1	Measuring ISP Response to VeriSign Site Finder Benjamin Edelman Berkman Center for Internet & Society Harvard Law School
2	Motivation Why measure generally Special implications of disabling Site Finder If find Site Finder disabled in many places If find Site Finder disabled in only a few places
3	Methodology I: Surveys / Form Submissions Problems Burden on respondents Representativeness Accuracy Results - Submissions to date report blocking by AOL Earthlink Time Warner Cable (partial?) San Bernardino County School System …
4	Methodology II: Site Finder Log Analysis Problem Availability of log files for analysis “[A]s a matter of privacy and operational security we do not provide our log files to any outside party” – Tom Galvin, VeriSign spokesman (email) Results Unknown
5	Methodology III: Inference from selected user clickstreams Idea: Get data about selected users’ / networks’ accesses to Site Finder content. Look for trends. Problems Obtaining necessary data Potential privacy concerns Users with non-default nameservers Users who request Site Finder content manually (other than by mistyping domain names) Works best for large networks
6	Reference: Inference from selected user clickstreams see also Technical Responses to Unilateral Internet Authority: The Deployment of VeriSign “Site Finder” and ISP Response http://cyber.law.harvard.edu/tlds/sitefinder with Jonathan Zittrain
7	Intro to Alexa Toolbar Provides search shortcut, related links, etc. 10+ million downloads. Active users unknown. Representativeness Users generally representative of Internet community. Possible under-emphasis on technical community. Possible over-emphasis on Southeast Asia.
8	Alexa Data Set Date through Sep. 29 as to requests for web pages on verisign.com by Alexa Toolbar users Data elements: User IP address (/24) Date & time URL requested
9	Some Noteworthy Networks: China
10	Some Noteworthy Networks: Adelphia
11	Aggregation: Idealizations
12	Aggregation: Data (ordinary, public Alexa rank chart)
13	Aggregation: Data (summed across all networks)
14	Additional Analysis with Add’l Data Possible data sources Site Finder web server log files Google Toolbar log files, other toolbars Could address Site Finder blocking by smaller networks by networks with fewer or no Alexa Toolbar users
15	Analyzing Page-Views on Site Finder Click on a “Did you mean?” link? Click on a “Popular Categories” link? Read the Terms of Use? Leave Site Finder without clicking on anything?
16	Analyzing Page-Views on Site Finder
17	Session-Based Analysis “What proportion of users click on…” rather than “What proportion of page-views are for…” Difficult using Alexa Toolbar data because low order byte of IP address is unavailable. 12% of user-sessions include at least one “Popular Category” view. (“Session” = “request from same /24 within 5 minutes”)
18	Benjamin Edelman