We need people to attack the system, quantify defenses, etc. Here are some example projects:
- What algorithm should we use to assign Guard flags such that a) we assign the flag to as many relays as possible, yet b) we minimize the chance that Alice will use an attacker's node as a guard? See the blog post for details.
- For various diversity metrics, how has the diversity of the Tor network changed over time? How robust is it to change or attack? These results can help us make better design decisions. See the blog post for details.
- If we prevent the really loud users from using too much of the Tor network, how much can it help? We've instrumented Tor's entry relays so they can rate-limit connections from users, and we've instrumented the directory authorities so they can change the rate-limiting parameters globally across the network. Which parameter values improve performance for the Tor network as a whole? How should relays adapt their rate-limiting parameters based on their capacity and based on the network load they see, and what rate-limiting algorithms will work best? See the blog post for details.
- Right now Tor clients are willing to reuse a given circuit for ten minutes after it's first used. The goal is to avoid loading down the network with too many circuit creations, yet to also avoid having clients use the same circuit for so long that the exit node can build a useful pseudonymous profile of them. Alas, ten minutes is probably way too long, especially if connections from multiple protocols (e.g. IM and web browsing) are put on the same circuit. If we keep fixed the overall number of circuit extends that the network needs to do, are there more efficient and/or safer ways for clients to allocate streams to circuits, or for clients to build preemptive circuits? Perhaps this research item needs to start with gathering some traces of what requests typical clients try to launch, so you have something realistic to try to optimize.
- The "website fingerprinting attack": make a list of a few hundred popular websites, download their pages, and make a set of "signatures" for each site. Then observe a Tor client's traffic. As you watch him receive data, you quickly approach a guess about which (if any) of those sites he is visiting. First, how effective is this attack on the deployed Tor design? The problem with all the previous attack papers is that they look at timing and counting of IP packets on the wire. But OpenSSL's TLS records, plus Tor's use of TCP pushback to do rate limiting, means that tracing by IP packets produces very poor results. The right approach is to realize that Tor uses OpenSSL, look inside the TLS record at the TLS headers, and figure out how many 512-byte cells are being sent or received. Then start exploring defenses: for example, we could change Tor's cell size from 512 bytes to 1024 bytes, we could employ padding techniques like defensive dropping, or we could add traffic delays. How much of an impact do these have, and how much usability impact (using some suitable metric) is there from a successful defense in each case?
- More coming soon. See also the "Research" section of the volunteer page for other topics.