Tot 81,4 procent van het geanalyseerde verkeer zou te herleiden zijn tot een ip-adres met een foutmarge van 6,4 procent.
Hier mag hier nog iets duidelijker bij dat het labomstandigheden zijn. Vergelijk in het paper figuur 5 en 6b en de conclusie van het paper:
Our traffic analysis attack works well in a controlled in-lab set-up with symmetric network paths and path capacities(and other properties) and least congestion and uncontrolled disturbances. In such an enviromnent we were always able to determine the identity of the anonymous client amidst, several others clients that were communicating with the server. We observed sharp contrast between the correlation of the server to exit traffic with that of the entry node to victim client and the correlation of the server to exit node traffic with entry node to non-victim clients’ traffic. This enabled easy
detection of the victim traffic. This is evident from the results presented in the previous section.
However, on moving to tests with real Tor relays, we didn’t see such high correlation for server to exit traffic and entry node to client traffic. This is because of existing path congestion and the traffic scheduling by Tor relays which distort the injected traffic pattern. The decrease in correlation is attributed to this distortion in injected pattern, thereby leading information loss.
:strip_exif()/u/56159/archer.gif?f=community)
/u/207598/radioheadbear.png?f=community)
