios. We believe that this affectis due to sending large number of packets with short intervals. As the trans-mission time depends on packet size, the packets larger than one MaximumSegment Size (MSS) are divided in to multiple frames and transmissions arenot considered until the last frame is arrived to its destination [29].

Table 3.1 and Table 3.2 shows the statistics of min, maximum (max),mean and standard deviation (std) values for upload and download scenarios.

From Table 3.1 and 3.2, it is evident that the lowest min OWD occursto be around 0.141919 ms for upload and 0.266135 ms for download respec-

22

CHAPTER 3. RESULTS 23

Figure 3.1: Minimum OWD distribution in GW

Table 3.1: Statistics of GW from 256kbps to 360kbps, UploadPL R Min Max Mean Std

[Bytes] [kbps] [ms] [ms] [ms] [ms]

40 256 0.166834 1.672149 0.39702 0.110304232 256 0.32115 1.731277 0.693318 0.107718488 256 0.267625 1.754046 0.951263 0.10336726 256 0.155807 47.20736 0.936873 0.5996261000 256 0.282049 1.81514 1.09914 0.1568751218 256 0.855923 2.02173 1.349776 0.0752311416 256 1.455784 1.977444 1.549432 0.027302

40 360 0.177324 1.294852 0.396405 0.111468232 360 0.326872 1.359165 0.568968 0.095515488 360 0.228763 1.428902 0.952375 0.089751726 360 0.141919 1.957238 0.889076 0.4425241000 360 0.357807 1.804352 1.062546 0.1808911218 360 0.329614 2.011299 1.327286 0.1104771416 360 1.407445 2.213776 1.541128 0.033827


Table 3.2: Statistics of GW from 512kbps to 1024kbps, DownloadPL R Min Max Mean Std


40 512 0.292242 2.281308 0.649931 0.135227232 512 0.458002 1.596868 0.791848 0.09149488 512 0.434637 1.986861 1.076259 0.129457726 512 0.263512 2.680838 1.012816 0.4038841000 512 0.808179 2.064169 1.277775 0.1941061218 512 1.055896 2.336562 1.546258 0.1125641416 512 1.569867 2.45893 1.778311 0.060285

40 1024 0.292242 1.917899 0.712179 0.141934232 1024 0.461996 1.653969 0.794184 0.124886488 1024 0.536442 2.417267 1.118465 0.238427726 1024 0.266135 2.115607 0.991115 0.3916111000 1024 0.770748 2.101183 1.274251 0.1930871218 1024 1.176894 20.84619 1.53025 0.1701591416 1024 1.556993 2.470791 1.774893 0.058848

tively. So, this brings us to conclusion that the min OWD induced by theGW is less than 1 ms and there are some oddities observed in max OWD atdata rates 256 and 1024 kbps.

The max OWD’s observed are 47.20736 ms and 20.84619 ms, which arequite varying from the normal maximum delay distribution. By lookinginto the individual traces it is noticed that there is a large delay spike.Figure 3.2 and Figure 3.3 show the spike found in the delay distributionduring evaluation. To verify this we did the second experiment run forthat particular payload size with same experiment settings explained before.From the second run it is noticed that there is a normal delay distributionand hence we conclude that these spikes are due to some network artifact.


Figure 3.2: Identification of delay spike at 726 bytes in Upload, GW

Figure 3.3: Identification of delay spike at 1218 bytes in Download, GW

3.1.1 Packet loss

The investigation of packet loss is carried out along with the OWD. Figure3.4 shows the percentage of packets loss in GW. From Figure 3.4, it isclear that the percentage of packets loss is less than 1% and these can beconsidered negligible in all cases. So, there are no significant losses occurredduring GW experimentation.


Figure 3.4: Percentage of packet loss, GW

3.1.2 Identification of sawtooth pattern

We also identified a sawtooth pattern in some individual traces of GW.Figure 3.5 shows the sawtooth pattern identified in GW for payload size726bytes of data rate 256kbps. From Figure 3.5, a slow drift is observedwith a periodic hop of 1.5 ms.

In [10], the authors found the same behavior done on UDP. They con-cluded that this behavior is due to the affect of USB modem. This patternis observed in both upload and download scenarios. The packet transmittedfrom sender is entered into two different interfaces. In upload scenario thepacket enters into the GW NIC card and then travels through the USB in-terface. Whereas, in download the packet originated from the receiver entersinto the USB interface and travels through the GW NIC card. With thisit is clear that the packet travels along same path but experiences differenttreatment.


Figure 3.5: Identification of sawtooth pattern in GW, Upload

3.1.3 Acknowledgment packets

We also considered the acknowledgment packets which are captured in theopposite direction. Figure 3.6 shows the min OWD distribution of acknowl-edgment packets in GW. There are two scenarios in Figure 3.6, as it showsthe min OWD for both upload and download scenarios. Here, upload repre-sents the acknowledgment packets (reverse direction) captured for the pack-ets (forward direction) in upload scenario. Whereas, download representsthe acknowledgment packets captured for the packets in download scenario.From Figure 3.6, it is observed that the min OWD has a small variation be-tween upload and download scenarios, as the min OWD is large for uploadthan download scenario. We suppose that this affect is due to the Windowsoperating system and merits further investigation.

Figure 3.6: Minimum OWD distribution in acknowledgment packets, GW


3.2 Evaluation of 3G UMTS Networks

The evaluation of mobile networks with respect to OWD are carried outfor three different Swedish mobile operator networks namely Tele2 (OP1),Tre (OP2) and Telia (OP3). In these operators OP1 and OP3 share thesame Radio Access (RA) network and the OP2 operator uses different RAnetwork. The experiments are conducted between months of October andDecember, 2010 with the sender being stationary.

3.2.1 OWD in Upload

Figure 3.7 shows the min OWD distribution for mobile operators OP1, OP2and OP3. When we look into the individual graphs it is observed that themin OWD is high at lower data rates of 8 and 16 kbps respectively. It isnoticed that the min OWD is less than 100 ms in most cases for all datarates and payload sizes taken into consideration. It is also observed that athigher data rates the min OWD has lower values.

Figure 3.7: Minimum OWD distribution in mobile operators, Upload

Table 3.3, 3.4 and 3.5 show the statistics for min, max, mean and stdvalues for mobile operators. From statistics it is clear that the lowest minOWD occurs at 128 kbps for all operators.


Tab

le3.

