Original title: Barnacle: Adaptive Multi-Leader Scheduling for DAG-Based Consensus
Zeno De Angeli, Alexandru Ianov Vitanov, Philipp Jovanovic, Lefteris Kokoris-Kogias +3 · September 2026
In DAG-based consensus, computers called validators suggest transaction blocks at the same time, and designated leader blocks finalize them. Using multiple leaders per round cuts waiting time. However, real-world systems use only one leader. If one of several leaders is slow, it causes head-of-line blockingWhen one slow component in a pipeline holds up everything that follows it, forcing the whole system to wait.: the whole pipeline stalls for at least one timeout, or several waves if forced to use a slow fallback rule. More leaders make this risk worse. To fix this, researchers created Barnacle, an add-on that adjusts the leader count while the system runs. Barnacle checks the committed graph directly to see how many slots finalize through the normal fast rule, without sending extra messages or using cryptography. It raises the leader count gradually and drops it sharply when trouble appears. Barnacle achieves 6-13% lower latency than a single leader when healthy, and avoids stalls during network degradation.
You're at a grocery store with several checkout lanes open. More lanes mean shorter lines and faster checkout. But if one register jams, that lane's customers pile up and back into the aisles — so the manager hesitates to open more lanes, since any one of them could jam at any moment. In this paper, the blockchain watches how often its fast commit path succeeds and uses that signal to open more leader lanes when things are running smoothly or pull back to a single lane when something slows down.
Every blockchain faces a tradeoff: more leaders mean faster transaction confirmation, but one slow leader freezes the pipeline. Barnacle lets blockchain operators get the speed of many leaders while automatically pulling back to safety when the network degrades.
The abstract does not report results from a production deployment — integration with the Sui blockchain is noted as ongoing. The sensitivity of the AIMD parameters to different network conditions is not discussed, and the evaluation is at the protocol level rather than in a fully operational system.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Barnacle: Adaptive Multi-Leader Scheduling for DAG-Based Consensus — Zeno De Angeli, Alexandru Ianov Vitanov, Philipp Jovanovic et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.