<The Network is aware>

using network data to detect dpi


Network transparency cuts both ways. It can be exploited to engage in surveillance of network operators by users as well as the other way around. In order to better understand DPI use and the scope of its deployment, the project makes use of crowdsourced data from a network monitoring test known as Glasnost. An Internet user who runs the Glasnost test can see whether BitTorrent is completely blocked by their ISP, slowed down (throttled), or running normally. Glasnost was developed by researchers at a German institute to detect blocking or throttling of BitTorrent and other peer to peer (P2P) file sharing protocols (Dischinger, Marcon, et al, 2010). Papers about their technical work can be accessed here. Our research team has written its own paper on the challenges we faced using the Glasnost data, which can be accessed here (Ashgari, Mueller, van Eeten and Wang, 2012)

Thanks to an initiative known as the Measurement Lab (MLab), supported by Google, the New America Foundation and the PlanetLab Consortium, the Glasnost test allows end users all over the world to test whether their ISP is blocking or throttling BitTorrent and other protocols. The results are stored and made available to researchers. The Glasnost data begins in April 2008 and continues to the present time. In 2009 there was a change in the test protocol to make it more reliable. Most of the data we display here are the result of tests run from January 1 2009 to the end of February 2012. Our processing of the MLab data builds on the research of Michel van Eeten, Johannes Bauer, Hadi Asghari, & Shirin Tabatabaie, (2010).

We display here three kinds of data: an Interactive table that allows you to select any ISP in the world with a sufficient quantity of tests to see whether it uses DPI to throttle or block BitTorrent, a list of the world’s “top throttlers” for each of the years 2009, 2010, 2011 and 2012, and charts for some selected countries, including Germany and Canada. 

The world’s top throttlers

Interactive table: BitTorrent manipulation by the world’s ISPs from 2009 to Q1 2012

Selected country charts

   (Canada, U.S.A., Taiwan, Hong Kong, Poland, Germany, Brazil, Great Britain)

BitTorrent Throttling - worldwide aggregate

Glasnost data, Q1 2009 - Q1 2012

This chart provides a very crude measurement of the overall extent to which ISPs worldwide use DPI to detect and manipulate a specific application (in this case, BitTorrent). It simply aggregates all the Glasnost tests quarterly and shows the percentage of valid tests for which the verdict is DPI use.

The data indicates that DPI use to detect and manipulate a specific application seems to be declining. There is a peak in Q2 2009 at 24%, and for all of 2009 it stays over 20%. In 2010 and the first half of 2011, it stays between 17% and 20%, and then for the rest of the time period it (Q3 2011 to Q1 2012) it declines to 15% - 17%.

The measurement is not definitive because it aggregates the results we have from user-initiated tests across all the world’s ISPs. The problem with that method is that it does not guarantee a representative sample of the entire user population proportionally distributed across each of the world’s ISPs. Thus, places such as Russia, Turkey or Indonesia, where many users may not be able to read the MeasurementLab test page or have high chance of being exposed to it, or places like China which not only has a different language/script but also may be blocked from taking the test, will be underrepresented in these statistics. Large North American ISPs and English-speaking countries such as GB and AU, on the other hand, would be overrepresented.