3:S

tati

stic

sof

mob

ile

oper

ator

sfr

om8k

bp

sto

16kb

ps,

Up

load

PL

RO

P1

OP

2O

P3

[Byte

s][k

bp

s]M

inM

ax

Mea

nS

tdM

inM

axM

ean

Std

Min

Max

Mea

nS

td[m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s]

408

74.3

5355.

2792.0

218

.43

69.6

148

0.60

88.6

623

.20

85.4

233

6.84

97.7

920

.13

232

875.0

4436.

5998.1

822

.66

60.1

911

07.5

216

3.38

115.

1776

.24

529.

4010

1.97

36.4

548

88

73.7

4499.

43336.

1737

.28

78.5

668

9.32

332.

9134

.86

86.6

464

0.09

343.

7043

.92

726

8146.

92716.

27163.

5038

.71

70.4

137

5.75

153.

2918

.64

86.1

160

6.30

168.

3946

.21

1000

875.7

9408.

90200.

5317

.74

186.

3075

9.84

249.

4714

2.92

72.3

043

4.09

203

25.9

012

18

873.7

1302.

32218.

2414

.62

206.

8286

5.33

228.

1857

.50

208.

8786

7.55

225.

7547

.12

1416

874.8

1944.

72265.

7110

9.49

69.6

256

2.12

265.

7636

.74

248.

7977

3.72

263.

7137

.47

4016

55.6

0484.

69113.

8233

.65

69.8

035

0.50

112.

3225

.26

105.

5564

6.37

121.

0726

.81

232

1676.0

9357.

13127.

6420

.06

69.3

287

9.95

130.

2767

.30

77.9

937

6.96

129.

7024

.47

488

16123.

14383.

53137.

0221

.70

69.2

639

9.00

132.

2419

.44

96.6

037

32.3

726

9.26

217.

1472

616

76.2

0385.

85171.

9721

.34

70.7

938

8.98

154.

7821

.54

76.0

710

74.5

837

3.39

144.

4810

00

1676.3

0526.

38202.

3021

.30

70.1

942

6.83

204.

4528

.90

75.5

443

7.68

202.

5622

.68

1218

1684.0

7471.

32221.

7027

.85

70.1

944

0.13

223.

3228

.08

76.8

149

1.07

222.

3424

.55

1416

1673.0

7482.

08261.

2218

.82

200.

3373

1.74

266.

0835

.04

76.3

742

4.19

264.

5523

.86


Tab

le3.

4:S

tati

stic

sof

mob

ile

oper

ator

sfr

om32

kb

ps

to64

kb

ps,

Up

load

PL

RO

P1

OP

2O

P3

[Byte

s][k

bp

s]M

inM

ax

Mea

nS

tdM

inM

axM

ean

Std

Min

Max

Mea

nS

td[m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s]

4032

74.7

2496.

32229.

7532

.02

68.9

464

05.1

923

2.73

259.

1811

6.92

824.

8023

4.59

43.5

223

232

76.4

7516.

23235.

1042

.07

172

521.

1823

8.40

34.2

476

.51

634.

8923

4.95

46.1

448

832

64.9

0430.

60220.

8834

.15

70.7

959

5.81

220.

4039

.41

7321

94.9

028

7.94

283.

7372

632

77.9

1931.

61198.

7862

.52

70.2

915

16.5

020

6.43

85.4

576

.90

3997

.07

499.

1077

8.58

1000

3275.7

9556.

49210.

5940

.58

113.

6165

5.42

240.

3955

.73

76.3

749

6.85

238.

9856

.22

1218

3283.7

2506.

31259.

2849

.75

80.9

060

8.59

262.

4157

.79

76.2

350

6.37

231.

7529

.57

1416

3284.3

3505.

92260.

3520

.91

68.8

086

4.18

285.

4386

.81

76.6

350

7.38

281.

2453

.79

4064

65.3

8558.

33168.

5935

.26

70.5

657

4.55

167.

0634

.31

77.2

154

7.33

167.

8134

.09

232

6464.8

6707.

67174.

0234

.83

71.0

316

85.8

418

1.19

100.

1557

.79

631.

6921

2.04

84.2

948

864

74.3

7450.

81154.

8125

.45

60.0

859

4.05

156.

0827

.42

77.1

845

1.18

155.

6028

.46

726

6477.0

9587.

21154.

5530

.06

62.6

051

8.63

160.

0939

.63

57.9

110

49.7

515

4.02

41.4

010

00

6476.4

1562.

67185.

4735

.30

80.4

454

1.95

182.

2440

.00

77.5

256

2.65

187.

0337

.47

1218

6477.0

4483.

98179.

0237

.97

68.9

052

7.01

180.

0643

.33

85.7

159

0.60

181.

4938

.51

1416

6464.1

6753.

74151.

3640

.12

70.4

768

7.70

159.

1742

.04

85.5

555

0.15

199.

9055

.78


Tab

le3.

5:S

tati

stic

sof

mob

ile

oper

ator

sfr

om12

8kb

ps

to36

0kb

ps,

Up

load

PL

RO

P1

OP

2O

P3

[Byte

s][k

bp

s]M

inM

ax

Mea

nS

tdM

inM

axM

ean

Std

Min

Max

Mea

nS

td[m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s]

4012

865.5

9631.

53100.

3136

.62

59.5

159

5.04

91.9

735

.85

65.5

784

0.18

104.

5438

.49

232

128

64.5

6831.

7699.2

738

.66

56.1

413

14.7

811

2.40

125.

4064

.94

949.

7611

1.28

75.2

648

812

855.0

41368

.14

97.5

283

.65

59.2

669

4.19

90.6

939

.28

64.6

456

9.93

98.9

940

.40

726

128

56.2

41310

.99

103.

4393

.37

49.7

381

3.91

87.3

346

.74

54.5

713

70.8

111

7.73

125.

7410

00

128

64.0

61310

.32

106.

2086

.75

60.2

411

87.2

310

3.08

53.4

163

.13

1290

.37

100.

8879

.96

1218

128

64.8

4689.

36106.

3936

.11

61.3

682

0.13

112.

3147

.03

65.8

765

1.57

107.

7235

.83

1416

128

52.4

0888.

9791.7

443

.15

48.8

482

2.33

94.3

450

.79

77.0

883

2.83

109.

8943

.16

4025

675.6

81400

.67

120.

0485

.57

60.9

715

72.6

411

0.16

92.4

464

.64

1402

.37

118.

1880

.06

232

256

56.5

21398

.61

117.

