List of Figures

1. Network traffic seen from different levels.
2. An a-b-t diagram illustrating a persistent HTTP connection.
3. A diagram illustrating the interaction between two BitTorrent peers.
4. Overview of Source-level Trace Replay.
5. An a-b-t diagram representing a typical ADU exchange in HTTP version 1.0.
6. An a-b-t diagram illustrating a persistent HTTP connection.
7. An a-b-t diagram illustrating an SMTP connection.
8. Three a-b-t diagrams representing three different types of NNTP interactions.
9. An a-b-t diagram illustrating a server push from a webcam using a persistent HTTP connection.
10. An a-b-t diagram illustrating Icecast audio streaming in a TCP connection.
11. Three a-b-t diagrams of connections taking part in the interaction between an FTP client and an FTP server.
12. An a-b-t diagram illustrating an NNTP connection in ``stream-mode'', which exhibits data exchange concurrency.
13. An a-b-t diagram illustrating the interaction between two BitTorrent peers.
14. A first set of TCP segments for the connection vector in Figure 3.1: lossless example.
15. A second set of TCP segments for the connection vector in Figure 3.1: lossy example.
16. Distributions of ADU sizes for the testbed experiments with synthetic applications.
17. Distributions of quiet time durations for the testbed experiments with synthetic applications.
18. Distributions of ADU sizes for the testbed experiments with synthetic applications.
19. Distributions of quiet time durations for the testbed experiments with synthetic applications.
20. Bodies of the A and B distributions for Abilene-I, Leipzig-II and UNC 1 PM.
21. Tails of the A and B distributions for Abilene-I, Leipzig-II and UNC 1 PM.
22. Bodies of the A and B distributions with per-byte probabilities for Abilene-I, Leipzig-II and UNC 1 PM
23. Bodies of the E distributions for Abilene-I, Leipzig-II and UNC 1 PM.
24. Bodies of the E distributions with per-byte probabilities for Abilene-I, Leipzig-II and UNC 1 PM.
25. Tails of the E distributions for Abilene-I, Leipzig-II and UNC 1 PM.
26. Average size of the epochs in each connection vector as a function of the number of epochs for Abilene-I, Leipzig-II and UNC 1 PM.
27. Average of the median size of the ADUs in each connection vector as a function of the number of epochs for Abilene-I, Leipzig-II and UNC 1 PM.
28. Average of the median size of the ADUs in each connection vector as a function of the number of epochs, for Leipzig-II.
29. Average of the median size of the ADUs in each connection vector as a function of the number of epochs for Abilene-I.
30. Bodies of the TA and TB distributions for Abilene-I, Leipzig-II and UNC 1 PM.
31. Tails of the TA and TB distributions for Abilene-I, Leipzig-II and UNC 1 PM.
32. Distribution of the durations of the quiet times between the final ADU and connection termination.
33. Bodies of the A and B distributions for the concurrent connections in Abilene-I, Leipzig-II and UNC 1 PM.
34. Tails of the A and B distributions for the concurrent connections in Abilene-I, Leipzig-II and UNC 1 PM.
35. Bodies of the TA and TB distributions for the concurrent connections in Abilene-I, Leipzig-II and UNC 1 PM.
36. Tails of the TA and TB distributions for the concurrent connections in Abilene-I, Leipzig-II and UNC 1 PM.
37. Bodies of the A distributions for UNC 1 AM, UNC 1 PM and UNC 7:30 PM.
38. Bodies of the B distributions for UNC 1 AM, UNC 1 PM and UNC 7:30 PM.
39. Bodies of the TB distributions for UNC 1 AM, UNC 1 PM and UNC 7:30 PM.
40. Tails of the TB distributions for UNC 1 AM, UNC 1 PM and UNC 7:30 PM.
41. Bodies of the TA distributions for three UNC traces.
42. Tails of the TA distributions for three UNC traces.
43. A set of TCP segments illustrating RTT estimation from connection establishment.
44. Two sets of TCP segments illustrating RTT estimation ambiguities in the presence of loss and early retransmission in connection establishment.
45. A set of TCP segments illustrating RTT estimation using the sum of two OSTTs.
46. A set of TCP segments illustrating the impact of delayed acknowledgments on OSTTs.
47. Comparison of RTT estimators for a synthetic trace: no loss and enabled delayed acknowledgments.
48. Comparison of RTT estimators for a synthetic trace: no loss and disabled delayed acknowledgments.
49. Comparison of RTT estimators for a synthetic trace: fixed loss rate of 1% for all connections.
50. Comparison of RTT estimators for a synthetic trace: loss rates uniformly distributed between 0% and 10%.
51. A set of TCP segments illustrating an invalid OSTT sample due to the interaction between loss and cumulative acknowledgments.
52. Comparison of RTT estimators for a synthetic trace: loss rates uniformly distributed between 0% and 10%.
53. Comparison of RTT estimators for synthetic traces: fixed loss rate of 1%; real RTTs up to 4 seconds.
54. Bodies of the RTT distributions for the five traces.
55. Bodies of the RTT distributions with per-byte probabilities for the five traces.
56. Comparison of the sum-of-minima and sum-of-medians RTT estimators for UNC 1 PM.
57. Comparison of the sum-of-minima and sum-of-medians RTT estimators for Leipzig-II.
58. Bodies of the distributions of maximum receiver window sizes for the five traces.
59. Bodies of the distributions of maximum receiver window sizes with per-byte probabilities for the five traces.
60. Measured loss rates from experiments with 1% loss rates applied only on one direction or on both directions of the TCP connections.
61. Bodies of the distributions of loss rates for the five traces.
