Anonymous Networking under Memory Constraints

Abstract

Chaum Mixes, a class of proxy servers or relays which use randomized reordering and batching of packets from multiple users to provide source anonymity, have been used extensively for anonymous remailing, browsing and peer-to-peer file sharing. In this work, an analytical framework is proposed to measure and optimize the anonymity provided by mixing strategies when the mixes have memory restrictions. Specifically, an information-theoretic metric of anonymity is proposed for buffer constrained mixes, and using Poisson traffic models, fundamental trade-offs between achievable anonymity and the buffer size are studied analytically. In particular, a buffer-constrained mixing strategy is proposed that is asymptotically optimal and obtains the best convergence rate amongst known mixing strategies. The strategy is generalized to a network of mixes, where the achievable anonymity is expressed as a function of the topology and the buffer constraints of individual mixes.