3586

.97

49.7

485

2.84

105.

7946

.75

57.6

383

2.78

118.

4149

.09

488

256

63.7

2770.

11107.

0938

.22

60.0

515

24.4

011

5.44

99.0

960

.12

2921

.85

135.

9620

2.07

726

256

65.6

31439

.37

123.

4195

.31

58.7

914

84.5

611

1.62

85.8

656

.00

1311

.16

307.

7316

9.37

1000

256

68.5

4889.

04109.

6337

.59

59.2

795

0.90

115.

2244

.06

58.6

887

3.06

112.

5737

.63

1218

256

63.2

1690.

21109

35.6

357

.81

1912

.64

129.

3212

6.55

57.4

483

2.04

111.

9637

.00

1416

256

53.5

1560.

93112.

7033

.96

49.6

210

10.4

711

7.90

46.9

176

.51

610.

8411

4.57

36.5

0

4036

056.0

51012

.21

183.

2270

.86

49.9

380

5.67

233.

7311

4.42

57.7

913

51.6

522

9.35

100.

8123

236

053.3

61440

.18

305.

4512

8.16

60.7

610

95.7

328

1.40

120.

2257

.41

665.

3524

6.86

101.

8748

836

064.4

51147

.90

290.

8710

4.68

60.2

288

2.07

228.

3311

2.06

76.2

413

11.8

538

1.08

144.

7972

636

058.2

61310

.83

205.

4763

.56

71.6

420

45.2

637

2.53

152.

3456

.00

1311

.16

307.

7316

9.37

1000

360

64.1

71070

.25

188.

4365

.85

6383

1.74

290.

4390

.59

66.7

112

52.0

524

6.01

95.2

112

18

360

54.7

11069

.50

259.

6711

3.50

49.9

914

31.2

331

8.32

216.

9965

.26

831.

8930

0.65

140.

5314

16

360

64.4

01347

.81

335.

6919

0.54

70.2

333

00.0

464

6.23

600.

3793

.36

1480

.74

325.

7614

5.90


Also it is noticed that there is a large OWD at initial stage of trans-mission. Figure 3.8 shows the 5 minute duration of OWD distribution forpayload size 40 bytes of data rate 32 kbps in OP2.

Figure 3.8: OWD Distribution for 5 minute duration

As shown in Figure 3.8, the packets experienced large OWD at initialstage and suddenly dropped in the later stage. The reason behind thisproblem is that the GW queues the packets while acquiring a dedicatedcommunication channel. As the queue builds if the sending rate is higherthan the assigned communication channel. Thus a new channel is reassignedfor the packets, resulting in large delay. This behavior is observed in allpayload sizes and data rates within the three operators.

3.2.2 OWD in Download

When we focus on the download scenario, it is noticed that the min OWDdistribution is less than 50 ms for all operators. Figure 3.9 shows the minOWD distribution for download scenario. If we have closer look in thesegraphs, it is evident that the min OWD is approximately same for all payloadsizes and data rates. Interestingly, the min OWD is less in OP2 whencompared with OP1 and OP3. Table 3.6 and Table 3.7 shows the lowestvalues of min OWD, obtained at 512kbps for OP1 and 3072 kbps in OP2and OP3.

There are also some oddities observed in three operators, the max OWD’sare shown in Table 3.6 and Table 3.7. Especially in OP2 at data rate 2048kbps, the standard deviation is too high i.e. around 1400 ms and the maxOWD value is 8819 ms. To investigate this, we checked the individual trace.From the OWD trace it is noticed that a large delay spike is found at theinitial stage of transmission. Figure 3.10 shows the spike identified in OP2.


Figure 3.9: Minimum OWD distribution in mobile operators, Download

Figure 3.10: Identification of delay spike in OP2, Download


Tab

le3.6

:S

tati

stic

sof

mob

ile

oper

ator

sfr

om36

0kb

ps

to51

2kb

ps,

Dow

nlo

adPL

RO

P1

OP

2O

P3

[Byte

s][k

bp

s]M

inM

ax

Mea

nS

tdM

inM

axM

ean

Std

Min

Max

Mea

nS

td[m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s]

4036

037.2

886.4

551.2

36.

1934

.55

511.

9851

.64

18.7

538

.76

649.

7552

.73

11.2

123

236

035.1

2237.

1353.2

712

.02

29.8

619

26.3

054

.42

68.0

938

.81

92.3

551

.65

5.69

488

360

35.2

5263.

5157.7

816

.22

34.0

968

4.18

58.3

843

.22

38.6

795

.66

52.2

06.

7872

636

039.3

5246.

1359.0

417

.06

29.5

126

7.12

45.8

910

.02

38.9

730

1.46

53.6

910

.27

1000

360

36.0

1238.

2553.0

011

.19

30.3

653

1.16

45.6

97.

2137

.59

88.8

450

.93

6.66

1218

360

35.2

0241.

8250.9

89.

4528

.41

110.

9245

.43

6.23

37.9

341

7.56

51.9

17.

8314

16

360

32.5

0604.

6651.8

215

.13

28.9

080

.48

45.2

75.

9842

.18

753.

1052

.08

9.38

4051

238.3

3199.

3351.7

17.

8529

.88

1121

.79

51.9

935

.44

38.7

292

.10

53.7

96.

8623

251

236.4

3227.

8054.9

010

.92

30.8

343

69.6

818

2.26

464.

6140

.19

1294

.91

55.0

431

.77

488

512

35.5

6474.

8657.8

813

.66

29.6

711

10.0

273

.51

94.1

940

.41

89.7

251

.91

6.99

726

512

33.4

6222.

1852.5

38.

3833

.09

92.3

144

.95

5.99

38.8

610

8.20

52.5

67.

4910

00

512

38.5

1294.

6954.4

611

.71

28.8

987

.19

46.3

05.

3041

.15

201.

7453

.30

7.99

1218

512

32.4

8614.

4652.5

513

.58

28.4

218

1.12

47.1

07.

3238

.98

130.

5354

.15

8.80

1416

512

33.2

9641.

1765.2

937

.47

28.0

116

96.0

846

.06

16.7

342

.79

688.

3853

.37

9.22


Tab

le3.

7:S

tati

stic

sof

mob

ile

oper

ator

sfr

om10

24kb

ps

to30

72kb

ps,

Dow

nlo

adPL

RO

P1

OP

2O

P3

[Byte

s][k

bp

s]M

inM

ax

Mea

nS

tdM

inM

axM

ean

Std

Min

Max

Mea

nS

td[m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s]

4010

24

34.4

4587.

