Byzantine Fault Tolerant Protocols with Near-Constant Work per Node without Signatures
Philipp Schneider
TL;DR
The paper addresses Byzantine fault-tolerant tasks for large-scale networks without cryptographic signatures by introducing a pre-computation phase that builds a system of witness committees. This precomputation enables execution with near-constant per-node work ($\widetilde{O}(1)$) and supports a suite of tasks—reliable broadcast, reliable aggregation, consensus, and common coins—through a modular protocol stack that operates asynchronously after the committees are in place. The core contributions are a Monte Carlo construction of $\Theta(n)$ near-constant-size committees with provable availability and agreement properties across phases, and a deterministic execution framework that leverages a broadcast tree to achieve efficient reliable broadcast, aggregation, and consensus without signatures. The approach promises scalability and modularity, with potential applications to sharding and large distributed data structures, while shifting the adversarial and timing assumptions primarily to the pre-computation phase.
Abstract
Numerous distributed tasks have to be handled in a setting where a fraction of nodes behaves Byzantine, that is, deviates arbitrarily from the intended protocol. Resilient, deterministic protocols rely on the detection of majorities to avoid inconsistencies if there is a Byzantine minority, which requires individual nodes to handle a communication workload that is proportional to the size of the network -- an intolerable disadvantage in large networks. Randomized protocols circumvent this by probing only small parts of the network, thus allowing for consistent decisions quickly and with a high level of confidence with communication that is near-constant in the network size. However, such protocols usually come with the drawback of limiting the fault tolerance of the protocol, for instance, by severely restricting the number or type of failures that the protocol can tolerate. We present randomized protocols to reliably aggregate and broadcast information, form consensus and compute common coins that tolerate a constant fraction of Byzantine failures, do not require cryptographic signatures and have a near-constant time and message complexity per node. Our main technique is to compute a system of witness committees as a pre-computation step almost optimally. This pre-computation step allows to solve the aforementioned distributed tasks repeatedly and efficiently, but may have far reaching further applications, e.g., for sharding of distributed data structures.