62. Bodies of the distributions of loss rates with per-byte probabilities for the five traces.
63. Breakdown of the byte throughput time series for Leipzig-II inbound.
64. Breakdown of the packet throughput time series for Leipzig-II inbound.
65. Breakdown of the byte throughput time series for Leipzig-II outbound.
66. Breakdown of the packet throughput time series for Leipzig-II outbound.
67. Breakdown of the byte throughput time series for Leipzig-II outbound.
68. Breakdown of the packet throughput time series for Leipzig-II outbound.
69. Breakdown of the byte throughput time series for Abilene-I Ipls/Clev.
70. Breakdown of the packet throughput time series for Abilene-I Ipls/Clev.
71. Breakdown of the byte throughput time series for Abilene-I Clev/Ipls.
72. Breakdown of the packet throughput time series for Abilene-I Clev/Ipls.
73. Breakdown of the byte throughput time series for UNC 1 PM inbound.
74. Breakdown of the packet throughput time series for UNC 1 PM inbound.
75. Breakdown of the byte throughput time series for UNC 1 PM outbound.
76. Breakdown of the packet throughput time series for UNC 1 PM outbound.
77. Breakdown of the byte throughput time series for the three UNC traces.
78. Breakdown of the packet throughput time series for the three UNC traces.
79. Byte throughput marginals of Leipzig-II inbound, its normal distribution fit, the marginal distribution of its Poisson arrival fit, and the normal distribution fit of this Poisson arrival fit.
80. Packet throughput marginals of Leipzig-II inbound, its normal distribution fit, the marginal distribution of its Poisson arrival fit, and the normal distribution fit of this Poisson arrival fit.
81. Byte throughput marginals of UNC 1 PM outbound, its normal distribution fit, the marginal distribution of its Poisson arrival fit, and the normal distribution fit of this Poisson arrival fit.
82. Packet throughput marginals of UNC 1 PM outbound, its normal distribution fit, the marginal distribution of its Poisson arrival fit, and the normal distribution fit of this Poisson arrival fit.
83. Quantile-quantile plots with simulation envelops for the marginal distribution of Leipzig-II inbound. The top four plots show byte throughput, while the four bottom plots show packet throughput.
84. Quantile-quantile plots with simulation envelops for the marginal distribution of UNC 1 PM outbound. The top four plots show byte throughput, while the four bottom plots show packet throughput.
85. Wavelet spectra of the packet throughput time series for Leipzig-II inbound and its Poisson arrival fit.
86. Wavelet spectra of the byte throughput time series for Leipzig-II inbound and its Poisson arrival fit.
87. Wavelet spectra of the packet throughput time series for Abilene-I.
88. Wavelet spectra of the byte throughput time series for Abilene-I.
89. Wavelet spectra of the packet throughput time series for UNC 1 PM.
90. Wavelet spectra of the byte throughput time series for UNC 1 PM.
91. Breakdown of the active connections time series for Leipzig-II.
92. Impact of the definition of active connection on Leipzig-II.
93. Breakdown of the active connections time series for Abilene-I.
94. Impact of the definition of active connection on Abilene-I.
95. Breakdown of active connections time series for UNC 1 PM using both definitions of active connection.
96. Impact of the time-of-day on the active connections time series for the three UNC traces.
97. Overview of Source-level Trace Replay.
98. Diagram of the network testbed where the experiments of this dissertation were conducted.
99. End-host architecture of the traffic generation system.
100. Bodies and tails of the A distributions for Leipzig-II and its source-level trace replays.
101. Bodies and tails of the B distributions for Leipzig-II and its source-level trace replays.
102. Bodies and tails of the E distributions for Leipzig-II and its source-level trace replays.
103. Bodies and tails of the TA distributions for Leipzig-II and its source-level trace replays.
104. Bodies and tails of the TB distributions for Leipzig-II and its source-level trace replays.
105. Bodies of the round-trip time and receiver window size distributions for Leipzig-II and its source-level trace replays.
106. Bodies the loss rate distributions for Leipzig-II and its source-level trace replays, with probabilities computed per connection (left) and per byte (right).
107. Bodies and tails of the A distributions for UNC 1 PM and its source-level trace replays.
108. Bodies and tails of the B distributions for UNC 1 PM and its source-level trace replays.
109. Bodies and tails of the E distributions for UNC 1 PM and its source-level trace replays.
110. Bodies and tails of the TA distributions for UNC 1 PM and its source-level trace replays.
111. Bodies and tails of the TB distributions for UNC 1 PM and its source-level trace replays.
112. Bodies of the round-trip time and receiver window size distributions for UNC 1 PM and its source-level trace replays.
113. Bodies of the loss rate distributions for UNC 1 PM and its source-level trace replays, with probabilities computed per connection (left) and per byte (right).
114. Bodies and tails of the A distributions for Abilene-I and its source-level trace replays.
115. Bodies and tails of the B distributions for Abilene-I and its source-level trace replays.
116. Bodies and tails of the E distributions for Abilene-I and its source-level trace replays.
117. Bodies and tails of the TA distributions for Abilene-I and its source-level trace replays.