2452.8

48.

1830

.88

545.

3250

.18

10.2

641

.10

1535

57.4

531

.91

232

1024

35.3

6497.

9957.5

513

.11

29.9

667

74.1

918

539

7.87

40.2

312

2.21

55.1

18.

7148

810

24

36.0

1492.

6259.0

116

.01

30.6

790

67.7

230

2.36

896.

6740

.72

394.

6056

.75

10.2

172

610

24

35.0

5600.

1552.6

37.

1229

.40

81.6

747

.65

6.70

38.2

928

3.25

54.7

18.

6110

00

1024

34.3

5307.

6057.3

213

.84

32.4

474

.56

46.4

25.

0733

.71

620.

4957

.82

17.9

512

18

1024

34.6

5607.

3954.2

910

.52

27.7

022

1.84

46.8

76.

9040

.75

405.

5760

.25

22.8

814

16

1024

31.6

62545

.69

94.4

318

5.85

28.2

819

0.72

46.2

85.

3941

.17

97.2

851

.62

6.22

4020

48

34.8

9729.

2652.4

69.

3727

.93

347.

5054

.58

14.3

434

.64

2855

.75

54.2

124

.80

232

2048

34.7

3497.

5758.0

914

.40

29.4

788

19.8

697

6.65

1400

.20

39.3

710

76.5

063

.34

34.1

848

820

48

35.3

8478.

1854.3

59.

5132

.29

620.

7962

.82

37.6

841

.30

640.

2555

.41

11.1

772

620

48

32.9

4216.

9353.0

27.

2030

.41

175.

7951

.78

8.68

35.6

219

9.40

52.4

47.

3310

00

2048

35.1

7317.

2657.4

012

.48

31.4

619

6.38

46.3

95.

1541

.15

1115

.73

56.1

213

.13

1218

2048

36.1

0611.

5855.0

8411

.31

28.0

310

2.91

46.8

35.

4241

.90

1384

.40

72.6

065

.09

1416

2048

34.8

61605

.66

67.6

077

.69

28.3

183

.13

46.6

25.

1241

.91

487.

1255

.71

15.0

2

4030

72

33.8

2480.

2455.6

811

.86

29.6

135

8.33

58.6

923

.00

34.0

321

5.19

55.4

28.

7923

230

72

34.6

9477.

5557.8

113

.68

32.0

641

61.1

731

2.46

521.

4341

.46

583.

6956

.97

12.6

148

830

72

37.1

5130.

2451.4

08.

4930

.31

1011

9.79

56.4

746

.55

40.7

591

.99

52.5

67.

3172

630

72

35.1

2491.

6856.4

513

.16

29.8

839

3.48

53.5

517

.66

40.7

948

2.50

55.4

99.

7110

00

3072

35.2

3590.

6155.1

011

.17

27.1

351

8.01

46.7

55.

9840

.47

694.

2156

.86

24.0

812

18

3072

35.3

9488.

9054.8

510

.22

27.7

527

7.38

46.5

85.

5639

.51

274.

3851

.65

7.49

1416

3072

41.6

21784

.83

73.6

097

.57

29.2

653

0.38

46.6

95.

7341

.77

684.

8256

.12

21.1

1


3.2.3 Acknowledgment packets

Figure 3.11 shows the acknowledgment packets captured for upload scenario.From Figure 3.11, it is evident that the min OWD is varied around 30 msfor all the three operators. In order to investigate the impact of packet size,we compared the acknowledgment packets with the packets, as the acknowl-edgment packets are smaller than the normal packets. So, by doing this wecan find the impact between these two traffic types. Now, if we comparethese acknowledgment packets with the packets in download scenario (see inFigure 3.9), the min OWD is approximately same for both acknowledgmentpackets and packets. The reason behind comparing these acknowledgmentpackets with the min OWD of packets in download scenario is that the ac-knowledgment packets in upload traverse in same path to the packets indownload scenario. From the comparison it is clear that there is no muchdifference found in min OWD between the two traffic types. Similarly, thisbehavior is observed in download acknowledgment packets for all operators.So, this brings us to conclusion that there is no much impact on packet sizebetween acknowledgment packets and packets in mobile networks.

Figure 3.11: Acknowledgment packets in mobile operators, Upload

3.2.4 Segment Distribution

We also observed some interesting patterns in the segment distribution formobile operators. Figure 3.12 shows the flow of segments taken at networklayer for payload size 726 bytes in OP3. To get clear picture in the segment


distribution we have plotted 100 packet samples from the trace file. .

Figure 3.12: Segment distribution in OP3, Upload

From Figure 3.12 it is observed that the variability of packet size in-creases as the data rate increases. Interestingly, there is a huge drift foundin the transmission of packets, as the packets vary from higher to lower sizes.The reason behind this is that, during the data transmission the sender TCPwindow field adjusts the rate of flow of the byte stream i.e. being advertisedby the receiver. This behavior is observed in all payload sizes within thethree operators for both upload and download scenarios.

To give more insight we have considered the data rate for 256 kbpsin Figure 3.12 and plotted, that data rate as a Complementary CumulativeDistribution Function (CCDF) curve. Figure 3.13 shows the CCDF for datarate 256 kbps.

From Figure 3.13, we observed that the 10% of packets are transmittedwith less than 70 bytes segment size and 90% of the packets experiencedsegment size greater than 70 bytes. The curve goes linear by increasing thesegment size up to 1500 bytes, which is the MTU at network layer. And 80%of the packets experienced the segment size as less than 1350 bytes. Thisshows the variation of smaller segments and large segments. As 10% of thepackets are sent with smaller size and 80% with larger size i.e. around 1000bytes. In some cases, the segments are transmitted around 1450 bytes andalso reached the MTU i.e. 1500 bytes. It’s a fact that TCP sends numberof segments according to the receiver advertised window. So, here we canobserve that the TCP sends at most one or more segments depending on thereceiver advertised window. These kinds of observations are approximatelysimilar for remaining payload sizes and data rates within all three operators.


