Anonymous communications systems transfer information over a network while concealing the source and destination of this information. One of the most widely-used systems for this purpose is Tor, a volunteer-operated network that uses several thousand relays to support the communication of several million users each day. Because any anonymous communication system is more useful if it can serve more users, Tor is designed to provide low-latency anonymous connectivity to the general Internet.
Because Tor has such a high ratio of users to relays, it is important to find ways to improve the performance of Tor while not compromising the anonymity it provides. Our group has worked on several projects seeking methods to improve Tor's performance and to understand the degradation of anonymity resulting from these methods. This includes understanding attacks on and improvements to many mechanisms used by Tor to improve performance, including load balancing, admission control, congestion control, hidden services, circuit selection, and circuit scheduling.
Another important aspect of improving Tor's performance is accurately measuring and modeling the Tor network. Our group developed and maintains shadow, a network emulator that allows accurate large-scale simulations of Tor's performance and security, and developed algorithms for producing accurate reduced-scale models of the network. However, a further challenge remains, which is accurately and privately measuring the parameters of the Tor network that depend on user behavior, in the face of adversaries that might seek to distort these measurements. We are working on protocols that combine techniques from cryptography and differential privacy to address this problem.
An increasingly common Internet phenomenon is the use of technical means by corporate or state entities to prevent users from finding some content or sites on the Internet. Some users respond to this censorship (content blocking) by using circumvention technology that hides the content from a censor. Notable circumvention technologies, including Tor and many VPN-based services, are sometimes blocked in response. This leads to an "arms race" between censors and circumvention techniques, with several important questions.
One set of problems is in distribution: slowing censors that may act as insiders from finding contact information for circumvention systems. Our group has designed a system that maximize the lifetime of bridges, unlisted Tor relays used for circumvention. We also examined the privacy risks of running a bridge (or other peer-to-peer based relay scheme). We investigated social-network based protocols for building a "membership-concealing" network that prevents a censor from finding relays. We're still interested in how best to distribute and reliably test live relay information.
Another set of problems in censorship circumvention are about making it more difficult to identify and block censored or circumvention traffic. Our group has worked extensively on this problem; we have designed censorship resistant algorithms for peer-to-peer lookups; found new attacks on systems that hide circumvention traffic in other, legitimate traffic; and designed special purpose systems that deliver uncensored video and social network contents without requiring relay discovery or special-purpose client software. We are also investigating attacks and deployment strategies for "decoy routing" systems, that can use a large set of "overt" destinations to disguise circumvention traffic.
Traditional cryptographic schemes attempt to provide confidentiality and authentication, while not protecting the identities of the parties to a communication. Our group has worked on cryptographic schemes for several applications that support integrity and confidentiality while preserving the privacy of participants. Recent applications include:
We've also worked extensively in the past on the security and privacy of protocols to locate information in peer-to-peer networks, as well as the reliability of the Internet routing system.