Paper 2022/171

Practical and Improved Byzantine Reliable Broadcast and Asynchronous Verifiable Information Dispersal from Hash Functions

Nicolas Alhaddad, Sisi Duan, Mayank Varia, and Haibin Zhang

Abstract

This paper improves upon two fundamental and closely related primitives in fault-tolerant distributed computing---Byzantine reliable broadcast (BRB) and asynchronous verifiable information dispersal (AVID). We make improvements asymptotically (for our AVID construction), concretely (much lower hidden constants), and practically (having 3 steps, using hash functions only, and avoiding using online error correction on the bulk data). The state of the art BRB protocol of Das, Xiang, and Ren (DXR BRB, CCS 2021) uses hash functions only and achieves a communication overhead of $O(nL + kn^2)$, where $n$, $L$, and $k$ are the number of replicas, the message length, and the security parameter, respectively. More precisely, DXR BRB incurs a concrete communication of $7nL + 2kn^2$, with a large constant 7 for the bulk data term (i.e., the $nL$ term). Das, Xiang, and Ren asked an open question if it is possible "from a practical point of view to make the hidden constants small." Two other limitations of DXR BRB that authors emphasized are that "higher computation costs due to encoding and decoding of the message" due to applying error correcting codes on bulk data and the fact that "in the presence of malicious nodes, each honest node may have to try decoding $f$ times" due to the use of an online error correcting algorithm. Meanwhile, the state of the art AVID protocols achieve $O(L+kn^2)$ communication assuming trusted setup. Apparently, there is a mismatch between BRB and AVID protocols: another natural open problem is whether it is possible to build a setup-free AVID protocol with $O(L+kn^2)$ communication. In this work, we answer all these open questions in the affirmative. We first provide a hash-based BRB protocol that improves concretely on DXR BRB, having low constants and avoiding using online error correction on bulk data. Our key insight is to encode the consistency proof, not just the message. Our technique allows disseminating the message and proof together. Then we provide the first setup-free AVID protocol achieving $O(L+kn^2)$ communication. Both our BRB and AVID protocols are practical because they have 3 steps, a multiplicative factor of 3 for the bulk data term, use hash functions only, and they avoid applying online error correction on bulk data.