Figure 3.13: CCDF for payload size 726 bytes in OP3, Upload

3.2.5 Packet loss

With respect to OWD, packet loss is also carried out for the mobile opera-tors. Figure 3.14, 3.15 illustrate the percentage of packet loss for all payloadsizes and data rates within the three operators for both upload and down-load scenarios. There are some observations noticed from these graphs, asthe packet loss increases with the increase in payload size and data rate. So,this shows that the packet loss has much impact on higher data rates and forlarger payload sizes. Interestingly, the packet loss is stable up to 232 bytesand for data rates up to 64 kbps. Later it is increased with the increase inpayload size and data rates. This behavior is observed in both upload anddownload scenarios within all the three operators. Also the percentage ofpacket loss is high in download when compared with upload scenario. Asin upload the percentage of packet loss is around 5% and for download itis around 20%. The reason behind this is that, due to transmitting largenumber of packets with short intervals, huge losses are occurred in downloadscenario during the transmission. The number of packets lost are presentedin the Appendix section.


Figure 3.14: Percentage of packet loss in mobile operators, Upload

Figure 3.15: Percentage of packet loss in mobile operators, Download

Chapter 4

Conclusions

4.1 Conclusions

The OWD and packet loss characteristics over TCP performance for threedifferent Swedish mobile networks have been reported. The characteristicshas reported for both uplink and downlink applications. Interesting obser-vations were found in GW analysis. In GW, the min OWD increases asthe payload size increases for data rates 8kbps, 16kbps and 32kbps respec-tively. For remaining data rates the min OWD exhibits an irregular pattern.This is observed in both upload and download scenarios. Also a sawtoothpattern was observed in GW, which is caused due to the affect of USB mo-dem. Moreover, in mobile operators it is noticed that in upload scenario thepackets experienced larger OWD at initial stage of transmission. Anotherinteresting pattern was observed in download scenario, the min OWD is lessin OP2 when compared with OP1 and OP3. Also it is noticed that thereis no impact on acknowledgment packets in mobile operators, as the minOWD is approximately same for both packets and acknowledgment packets.Regarding packet loss the percentage of packet loss is high in larger datarates and payload sizes, as the packet loss increases with the increase inpayload size and data rate.

4.2 Future work

Our current study is based upon OWD on TCP behavior in 3G mobilenetworks. The future work will be to calculate RTT and the experimentswill be conducted with the same experimental settings, which were used forOWD. This can show the impact between these two metrics in 3G mobilenetworks.

40

Bibliography

[1] Third of Asia-Pacific 3G Subscribers Expected to usepc dongles, [Online; Verified June] Available:2009.http://www.cellular-news.com/story/35580.php.

[2] Patrik Arlos. On the Quality of Computer Network Measurements. PhDthesis, Blekinge Institute of Technology, 2005.

[3] Constantinos Dovrolis, editor. Passive and Active Network Measure-ment, 6th International Workshop, PAM 2005, Boston, MA, USA,March 31 - April 1, 2005, Proceedings, volume 3431 of Lecture Notesin Computer Science. Springer, 2005.

[4] Arvind Krishnamurthy and Bernhard Plattner, editors. Passive and Ac-tive Network Measurement, 11th Internatinoal Conference, PAM 2010,Zurich, Switzerland, April 7-9, 2010, Proceedings, volume 6932 of Lec-ture Notes in Computer Science. Springer, 2010.

[5] J.M. Cano-Garcia, E. Gonzalez-Parada, and E. Casilari. Experimentalanalysis and characterization of packet delay in umts networks. pages396 – 407, Berlin, Germany, 2006.

[6] Patrik Arlos and Markus Fiedler. Influence of the packet size on theone-way delay on the uplink in 3G networks. In PAM2010, pages 61–70,2010.

[7] L. Tomaciello, L. De Vito, and S. Rapuano. One-way delay measure-ment: State of art. In Instrumentation and Measurement TechnologyConference, 2006. IMTC 2006. Proceedings of the IEEE, pages 218–223, April 2006.

[8] F. Vacirca, F. Ricciato, and R. Pilz. Large-scale rtt measurements froman operational umts/gprs network. pages 190 – 7, Los Alamitos, CA,USA, 2005.

[9] Darryl Veitch, Satish Babu, and Attila Pasztor. Robust synchronizationof software clocks across the internet. In IMC ’04: Proceedings of the

41

BIBLIOGRAPHY 42

4th ACM SIGCOMM conference on Internet measurement, pages 219–232, New York, NY, USA, 2004. ACM.

[10] Ravichandra Kommalapati. Performance of 3G Data Services OverMobile Network in Sweden. Master’s thesis, Blekinge Institute of Tech-nology, 2010.

[11] K. Pentikousis, M. Palola, M. Jurvansuu, and P. Perala. Active goodputmeasurements from a public 3g/umts network. IEEE CommunicationsLetters, 9(9):802 – 4, Sept. 2005.

[12] Venkata Santosh Pemmaraju. Real-Time Live RTT Analyzer. Master’sthesis, Blekinge Institute of Technology, 2010.

[13] C.-F. Chiasserini and M. Meo. Impact of arq protocols on qos in 3gppsystems. IEEE Transactions on Vehicular Technology, 52(1):205 – 15,2003/01/.

[14] Prasad Calyam, Weiping Mandrawa, Mukundan Sridharan, Arif Khan,and Paul Schopis. H.323 beacon: an h.323 application related end-to-end performance troubleshooting tool. In Proceedings of the ACMSIGCOMM workshop on Network troubleshooting: research, theory andoperations practice meet malfunctioning reality, NetT ’04, pages 241–246, New York, NY, USA, 2004. ACM.

[15] C.M. White, E.J. Daniel, and K.A. Teague. A network performanceapplication for modeling, simulation, and characterization of packetnetwork behavior. volume Vol.1, pages 555 – 9, Piscataway, NJ, USA,2003.

[16] Jin hee Choi and Chuck Yoo. One-way delay estimation and its appli-cation. Computer Communications, 28:819–828, 2005.

[17] J. Prokkola, M. Hanski, M. Jurvansuu, and M. Immonen. Measuringwcdma and hsdpa delay characteristics with qosmet. In Communica-tions, 2007. ICC ’07. IEEE International Conference on, pages 492–498, june 2007.

[18] M. Jurvansuu, J. Prokkola, M. Hanski, and P. Perala. Hsdpa perfor-mance in live networks. In Communications, 2007. ICC ’07. IEEEInternational Conference on, pages 467 –471, june 2007.

[19] Tele2 mobile operator, [Online; Verified November] Available:2009.http://www.tele2.se.

[20] Tre mobile operator, [Online; Verified November] Available:2009.http://www.tre.se.

BIBLIOGRAPHY 43

[21] Telia mobile operator, [Online; Verified November] Available:2009.http://www.telia.se.

