Steelhead: Interleaving Partially Synchronous and Asynchronous Commit Rules on a Shared DAG
This study addresses the inherent tension in consensus protocols, which are efficient under partial synchrony yet prone to stalling in asynchronous environments. To resolve this, we propose a hybrid consensus algorithm built upon a shared directed acyclic graph (DAG). The approach alternates between partially synchronous and asynchronous commit rules, dynamically adjusting execution periods to balance performance with liveness. It enables adaptive mode switching without additional communication overhead and incorporates a hidden leader mechanism to guarantee asynchronous liveness. The protocol is formally verified using Lean 4. Experimental results demonstrate that the proposed method achieves throughput comparable to synchronous protocols under favorable network conditions while preserving liveness approaching that of asynchronous protocols in adverse environments, thereby effectively reconciling efficiency with robustness.