| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 1.5 MB | Adobe PDF |
Abstract(s)
Multi-Party Computation (MPC) has recently gained interest as a tool to perform secure, distributed computations. However, current work presents limitations that hinder their practical application, namely not supporting a framework suited for long computations, assuming a fixed participant size, not tolerating faults, and running in strictly synchronous environments. We present
MPCServe, a new practical Multi-Party Computation framework that dynamically performs computations while adopting a MPC-as-a-Service model for ease of usage. Our framework allows
lightweight clients to outsource their privacy-preserving computations on encrypted data to a set
of untrusted servers while guaranteeing computational output in the presence of t Byzantine faults
assuming a total of at least n > 3t servers. MPCServe extends COBRA, a confidential Byzantine
Fault Tolerance State Machine Replication framework that uses Dynamic Proactive Secret Sharing
(DPSS) for storing data with high levels of privacy, integrity, and availability. Leveraging its faulttolerance guarantees and the homomorphic properties of DPSS, MPCServe builds a fluid-style,
maliciously secure MPC infrastructure for asynchronous networks that allows servers to join and
leave during the computational effort.
Description
Tese de Mestrado, Engenharia Informática, 2025, Universidade de Lisboa, Faculdade de Ciências
Keywords
Computação Segura entre Pares Partilha de Segredos Tolerância a Faltas Bizantinas Teses de mestrado - 2025