[22] Huawei E220 USB Modem, [Online; Verified November] Available:2009.http://www.huawei.com/mobileweb/en/products/view.do?id=282.

[23] Patrik Arlos, Markus Fiedler, and Arne A. Nilsson. A distributed pas-sive measurement infrastructure. In Dovrolis [3], pages 215–227.

[24] Patrik Arlos and Markus Fiedler. A method to estimate the times-tamp accuracy of measurement hardware and software tools. In SteveUhlig, Konstantina Papagiannaki, and Olivier Bonaventure, editors,PAM2007, volume 4427 of Lecture Notes in Computer Science, pages197–206. Springer, 2007.

[25] Network time protocol (version 3) specification, implementation, 1992.

[26] GPS Clocks for Computer Time Synchronization, [Online; VerifiedNovember] Available:2009. http://www.gpsclock.com/gps/g1.html.

[27] Endace Measurement Systems, [Online; Verified November] Avail-able:2009. http://www.endace.com.

[28] D-link DUB-E100 fast Ethernet USB adaptor, [Online; Verified Novem-ber] Available:2009. http://www.dlink.com/products/?pid=133.

[29] TCP throughput considerations, Available:2011.http://msdn.microsoft.com/en-us/library/ms819738.aspx.

Appendix

44

Appendix A

Experimental Results

A.1 Gateway

A.1.1 Upload

45

APPENDIX A. EXPERIMENTAL RESULTS 46

Table A.1: Statistics of GW from 8kbps to 16kbps, UploadPL R Min Max Mean Std


40 8 0.205755 1.329363 0.647857 0.083448232 8 0.443995 1.295745 0.829513 0.092665488 8 0.67395 1.353919 1.071583 0.095851726 8 0.933349 1.488745 1.291485 0.0896231000 8 1.148224 1.672864 1.264347 0.0433021218 8 1.301527 1.580179 1.466244 0.0396651416 8 1.526237 1.768232 1.651175 0.036837

40 16 0.172853 1.363039 0.585803 0.110876232 16 0.385225 1.232147 0.770757 0.130787488 16 0.637174 1.377642 1.064905 0.095476726 16 0.884473 1.427531 1.290352 0.0913221000 16 1.109302 1.354933 1.232437 0.0413941218 16 1.315236 2.038658 1.441494 0.0471451416 16 1.50764 1.752317 1.630215 0.039062

Table A.2: Statistics of GW from 32kbps to 128kbps, UploadPL R Min Max Mean Std


40 32 0.179351 1.322806 0.535076 0.090545232 32 0.366271 1.520396 0.764604 0.114641488 32 0.61506 1.42467 1.060235 0.09836726 32 0.890255 1.973391 1.29011 0.0967941000 32 1.083911 1.370549 1.205117 0.0396441218 32 1.294434 1.908362 1.412 0.0429751416 32 1.475811 2.258837 1.596472 0.048183

40 64 0.183642 1.208305 0.412946 0.079521232 64 0.378728 2.016127 0.780338 0.102141488 64 0.249803 1.772762 1.057672 0.101872726 64 0.841856 1.837075 1.238344 0.1340621000 64 1.075745 1.346707 1.182711 0.035571218 64 1.283705 2.113163 1.387657 0.0455831416 64 0.329614 2.33084 1.371337 0.138942

40 128 0.161648 6.540001 0.396639 0.091804232 128 0.372231 1.498163 0.712169 0.096462488 128 0.353992 1.828432 1.053542 0.099407726 128 0.418306 1.66738 1.08404 0.133781000 128 1.074433 1.583815 1.17207 0.0289571218 128 0.95129 2.13021 1.373852 0.0346241416 128 1.446009 1.852691 1.561269 0.028921


A.1.2 Download


Table A.3: Statistics of GW from 8kbps to 16kbps, DownloadPL R Min Max Mean Std


40 8 0.294924 1.478195 0.7495 0.111207232 8 0.53376 1.42175 0.969011 0.085555488 8 0.832975 1.833021 1.22038 0.085856726 8 1.021862 2.111196 1.413364 0.1012651000 8 1.268506 1.707673 1.508648 0.0789551218 8 1.483619 1.893222 1.708112 0.0752651416 8 1.656294 2.020359 1.892877 0.065792

40 16 0.289738 2.832174 0.682366 0.118232232 16 0.551939 1.591086 0.972536 0.081819488 16 0.782728 1.906633 1.220556 0.082036726 16 0.991762 1.729965 1.402272 0.0977881000 16 1.271427 1.734733 1.482552 0.0660831218 16 1.464069 2.490103 1.689915 0.0780941416 16 1.651645 2.427757 1.875065 0.07257



40 32 0.274539 1.753986 0.638675 0.10296232 32 0.510216 1.786351 0.971309 0.08806488 32 0.795126 1.877487 1.218288 0.079851726 32 0.941515 1.680255 1.390668 0.1081691000 32 1.230776 2.064586 1.457488 0.0702231218 32 1.428306 1.832068 1.657507 0.065181416 32 1.612008 2.359808 1.836233 0.073575

40 64 0.300884 1.415134 0.626394 0.088727232 64 0.520169 2.157628 0.96957 0.089276488 64 0.564039 1.894534 1.216722 0.087915726 64 0.471413 2.087772 1.381589 0.1123191000 64 1.225888 2.162159 1.436529 0.0622791218 64 1.399338 1.939535 1.635327 0.0619421416 64 1.583457 2.215922 1.807896 0.067727

40 128 0.299335 1.456619 0.630494 0.105831232 128 0.475824 1.75786 0.85811 0.113809488 128 0.563979 1.9297 1.214598 0.090129726 128 0.944436 2.117932 1.359615 0.1162811000 128 1.212895 2.093196 1.423973 0.0564471218 128 1.412332 2.300083 1.616679 0.0612671416 128 1.591206 2.185941 1.78939 0.064398




40 256 0.272751 1.847386 0.54372 0.085085232 256 0.384033 2.284586 0.797662 0.099018488 256 0.444889 1.984298 1.150976 0.105501726 256 0.262141 2.084851 1.115854 0.378021000 256 0.813246 2.056658 1.3443 0.1600611218 256 1.168966 2.34735 1.607933 0.063531416 256 1.578868 2.434075 1.783712 0.05888