118. Bodies and tails of the TB distributions for Abilene-I and its source-level trace replays.
119. Bodies of the round-trip time and receiver window size distributions for Abilene-I and its source-level trace replays.
120. Bodies of the loss rate distributions for Abilene-I and its source-level trace replays, with probabilities computed per connection (left) and per byte (right).
121. Byte throughput time series for Leipzig-II inbound and its four types of source-level trace replay.
122. Byte throughput time series for Leipzig-II outbound and its four types of source-level trace replay.
123. Packet throughput time series for Leipzig-II inbound and its four types of source-level trace replay.
124. Packet throughput time series for Leipzig-II outbound and its four types of source-level trace replay.
125. Byte throughput marginals for Leipzig-II inbound and its four types of source-level trace replay.
126. Byte throughput marginals for Leipzig-II outbound and its four types of source-level trace replay.
127. Packet throughput marginals for Leipzig-II inbound and its four types of source-level trace replay.
128. Packet throughput marginals for Leipzig-II outbound and its four types of source-level trace replay.
129. Wavelet spectra of the byte throughput time series for Leipzig-II inbound and its four types of source-level trace replay.
130. Wavelet spectra of the byte throughput time series for Leipzig-II outbound and its four types of source-level trace replay.
131. Wavelet spectra of the packet throughput time series for Leipzig-II inbound and its four types of source-level trace replay.
132. Wavelet spectra of the packet throughput time series for Leipzig-II outbound and its four types of source-level trace replay.
133. Active connection time series for Leipzig-II and its four types of source-level trace replay.
134. Byte throughput time series for UNC 1 PM inbound and its four types of source-level trace replay.
135. Byte throughput time series for UNC 1 PM outbound and its four types of source-level trace replay.
136. Packet throughput time series for UNC 1 PM inbound and its four types of source-level trace replay.
137. Packet throughput time series for UNC 1 PM outbound and its four types of source-level trace replay.
138. Byte throughput marginals for UNC 1 PM inbound and its four types of source-level trace replay.
139. Byte throughput marginals for UNC 1 PM outbound and its four types of source-level trace replay.
140. Packet throughput marginals for UNC 1 PM inbound and its four types of source-level trace replay.
141. Packet throughput marginals for UNC 1 PM outbound and its four types of source-level trace replay.
142. Wavelet spectra of the byte throughput time series for UNC 1 PM inbound and its four types of source-level trace replay.
143. Wavelet spectra of the byte throughput time series for UNC 1 PM outbound and its four types of source-level trace replay.
144. Wavelet spectra of the packet throughput time series for UNC 1 PM inbound and its four types of source-level trace replay.
145. Wavelet spectra of the packet throughput time series for UNC 1 PM outbound and its four types of source-level trace replay.
146. Active connection time series for UNC 1 PM and its four types of source-level trace replay.
147. Byte throughput time series for UNC 1 AM inbound and its four types of source-level trace replay.
148. Byte throughput time series for UNC 1 AM outbound and its four types of source-level trace replay.
149. Packet throughput time series for UNC 1 AM inbound and its four types of source-level trace replay.
150. Packet throughput time series for UNC 1 AM outbound and its four types of source-level trace replay.
151. Byte throughput marginals for UNC 1 AM inbound and its four types of source-level trace replay.
152. Byte throughput marginals for UNC 1 AM outbound and its four types of source-level trace replay.
153. Packet throughput marginals for UNC 1 AM inbound and its four types of source-level trace replay.
154. Packet throughput marginals for UNC 1 AM outbound and its four types of source-level trace replay.
155. Wavelet spectra of the byte throughput time series for UNC 1 AM inbound and its four types of source-level trace replay.
156. Wavelet spectra of the byte throughput time series for UNC 1 AM outbound and its four types of source-level trace replay.
157. Wavelet spectra of the packet throughput time series for UNC 1 AM inbound and its four types of source-level trace replay.
158. Wavelet spectra of the packet throughput time series for UNC 1 AM outbound and its four types of source-level trace replay.
159. Active connection time series for UNC 1 AM and its four types of source-level trace replay.
160. Byte throughput time series for UNC 7:30 PM inbound and its four types of source-level trace replay.
161. Byte throughput time series for UNC 7:30 PM outbound and its four types of source-level trace replay.
162. Packet throughput time series for UNC 7:30 PM inbound and its four types of source-level trace replay.
163. Packet throughput time series for UNC 7:30 PM outbound and its four types of source-level trace replay.
164. Byte throughput marginals for UNC 7:30 PM inbound and its four types of source-level trace replay.
165. Byte throughput marginals for UNC 7:30 PM outbound and its four types of source-level trace replay.
166. Packet throughput marginals for UNC 7:30 PM inbound and its four types of source-level trace replay.
167. Packet throughput marginals for UNC 7:30 PM outbound and its four types of source-level trace replay.
