doubleyewdee, even if Alexa doesn't monitor or record DNS lookup failures, I don't think this presents a problem given our method of analysis.
In general, we look for a drop-off in Site Finder page views. So if Site Finder page views were high from a given ISP, then dropped off dramatically and suddenly, we notice this and classify the ISP as blocking Site Finder as of the corresponding date. It doesn't matter whether Alexa's other log data shows the dns-lookup-failure'd domains as msn logs, as dns lookup failures, as something else, or as nothing at all (so long as they don't show them as Site Finder, which they definitely would not) -- we'd still see the distinctive drop-off in Site Finder traffic.
The result you describe could come from one of at least two separate possibilities:
1) As another comment mentioned, some popups are apparently delayed relative to the event that triggers the popup. A specific example may be helpful. Gator could Fedex the following service: "Five clicks after a user exits ups.com, show the Fedex popup ad." Or, alternatively, "Five minutes..." I believe I've seen signs of both of these methods. For example, if I go to an online travel site, then go to harvard.edu (for which most Gator popup ads are about education), I might well be shown a Gator popup about travel.
2) Gator does target some ads at all sites (or almost all sites). The report describes this as follows:
Testing indicates that, in addition to advertisements targeted at specific web sites, Gator also shows certain additional advertisements subsequent to user requests for numerous web sites. (Gator seems to prevent such advertisements from being displayed subsequent to user requests for certain sites, including Gator's own site and sites specified in court orders). As of May 2003, an additional 17 advertisements were placed in this category of general advertisements, including advertisements about replacement ink cartridges, cell phones, lotteries, business cards, and DVD rental. Using the testing system below, these ads can be viewed by specifying the string
notrgs as the domain name to check.
Sorry to take so long to respond -- I've been offline all day, travelling, and haven't had connectivity.
This example is real, not fabricated! I saw it in Gator's configuration files, which I've preserved for posterity, and also confirmed it with a hands-on test.
Right. Multiple tests using our testing system ( http://cyber.law.harvard.edu/filtering/china/test ) have indicated that Altavista.com is inaccessible from China.
2) There's something else on the first IP that China doesn't like. You might be able to find it (try a google search on "china site:www.caltech.edu"). You might not.
Remember, all indications to date (to my knowledge, at least) are the filtering operates at the granularity of an entire IP address. If so, there's no way to block just one page or directory or even domain name (if multiple domain names are hosted on the same IP, as is of course standard these days). So China might "have" to block www.caltech.edu notwithstanding that the page they "want" to block is www.caltech.edu/deep/link/path/filename.html .
Any thoughts on what technology or technologies China might be using to block access to specific web pages? My testing to date suggests that blocking is generally at the level of an entire IP -- affecting all web pages (and all virtual servers) hosted from that IP. This sort of blocking is straightforward -- just load a "blackhole" file into the router, and that's that. On the other hand, blocking specific pages is harder -- usually the kind of task that requires a proxy, with all the cost, complexit, and performance implications that entails. Is it your sense that that's what China is doing? Or something else? Any references available?
Many thanks to all who suggested sites. Logs reflect a substantial number of newly-found blocked sites -- many of them going beyond the sites Professor Zittrain and I have been testing to date. (We've obtained sites tested to date using extraction from Yahoo and from Google.)
There have been several messages in this forum about getting multiple results re certain sites of interest, slashdot and google in particular. It's certainly odd to see a given site listed as both "reported to be accessible" and "reported to be inaccessible" mere minutes apart. But as it turns out that's not an unusual result in our testing of China -- perhaps reflecting network congestion, perhaps round-robin DNS (using multiple IPs for a given host, and only one or several but not all IPs are blocked), perhaps some other factor yet to be determined. Accordingly, in our "final reporting" of blocked URLs (as distinguished from the preliminary results posted in real-time by the testing system), we'll report as "inaccessible" only those sites that pass a reasonably rigorous test ("inaccessible on at least 60% of tests, and inaccessible from at least three testing locations"). Meanwhile, http://code.law.harvard.edu/filtering/list.html should be taken only for what it is; I've added a link to the FAQ to the top of that page to encourage folks to read more on this subject. Most importantly, look for our forthcoming report -- expected later this fall -- which will detail the sites we in fact found to be inaccessible, consistently, over an extended period of time.
Finally, a few readers asked whether the "test another url" form is working properly. I can confirm that it is. However, the testing site uses frames, so the URL listed in the Address Bar can in some instances get out of sync with what's really being tested. With any luck I'll fix this later this afternoon. My apologies for the confusion.