40 360 0.268221 1.544178 0.559798 0.105977232 360 0.389695 1.629412 0.81241 0.120375488 360 0.543773 1.915991 1.107181 0.123442726 360 0.275195 2.113998 1.035017 0.4103351000 360 0.782907 2.113104 1.310004 0.183011218 360 1.128793 2.237559 1.590179 0.1014521416 360 1.523674 2.424002 1.774649 0.066514



40 2048 0.434101 2.21926 1.05243 0.138848232 2048 0.439882 2.091825 1.050924 0.172093488 2048 0.696897 2.202928 1.063187 0.169915726 2048 0.349581 2.114177 1.248325 0.0986861000 2048 0.446319 2.328038 1.270288 0.185281218 2048 1.154125 2.462447 1.524466 0.107351416 2048 1.54382 2.649546 1.772554 0.062764

40 3072 0.554144 2.577066 1.682594 0.221445232 3072 0.457585 3.488719 1.756847 0.199992488 3072 0.427664 2.67762 1.380825 0.313334726 3072 0.311911 2.673388 1.537201 0.3553071000 3072 0.943541 2.644301 1.498177 0.174441218 3072 0.961304 3.231108 1.628928 0.0971471416 3072 1.5046 5.472958 1.767829 0.07608


A.2 Mobile operators

A.2.1 Download


Tab

leA

.7:

Sta

tist

ics

ofm

obil

eop

erat

ors

from

8kb

ps

to16

kb

ps,

Dow

nlo

adsP

LR

OP

1O

P2

OP

3[B

yte

s][k

bp

s]M

inM

ax

Mea

nS

tdM

inM

axM

ean

Std

Min

Max

Mea

nS

td[m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s]

408

34.6

783.5

342

.99

7.09

329

.15

520.

9654

.10

81.0

540

.48

598.

8345

.97

15.0

823

28

32.7

694.7

140

.82

5.66

30.1

513

4.90

40.4

48.

1436

.35

654.

6646

.26

18.5

848

88

36.1

372.7

645

.53

6.79

31.4

622

8.22

42.8

512

.43

38.2

385

.28

46.9

86.

2172

68

38.1

5613.

3248

.29

28.8

129

.69

56.9

842

.02

5.32

38.4

172

.31

47.1

04.

9310

00

835.9

281.6

249

.28

7.44

30.9

950

1.59

46.2

326

.93

42.2

399

.17

54.7

57.

3012

18

835.5

2168.

8849

.74

10.5

729

.77

344.

3943

.42

20.0

041

.36

80.5

556

.33

6.23

1416

835.3

6504.

0763

.70

54.0

230

.54

65.7

544

.91

6.15

41.6

210

3.96

52.1

18.

07

4016

35.5

7626.

3742

.73

15.1

729

.92

61.0

041

.26

4.79

40.7

874

.75

45.6

84.

3623

216

34.8

377.0

439

.48

5.39

29.3

724

0.54

40.7

110

.86

40.6

465

5.84

46.7

915

.48

488

1634.9

7599.

9745

.37

17.3

131

.11

136.

3041

.27

6.52

35.9

184

.09

44.8

36.

0872

616

37.3

0141.

8846

.72

6.92

33.6

512

0.83

42.3

46.

2639

.63

72.5

547

.52

5.12

1000

1636.3

583.1

348

.18

7.31

34.1

258

.36

43.5

25.

0639

.55

80.8

150

.16

6.32

1218

1634.6

1493.

1758

.63

45.6

331

.21

59.3

743

.77

6.05

40.8

349

8.42

52.2

421

.40

1416

1635.5

2189.

8756

.52

17.5

129

.60

71.1

744

.57

5.79

44.0

087

.81

53.2

46.

34


Tab

leA

.8:

Sta

tist

ics

ofm

obil

eop

erat

ors

from

32kb

ps

to64

kb

ps,

Dow

nlo

adsP

LR

OP

1O

P2

OP

3[B

yte

s][k

bp

s]M

inM

ax

Mea

nS

tdM

inM

axM

ean

Std

Min

Max

Mea

nS

td[m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s]

4032

35.8

067.6

149

.34

4.44

31.5

463

.34

44.9

44.

7841

.92

76.0

249

.48

5.51

232

3235.3

3464.

9445

.78

13.3

630

.04

62.6

742

.99

5.36

41.2

268

0.41

49.9

420

.93

488

3234.1

6624.

6444

.33

16.9

530

.12

541.

9743

.29

12.4

939

.15

583.

6956

.68

12.3

272

632

35.0

3168.

1348

.37

9.29

29.7

721

8.94

43.9

78.

8439

.19

91.9

952

.44

7.32

1000

3236.9

897.0

851

.09

6.82

35.0

166

.89

44.8

44.

8839

.46

95.8

050

.18

6.76

1218

3238.6

0648.

2953

.75

32.7

760

.984

.41

50.8

67.

2140

.92

148.

4051

.78

7.93

1416

3239.0

9691.

7453

.89

26.9

729

.55

82.8

945

.18

6.32

42.9

376

.55

52.5

15.

69

4064

35.4

1328.

0658

.03

11.3

629

.99

72.3

944

.85

4.82

40.2

567

5.08

50.7

221

.01

232

6436.2

8211.

9756

.11

13.0

938

.22

540.

6454

.24

21.7

842

.13

375.

0854

.82

13.9

448

864

34.2

6204.

5651

.14

9.98

29.3

330

0.19

48.3

011

.51

33.9

274

.81

49.3

95.

8972

664

34.4

8554.

4048

.37

14.6

429

.31

141.

4343

.36

6.29

41.2

724

7.85

50.3

610

.01

1000

6435.3

2230.

5552

.04

9.38

30.1

791

.53

44.7

05.

1634

.14

103.

9652

.03

8.49

1218

6435.5

1167.

6052

.57

11.3

030

.59

70.2

945

.41

5.60

41.6

181

.32

52.4

36.

6114

16

6433.8

5359.

3451

.12

13.6

829

.43

66.0

345

.22

5.81

45.8

269

8.36

66.2

421

.83


Tab

leA

.9:

Sta

tist

ics

ofm

obil

eop

erat

ors

from

128k

bp

sto

256k

bp

s,D

own

load

sPL

RO

P1

OP

2O

P3

[Byte

s][k

bp

s]M

inM

ax

Mea

nS

tdM

inM

axM

ean

Std

Min

Max

Mea

nS

td[m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s][m

s]

4012

842.4

5156.

3955

.22

8.21

39.8

450

4.86

47.7

512

.04

42.3

162

8.65

49.3

412

.95

232

128

35.8

7316.

