Karaoke: Fast And Strong Metadata Privacy With Low Noise

Authors:
David Lazar MIT CSAIL
Yossi Gilad MIT CSAIL
Nickolai Zeldovich MIT CSAIL

Introduction:

Karaoke is a system for low-latency metadata-private communication.

Abstract:

Karaoke is a system for low-latency metadata-private communication. Karaoke provides differential privacy guarantees, and scales better with the number of users than prior such systems (Vuvuzela and Stadium). Karaoke achieves high performance by addressing two challenges faced by prior systems. The first is that differential privacy requires continuously adding noise messages, which leads to high overheads. Karaoke avoids this using optimistic indistinguishability: in the common case, Karaoke reveals no information to the adversary, and Karaoke clients can detect precisely when information may be revealed (thus requiring less noise). The second challenge lies in generating sufficient noise in a distributed system where some nodes may be malicious. Prior work either required each server to generate enough noise on its own, or used expensive verifiable shuffles to prevent any message loss. Karaoke achieves high performance using efficient noise verification, generating noise across many servers and using Bloom filters to efficiently check if any noise messages have been discarded. These techniques allow our prototype of Karaoke to achieve a latency of 6.8 seconds for 2M users. Overall, Karaoke's latency is 5x to 10x better than Vuvuzela and Stadium.

You may want to know: