Fast, Metadata-private Anonymous Broadcast

Abstract

This thesis presents Trellis: a mix-net based anonymous broadcast system with cryptographic security guarantees. Trellis can be used to anonymously publish documents or communicate with other users, all while assuming full network surveillance. In Trellis, users send messages through a set of servers in successive rounds. The servers mix and post the messages to a public bulletin board, hiding which senders sent which messages.

Trellis hides all network-level metadata, remains robust to changing network conditions, guarantees availability to honest users, and scales with the number of mix servers. Trellis provides three to five orders of magnitude faster performance and better network robustness compared to Atom, the state-of-the-art anonymous broadcast system with a similar threat model.

In achieving these guarantees, Trellis contributes: (1) a simpler theoretical mixing analysis for a routing mix network constructed with a fraction of malicious servers, (2) anonymous routing tokens for verifiable random paths, and (3) lightweight blame protocols built on top of onion routing to identify and eliminate malicious parties.

We implement and evaluate Trellis in a networked deployment. With 32 servers located across four geographic regions, Trellis achieves a throughput of 200 bits per second with 100,000 users. With 64 servers, Trellis achieves a throughput of 320 bits per second. Trellis’s throughput is only 100 to 1000× slower compared to Tor (which has 2M daily users) and is therefore potentially deployable at a smaller “enterprise” scale. Our implementation is open-source.