8256

.76

13.6

529

.61

1011

.93

58.7

124

.91

41.5

188

.61

57.7

16.

6848

812

834.6

5613.

2653

.24

16.5

129

.81

198.

8748

.99

25.1

840

.53

82.7

849

.77

5.72

726

128

34.4

5200.

0152

.63

8.93

28.9

684

.70

48.8

67.

1741

.56

499.

7558

.48

17.5

110

00

128

34.8

2281.

7651

.81

10.3

329

.65

71.6

345

.03

6.25

38.3

064

1.09

56.5

015

.77

1218

128

36.6

9646.

6857

.21

18.8

432

.34

81.5

048

.70

6.87

35.4

269

4.21

56.9

224

.02

1416

128

33.4

8755.

8550

.61

19.5

727

.75

277.

3846

.23

5.91

45.3

633

6.33

60.3

710

.21

4025

640.3

0123.

2152

.38

7.43

29.2

555

9.10

51.7

225

.92

39.5

610

8.13

51.6

16.

4823

225

635.8

2229.

7253

.84

12.4

129

.87

1386

.30

51.8

450

.41

39.0

410

1.70

50.7

76.

2848

825

635.8

61802

.860

.57

69.6

129

.53

852.

3465

.18

73.2

839

.19

274.

7451

.42

7.87

726

256

36.0

0149.

2752

.83

8.37

29.4

323

1.08

46.2

76.

9137

.74

94.3

250

.78

7.34

1000

256

35.1

5248.

8152

.34

10.0

230

.87

72.8

646

.10

5.33

34.1

695

.50

50.8

06.

0812

18

256

34.7

5219.

5349

.66

7.82

28.8

011

9.68

46.0

67.

1641

.72

114.

6652

.08

6.87

1416

256

32.5

6247.

7949

.92

12.1

328

.79

71.3

044

.00

7.35

41.4

494

.96

55.4

07.

86


A.2.2 Packet loss


Table A.10: Packet loss from 40 bytes to 1000 bytes, UploadPL R N δ

[Bytes] [kbps] OP1 OP2 OP3

40 8 7500 1 0 140 16 15000 1 1 140 32 30000 0 3 340 64 60000 40 32 340 128 120000 12 29 1240 256 240000 180 113 12740 360 337500 205 162 227

232 8 1293 0 0 0232 16 2586 1 3 1232 32 5172 12 6 8232 64 10345 7 5 185232 128 20690 15 22 25232 256 41379 140 87 182232 360 58190 155 159 161

488 8 615 0 1 0488 16 1230 3 2 2488 32 2459 2 3 6488 64 4918 6 8 3488 128 9836 10 13 18488 256 19672 171 146 164488 360 27664 166 198 172

726 8 413 0 0 0726 16 827 1 1 3726 32 1653 4 12 8726 64 3306 2 45 3726 128 6612 20 18 96726 256 13223 185 126 190726 360 18595 175 147 171

1000 8 300 1 1 11000 16 600 0 11 01000 32 1200 4 3 51000 64 2400 2 43 31000 128 4800 49 14 141000 256 9600 53 245 1581000 360 13500 171 175 173


Table A.11: Packet loss from 1218 bytes to 1416 bytes, UploadPL R N δ


1218 8 246 1 0 01218 16 493 0 2 11218 32 985 5 3 01218 64 1970 4 23 61218 128 3941 62 115 101218 256 7882 158 169 1991218 360 11084 148 237 166

1416 8 212 2 1 01416 16 424 0 0 41416 32 847 2 4 61416 64 1695 4 4 81416 128 3390 56 60 1001416 256 6780 411 305 1531416 360 9534 161 243 145


Table A.12: Packet loss from 40 bytes to 488 bytes, DownloadPL R N δ


40 8 7500 1 0 040 16 15000 1 3 040 32 30000 0 1 340 64 60000 0 1 440 128 120000 1 1 140 256 240000 44 2 7240 360 337500 243 9 19740 512 480000 937 5 60140 1024 960000 3205 87 270840 2048 1920000 8282 720 528440 3072 2880000 9844 1931 7840

232 8 1293 1 0 1232 16 2586 0 0 1232 32 5172 1 0 0232 64 10345 3 2 2232 128 20690 2 2806 2232 256 41379 0 7642 121232 360 58190 99 11 261232 512 82759 188 141 785232 1024 165517 1015 992 2497232 2048 331034 3212 2675 4343232 3072 496552 6238 3921 7016

488 8 615 0 0 1488 16 1230 1 1 0488 32 2459 0 1 1488 64 4918 2 1 1488 128 9836 2 1 1488 256 19672 86 2 86488 360 27664 395 5 182488 512 39334 1345 14 792488 1024 78689 3201 64 2313488 2048 157377 6201 1035 5088488 3072 236066 9838 1927 8300




726 8 413 1 1 1726 16 827 1 0 0726 32 1653 0 1 0726 64 3306 1 1 1726 128 6612 1 2 1726 256 13223 101 5 97726 360 18595 310 30 248726 512 26446 979 113 604726 1024 52893 3048 712 2598726 2048 105785 6146 1531 5551726 3072 158678 9472 2772 7828

1000 8 300 0 0 01000 16 600 0 1 11000 32 1200 0 1 11000 64 2400 1 2 31000 128 4800 0 15 81000 256 9600 95 294 871000 360 13500 348 876 2601000 512 19200 982 1600 8261000 1024 38400 3258 3939 27761000 2048 76800 5988 7981 48421000 3072 115200 10300 12379 7680




1218 8 246 0 1 11218 16 493 1 0 01218 32 985 1 1 11218 64 1970 0 4 31218 128 3941 8 11 531218 256 7882 108 379 701218 360 11084 272 825 2251218 512 15764 1245 1457 5421218 1024 31527 3234 4024 23991218 2048 63054 6229 7935 50741218 3072 94581 9588 11955 12686

1416 8 212 2 2 01416 16 424 0 5 11416 32 847 2 2 01416 64 1695 3 8 31416 128 3390 8 262 71416 256 6780 179 440 4401416 360 9534 339 1014 7711416 512 13559 934 1516 15321416 1024 27119 2893 3887 42951416 2048 54237 5742 7785 78741416 3072 81356 7753 12602 12646

Evaluation of TCP Performance in 3G Mobile Networks

Documents

ieee international

gw nic card

mobile broadband

network time

ip id number

mobile operator

active network

3g data services