168. Wavelet spectra of the byte throughput time series for UNC 7:30 PM inbound and its four types of source-level trace replay.
169. Wavelet spectra of the byte throughput time series for UNC 7:30 PM outbound and its four types of source-level trace replay.
170. Wavelet spectra of the packet throughput time series for UNC 7:30 PM inbound and its four types of source-level trace replay.
171. Wavelet spectra of the packet throughput time series for UNC 7:30 PM outbound and its four types of source-level trace replay.
172. Active connection time series for UNC 7:30 PM and its four types of source-level trace replay.
173. Byte throughput time series for Abilene-I Clev/Ipls and its four types of source-level trace replay.
174. Byte throughput time series for Abilene-I Ipls/Clev and its four types of source-level trace replay.
175. Packet throughput time series for Abilene-I Clev/Ipls and its four types of source-level trace replay.
176. Packet throughput time series for Abilene-I Ipls/Clev and its four types of source-level trace replay.
177. Byte throughput marginals for Abilene-I Clev/Ipls and its four types of source-level trace replay.
178. Byte throughput marginals for Abilene-I Ipls/Clev and its four types of source-level trace replay.
179. Packet throughput marginals for Abilene-I Clev/Ipls and its four types of source-level trace replay.
180. Packet throughput marginals for Abilene-I Ipls/Clev and its four types of source-level trace replay.
181. Wavelet spectra of the byte throughput time series for Abilene-I Clev/Ipls and its four types of source-level trace replay.
182. Wavelet spectra of the byte throughput time series for Abilene-I Ipls/Clev and its four types of source-level trace replay.
183. Wavelet spectra of the packet throughput time series for Abilene-I Clev/Ipls and its four types of source-level trace replay.
184. Wavelet spectra of the packet throughput time series for Abilene-I Ipls/Clev and its four types of source-level trace replay.
185. Active connection time series for Abilene-I and its four types of source-level trace replay.
186. Bodies of the distributions of connection inter-arrivals for UNC 1 PM and 1 AM, and their exponential fits.
187. Tails of the distributions of connection inter-arrivals for UNC 1 PM and 1 AM, and their exponential fits.
188. Bodies of the distributions of connection inter-arrivals for Abilene-I and Leipzig-II, and their exponential fits.
189. Tails of the distributions of connection inter-arrivals for Abilene-I and Leipzig-II, and their exponential fits.
190. Average offered load vs. number of connections for 1,000 Poisson resamplings of UNC 1 PM.
191. Histogram of the average offered loads in 1,000 Poisson resamplings of UNC 1 PM.
192. Tails of the distributions of connection sizes for UNC 1 PM.
193. Analysis of the accuracy of connection-driven Poisson Resampling from 6,000 resamplings of UNC 1 PM (1,000 for each target offered load).
194. Comparison of average offered load vs. number of connections for 1,000 connection-driven Poisson resamplings and 1,000 byte-driven Poisson resamplings of UNC 1 PM.
195. Histogram of the average offered loads in 1,000 byte-driven Poisson resamplings of UNC 1 PM.
196. Analysis of the accuracy of byte-driven Poisson Resampling from 4,000 resamplings of UNC 1 PM (1,000 for each target offered load).
197. Analysis of the accuracy of byte-driven Poisson Resampling using source-level traces replay: replays of three separate resamplings of UNC 1 PM for each target offered load, illustrating the scaling down of load from the original 177.36 Mbps.
198. Analysis of the accuracy of byte-driven Poisson Resampling using testbed experiments: replay of one resampling of UNC 1 AM for each target offered load, illustrating the scaling up of load from the original 91.65 Mbps.
199. Connection arrival time series for UNC 1 PM (dashed line) and a Poisson arrival process with the same mean (solid line).
200. Connection arrival time series for UNC 1 AM and a Poisson arrivals process with the same mean.
201. Wavelet spectra of the connection arrival time series for UNC 1 PM and a Poisson arrival process with the same mean.
202. Wavelet spectra of the connection arrival time series for UNC 1 AM and a Poisson arrival process with the same mean.
203. Block resamplings of UNC 1 PM: impact of different block lengths on the wavelet spectrum of the connection arrival time series.
204. Block resamplings of UNC 1 AM: impact of different block lengths on the wavelet spectrum of the connection arrival time series.
205. Block resamplings of UNC 1 PM: average offered load vs. number of connection vectors (left) and corresponding histograms of average offered loads (right) in 3,000 resamplings.
206. Wavelet spectra of several random subsamplings of the connection vectors in UNC 1 PM (left) and 1 AM (right)
207. Analysis of the accuracy of byte-driven Block Resampling using source-level trace replay: replays of two separate resamplings of UNC 1 PM for each target offered load, illustrating the scaling down of load from the original 177.36 Mbps.
208. Analysis of the accuracy of byte-driven Block Resampling using source-level trace replay: replay of one resampling of UNC 1 AM for each target offered load, illustrating the scaling up of load from the original 91.65 Mbps.
209. Wavelet spectra of the packet arrival time series for UNC 1 PM and the source-level trace replays of two block resamplings of this trace.
210. Wavelet spectra of the packet arrival time series for UNC 1 PM and the source-level trace replays of three Poisson resamplings of this trace.

