'Lies, Damn Lies, and Statistics': A Critical Assessment of Preferential Attachment-type Network Models of the Internet

author:Walter Willinger, AT&T Labs, Inc. - Research
published: Nov. 22, 2007, recorded: October 2007, views: 244

Slides

Slides
0:00	A Critical Assessment of Preferential Attachment-type Network Models of the Internet
0:16	Acknowledgments
1:35	Recap: What “Network Science” says about the Internet - part 1
3:44	Recap: What “Network Science” says about the Internet - part 2
5:45	Basic Question
7:10	The Internet: The User Perspective
7:47	7The Internet: The Engineering Perspective
8:41	The Internet is a LAYERED Network
9:02	Internet Architecture: Vertical Decomposition
9:31	Internet Architecture: Horizontal Decomposition
10:03	Internet Architecture: Hourglass Design
11:11	Hourglass Design: Internet as Critical Infrastructure
11:41	Internet Connectivity/Topology - part 1
12:15	Internet Connectivity/Topology - part 2
12:28	The Many Facets of Internet Connectivity/Topology
14:12	Internet Connectivity/Topology
14:45	On Measuring Internet Connectivity
15:36	Back to our Basic Question
15:47	On Data Hygiene
15:49	On Measuring the Internet’s Router-level Topology - part 1
16:50	On Measuring the Internet’s Router-level Topology - part 2
17:09	On Measuring the Internet’s Router-level Topology - part 3
17:12	On Measuring the Internet’s Router-level Topology - part 4
17:17	On Measuring the Internet’s Router-level Topology - part 5
17:30	On Measuring the Internet’s Router-level Topology - part 6
17:37	HOWEVER: Problems with existing measurements
17:57	The Internet: The Engineering Perspective - part 1
18:16	HOWEVER: Problems with existing measurements
18:53	The Internet: The Engineering Perspective - part 1
19:10	The Internet: The Engineering Perspective - part 2
19:56	The Internet: The Engineering Perspective - part 3
20:38	HOWEVER: Problems with existing measurements
21:51	HOWEVER: Additional problems with existing measurements
22:25	On Statistical Rigor
22:34	How to lie with statistics …
24:08	Size-Frequency vs. Size-Rank PlotsorNon-cumulative vs. Cumulative - part 1
24:41	Size-Frequency vs. Size-Rank PlotsorNon-cumulative vs. Cumulative - part 2
25:40	On Model Validation
25:43	Taking Model validation more serious …
26:00	Cisco 12000 Series Routers
26:19	Router Technology Constraint
27:09	Back to the Basic Question
27:21	Network Science and the Internet: “Lies, damned lies and statistics”
28:26	The “Math” Perspective of the Internet
29:32	How to avoid such Fallacies?
30:25	Internet Modeling: An Engineering Perspective
33:05	Heuristically Optimized Topologies (HOT) - part 1
34:26	Heuristically Optimized Topologies (HOT) - part 2
34:59	HOT-type Network Models
35:27	Some Implications of this Engineering Perspective
36:01	Further Implications of this Engineering Perspective - part 1
36:42	Further Implications of this Engineering Perspective - part 2
38:57	Further Implications of this Engineering Perspective - part 3
41:15	A Reminder
42:43	http://hot.caltech.edu/topology.html

Report a problem or upload files

If you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.

Lecture popularity: You need to login to cast your vote.

Description

Basic Question: Do the available Internet-related connectivity measurements and their analysis support the sort of claims that can be found in the existing complex networks literature? Key Issues: What about data hygiene? What about statistical rigor? What about model validation? Author discusses some of the main problems and challenges associated with measuring, inferring, and modeling various types of Internet-related connectivity structures. To this end, he uses some known examples to illustrate the need to understand the process by which Internet connectivity measurements are obtained, explore the sensitivity of inferred graph properties to known ambiguities in the data, be more critical with respect to the dominant, preferential attachmenttype network modeling paradigm, and be more serious/ambitious when it comes to model validation. Ignoring any of these issues is bound to produce results that are best described by the well-known aphorism "lies, damned lies, and statistics."