Yong Wang of Northwestern University presented an interesting paper at the USENIX Symposium on Networked Systems Design and Implementation (NSDI) earlier this month. It explains how to geographically locate an IP address with a median error of about a half-mile square without any cooperation from the client on that IP address -- i.e. you or your computer.
This is a radical improvement over other "client independent" geolocation methods. On the dataset Wang and his co-authors used, they found that their method is 50 times more accurate than the best previous system.
Like most IP geolocation methods, the authors' method depends on network delay measurements; the time it takes for information to travel between different computers on a network. The authors apply such measurements in a sophisticated way, but their key insight is that the Web is full of landmarks. In their own words:
"[Many] entities host their Web services locally. Moreover, such Websites often provide the actual geographical location of the entity (e.g., business and university) in the form of a postal address. We demonstrate that the information provided in this way represents a precious resource, i.e., it provides access to a large number of highly accurate landmarks that we can exploit to achieve equally accurate geolocation results."
Essentially, the authors locate as many computers as they can on the Web using published street addresses, and then use network delay measurements to locate an IP address relative to those landmarks. Simple!
What are the applications of this? The authors give a nod to enforcement of "location-based access restrictions" and "context-aware security" and other serious-minded purposes, but the real application is online advertising.
"For example, knowing that a Web user is from New York is certainly useful, yet knowing the exact part of Manhattan where the user resides enables far more effective advertising, e.g., of neighboring businesses."
There is something seductive about such hyper-local advertising. I'd rather see today's specials for a nearby restaurant when reading The New York Times online than advertisements for shoes and groupons.
The authors' system would make that kind of advertising feasible. The restaurant would not be buying many ads; their specials would only be shown to local Times readers, within a block or two of the restaurant, or however they want to narrow the target audience. It's not hard to imagine such micro advertising becoming a big thing fast.
But, if we're concerned about how much Web advertisers know about us, how comfortable could we be about them knowing where we are? As the authors note, their method becomes more effective as population density goes up, so if you're living in an urban center, chances are the authors' method, well implemented, could do better than a half-mile square. Do you want advertisers -- or anyone -- to know what block you're on, or even what building you are in?