Abbreviations

ACKPositive acknowledgment TCP segment

ADUApplication Data Unit

APIApplication Programming Interface

AQMActive Queue Management

BGPBorder Gateway Protocol

BPFBerkeley Packet Filter

C.I.Confidence Interval

CCDFComplementary Cumulative Distribution Function

CDFCumulative Distribution Function

DAGData Acquisition and Generation

FIFOFirst-In First-Out

FINTCP control flag indicating ``no more data from sender''

FTPFile Transfer Protocol

GBGigabyte

GPSGlobal Positioning System

HTMLHyperText Markup Language

HTTPHyperText Transfer Protocol

I/OInput/Output

ICMPInternet Control Message Protocol

IPInternet Protocol

IRCInternet Relay Chat

ISPInternet Service Provider

K-SKolmogorov-Smirnov test

KBKilobyte

KppsKilo packet per second

LRDLong-Range Dependence

MBMegabyte

MIMEMultipurpose Internet Mail Extensions

MSSMaximum Segment Size

MTUMaximum Transmission Unit

MbpsMegabit per second

NNTPNetwork News Transfer Protocol

OSTTOne-Side Transit Time

PMAPassive Measurement and Analysis

Q-QQuantile-Quantile

REDRandom Early Detection

RFCRequest For Comments

RSTTCP control flag indicating ``connection reset''

RTTRound-Trip Time

SMTPSimple Mail Transfer Protocol

SSHSecure Shell

SYNSynchronize TCP control segment

SYN-ACKPositive acknowledgement of SYN segment

TCPTransport Control Protocol

UDPUser Datagram Protocol

UNCUniversity of North Carolina at Chapel Hill

URLUniversal Resource Locator