Slashdot Mirror
Academics Build a New Tor Client Designed To Beat the NSA
An anonymous reader writes: In response to a slew of new research about network-level attacks against Tor, academics from the U.S. and Israel built a new Tor client called Astoria designed to beat adversaries like the NSA, GCHQ, or Chinese intelligence who can monitor a user's Tor traffic from entry to exit. Astoria differs most significantly from Tor's default client in how it selects the circuits that connect a user to the network and then to the outside Internet. The tool is an algorithm designed to more accurately predict attacks and then securely select relays that mitigate timing attack opportunities for top-tier adversaries.
no source code == no story
TOR was originally developed by the Navy to hide CIA and NSA traffic. It was released to the public specifically to allow everybody's lesser-importance traffic to provide cover for said spies.
Almost everyone involved in developing Tor was (or is) funded by the US government
The problem with link padding is that it would be very costly for Tor nodes and for usability.
Firstly, link padding would require rate-limiting each link to something quite small to keep bandwidth reasonable. If you think Tor is slow now, it would be much slower with padding.
Secondly, link padding also requires batching circuit construction. If a new link comes in, you can't immediately allow the Tor user to open a new link out. You have to wait and batch multiple outgoing link requests. That increases latency significantly to something much more than people already tolerate. Likewise, when a circuit is destroyed you can't immediately close all the links. You have to batch closure. In the meantime those links are just eating up bandwidth.
Thirdly, link padding _ideally_ requires propagating packet delays, similar to the the way you batch circuit constructions and closure. If the network did this, it would be trivial to DoS the Tor network because the network would amplify disruptions. But in practice I don't think this would ever be implemented.
Tor has succeeded mostly because of it's popularity. Even with link padding improving the security, you still need a large, active community using the network to maintain anonymity. Basically, as is typical you must rely on the pr0n and file-sharing subset to build the critical mass. Those folks are especially sensitive to bandwidth and latency.
Yes, link padding (which is the basis of Wei Dai's original pipenet proposal*, which itself predated the Navy's Onion Routing project) is the ideal. It's basically how e-mail mixers work. But it would also make the network as useable as e-mail mixers are, which is not very useable in the context of the web.
* http://www.weidai.com/pipenet.txt