I worked on this case as an expert for the library plaintiffs and for the ACLU, and I thought folks here might be interested in my expert report.
The brunt of my research was documenting some 6000+ examples of overblocking -- blocking of sites that weren't consistent with respective filtering programs' blocking definitions and policies.
1) I agree that some portions of content on some of the sites on my list have been correctly categorized. But in the instance you described, it sounds like the specific URL on my list doesn't contain content meeting filtering programs' category definitions. As a result, even if there's reason to categorize other content on that same server, there's no need to categorize this specific page.
(To put this a different way: Many of the filtering programs seem to classify entire sites -- all content on an entire domain name, for example. But there's no reason why pages couldn't instead be rated on a page-by-page basis [and indeed some filtering companies report that they do this, too, in at least some instances]. To the extent that programs fail to do review and separately categorize every individual page, they may overblock pages without content meeting their criteria.)
2) There's no doubt that some URLs on my lists actually do meet filtering companies' category definitions. I'm no librarian, and neither am I otherwise trained in content categorization, so it wasn't my job to identify this content. (Plus, as you can imagine, it's a large task to view many thousands of sites!) Instead, librarians reviewed certain of the sites (including a random sample of the entire list) to attempt to estimate the proportion of sites from my lists that are, in their professional opinions, suitable for use within a library. It's my understanding that the results of their study are forthcoming.
Troller_Park_Trash, If you're already knowledgeable about the means of operation of filtering software, you may find that the most new & interesting part of the http://cyber.law.harvard.edu/people/edelman/mul-v- us/ page is the Appendices listing specific sites that have been, by and large, wrongly classified by filtering programs.
Regarding the blacking out of certain text from my report: As http://cyber.law.harvard.edu/people/edelman/mul-v- us/ mentions, a protective order (from the court in which the underlying case is pending) limits distribution of certain portions of my report -- namely anything I learned from reviewing confidential documents from filtering companies, or from attending confidential portions of depositions of their employees. But the work you, and most others here, are likely to find of greatest interest is the listings of specific sites blocked. (I'm presently adding a bit of text and formatting to help folks find this content more quickly and easily.)
Slashdot Mirror
In general, we look for a drop-off in Site Finder page views. So if Site Finder page views were high from a given ISP, then dropped off dramatically and suddenly, we notice this and classify the ISP as blocking Site Finder as of the corresponding date. It doesn't matter whether Alexa's other log data shows the dns-lookup-failure'd domains as msn logs, as dns lookup failures, as something else, or as nothing at all (so long as they don't show them as Site Finder, which they definitely would not) -- we'd still see the distinctive drop-off in Site Finder traffic.
Ben Edelman
Berkman Center for Internet & Society
Harvard Law School
1) As another comment mentioned, some popups are apparently delayed relative to the event that triggers the popup. A specific example may be helpful. Gator could Fedex the following service: "Five clicks after a user exits ups.com, show the Fedex popup ad." Or, alternatively, "Five minutes..." I believe I've seen signs of both of these methods. For example, if I go to an online travel site, then go to harvard.edu (for which most Gator popup ads are about education), I might well be shown a Gator popup about travel.
2) Gator does target some ads at all sites (or almost all sites). The report describes this as follows:
See the ads in this "show everywhere" category.
Ben Edelman
Berkman Center for Internet & Society
Harvard Law School
This example is real, not fabricated! I saw it in Gator's configuration files, which I've preserved for posterity, and also confirmed it with a hands-on test.
Ben Edelman
Berkman Center for Internet & Society
Harvard Law School
Right. Multiple tests using our testing system ( http://cyber.law.harvard.edu/filtering/china/test ) have indicated that Altavista.com is inaccessible from China.
1) They're on different IPs.
and,
2) There's something else on the first IP that China doesn't like. You might be able to find it (try a google search on "china site:www.caltech.edu"). You might not.
Remember, all indications to date (to my knowledge, at least) are the filtering operates at the granularity of an entire IP address. If so, there's no way to block just one page or directory or even domain name (if multiple domain names are hosted on the same IP, as is of course standard these days). So China might "have" to block www.caltech.edu notwithstanding that the page they "want" to block is www.caltech.edu/deep/link/path/filename.html .
