MPC primer: Arcium's BDOZ and Cerberus protocol explained
Quote356 words13 likes
Walks through Arcium's MPC primer: secret sharing, Beaver triples, BDOZ MACs for malicious security, and Cerberus optimizations including multicast and identifiable cheating. Explains the offline/online phase split.
Key claims
Multicast reduces network scaling from O(n^2) to O(n), halving bandwidth vs BDOZ
Computing parties assumed to co-locate under 5ms latency
Cerberus adds identifiable abort via the BMRS24 compiler
MPC is a key technology to enable confidential computing at scale.
Yet, few understand what it actually is and why Arcium is the global leader in the vertical.
Let me give you a TL;DR on the primer by @arcium below. ↓
➤ MPC enables trustless information processing and allows distrusting parties to jointly compute on private data without revealing inputs, thereby solving the fundamental problem of cooperation without trust, moving from "should not be broken, or else xyz" contracts to cryptographic guarantees.
➤ Arcium's two-phase architecture separates heavy cryptography from computation: Offline phase generates preprocessing "fuel", doing all the heavy cryptographic lifting, while a lean online phase consumes this fuel for actual computations, dramatically improving UX and computation speed.
➤ Secrets split into random-looking shares where sum equals original value. The involved parties then compute locally on shares for addition, use Beaver triples for multiplication requiring minimal communication, with all operations forming arithmetic circuits.
➤ The BDOZ protocol provides malicious security through MACs (Message Authentication Codes), which prevent forgery as a sender cannot create valid MAC for incorrect data without receiver's key. Meanwhile the receiver doesn't know sender's share in advance, enabling verifiable computation where cheating is always detected.
➤ Arcium's Cerberus extends BDOZ with critical optimizations. It supports multiple prime fields with binary backend conversions, native elliptic curve arithmetic enabling direct cryptographic protocol execution (ElGamal, Bulletproofs), becoming coSNARKs without EC emulation overhead.
➤ Fiat-Shamir transform enables single MAC for multiple shares via random linear combinations, multicast reduces network scaling from O(n²) to O(n), asymptotically halving bandwidth vs vanilla BDOZ.
➤ Unlike BDOZ's secure-with-abort where parties can only exit without proof, Cerberus integrates BMRS24 compiler allowing cheating party identification through externally verifiable proofs.
➤ Arcium leverages Solana for decentralization without strict necessity of geographic distribution. Computing parties can co-locate (<5ms latency assumption) while maintaining trustlessness through network guarantees, solving MPC's traditional latency bottleneck.
➤ Arcium enables privacy-preserving computation in use cases ranging from AI or healthcare to R&D and military/geopolitics, shifting trust from backed by institutional trust and social/legal consequences to cryptographically enforced impossibility of misbehaviour.
More depth on this in the article linked here ↓
QUOTING @Arcium
An Introduction to Multiparty Computation We’re kicking off a technical series exploring how MPC reshapes trust and how Arcium uses it to make privacy programmable on blockchains. Enjoy the first article below. t.co/TA4JDJjV9P
An Introduction to Multiparty Computation We’re kicking off a technical series exploring how MPC reshapes trust and how Arcium uses it to make privacy programmable on blockchains. Enjoy the first article below. t.co/TA4JDJjV9P