Any thoughts on what technology or technologies China might be using to block access to specific web pages? My testing to date suggests that blocking is generally at the level of an entire IP -- affecting all web pages (and all virtual servers) hosted from that IP. This sort of blocking is straightforward -- just load a "blackhole" file into the router, and that's that. On the other hand, blocking specific pages is harder -- usually the kind of task that requires a proxy, with all the cost, complexit, and performance implications that entails. Is it your sense that that's what China is doing? Or something else? Any references available?
Many thanks to all who suggested sites. Logs reflect a substantial number of newly-found blocked sites -- many of them going beyond the sites Professor Zittrain and I have been testing to date. (We've obtained sites tested to date using extraction from Yahoo and from Google.)
There have been several messages in this forum about getting multiple results re certain sites of interest, slashdot and google in particular. It's certainly odd to see a given site listed as both "reported to be accessible" and "reported to be inaccessible" mere minutes apart. But as it turns out that's not an unusual result in our testing of China -- perhaps reflecting network congestion, perhaps round-robin DNS (using multiple IPs for a given host, and only one or several but not all IPs are blocked), perhaps some other factor yet to be determined. Accordingly, in our "final reporting" of blocked URLs (as distinguished from the preliminary results posted in real-time by the testing system), we'll report as "inaccessible" only those sites that pass a reasonably rigorous test ("inaccessible on at least 60% of tests, and inaccessible from at least three testing locations"). Meanwhile, http://code.law.harvard.edu/filtering/list.html should be taken only for what it is; I've added a link to the FAQ to the top of that page to encourage folks to read more on this subject. Most importantly, look for our forthcoming report -- expected later this fall -- which will detail the sites we in fact found to be inaccessible, consistently, over an extended period of time.
Finally, a few readers asked whether the "test another url" form is working properly. I can confirm that it is. However, the testing site uses frames, so the URL listed in the Address Bar can in some instances get out of sync with what's really being tested. With any luck I'll fix this later this afternoon. My apologies for the confusion.
The brunt of my research was documenting some 6000+ examples of overblocking -- blocking of sites that weren't consistent with respective filtering programs' blocking definitions and policies.
My report is available at http://cyber.law.harvard.e du/people/edelman/mul-v-us
Ben Edelman
Berkman Center for Internet & Society
Harvard Law School
Good point. I've tweaked some of the language to make this more clear. Thanks for the suggestion.
Sh00z, two thoughts:
1) I agree that some portions of content on some of the sites on my list have been correctly categorized. But in the instance you described, it sounds like the specific URL on my list doesn't contain content meeting filtering programs' category definitions. As a result, even if there's reason to categorize other content on that same server, there's no need to categorize this specific page.
(To put this a different way: Many of the filtering programs seem to classify entire sites -- all content on an entire domain name, for example. But there's no reason why pages couldn't instead be rated on a page-by-page basis [and indeed some filtering companies report that they do this, too, in at least some instances]. To the extent that programs fail to do review and separately categorize every individual page, they may overblock pages without content meeting their criteria.)
2) There's no doubt that some URLs on my lists actually do meet filtering companies' category definitions. I'm no librarian, and neither am I otherwise trained in content categorization, so it wasn't my job to identify this content. (Plus, as you can imagine, it's a large task to view many thousands of sites!) Instead, librarians reviewed certain of the sites (including a random sample of the entire list) to attempt to estimate the proportion of sites from my lists that are, in their professional opinions, suitable for use within a library. It's my understanding that the results of their study are forthcoming.
Troller_Park_Trash, If you're already knowledgeable about the means of operation of filtering software, you may find that the most new & interesting part of the http://cyber.law.harvard.edu/people/edelman/mul-v- us/ page is the Appendices listing specific sites that have been, by and large, wrongly classified by filtering programs.
- us/index-subset.html ("Blocked Site Archives - Subset with Linked Pages - Appendix A") gives information about 395 such URLs. You'll likely find yourself surprised that many of these are blocked -- I know I was.
- us/ mentions, a protective order (from the court in which the underlying case is pending) limits distribution of certain portions of my report -- namely anything I learned from reviewing confidential documents from filtering companies, or from attending confidential portions of depositions of their employees. But the work you, and most others here, are likely to find of greatest interest is the listings of specific sites blocked. (I'm presently adding a bit of text and formatting to help folks find this content more quickly and easily.)
For example, http://cyber.law.harvard.edu/people/edelman/mul-v
Regarding the blacking out of certain text from my report: As http://cyber.law.harvard.edu/people/edelman/mul-v
Ben Edelman