# Shoal: Improving DAG-BFT Latency And Robustness

**Authors:** Alexander Spiegelman, Balaji Aurn, Rati Gelashvili, Zekun Li  
**arXiv:** [2306.03058](https://arxiv.org/abs/2306.03058)  
**Date:** 2023-06-05  
**Summary page:** https://www.summarizepaper.com/en/paper/2306.03058v1/

## Key points

- Shoal is a protocol-agnostic framework designed to enhance Narwhal-based consensus protocols
- Bullshark, built on Narwhal's DAG, can order over 100k transactions per second but has increased latency compared to leader-based BFT protocols
- Shoal reduces latency in Narwhal-based consensus by incorporating leader reputation and pipelining support
- Shoal introduces "Prevalent Responsiveness" which eliminates timeouts in most scenarios, surpassing traditional BFT protocols' Optimistic Responsiveness
- Experimental evaluations show up to a 40% reduction in latency during failure-free executions and up to an 80% reduction in executions with failures compared to vanilla Bullshark implementation

## Summary

In this paper, the authors introduce Shoal, a protocol-agnostic framework designed to enhance Narwhal-based consensus protocols. The Narwhal system is a state-of-the-art Byzantine fault-tolerant scalable architecture that utilizes a directed acyclic graph (DAG) of messages among validators in a Blockchain network. Bullshark, a zero-overhead consensus protocol built on top of Narwhal's DAG, can order over 100k transactions per second but suffers from increased latency due to DAG construction compared to leader-based BFT consensus protocols. Shoal incorporates leader reputation and pipelining support for the first time, significantly reducing latency in Narwhal-based consensus. By leveraging properties of the DAG construction and leader reputation mechanism, Shoal introduces "Prevalent Responsiveness," which eliminates timeouts in all but extremely rare scenarios. This property surpasses the desired Optimistic Responsiveness found in traditional BFT protocols. The authors integrated Shoal with Bullshark in an open-source Blockchain project and conducted experimental evaluations. Results show up to a 40% reduction in latency during failure-free executions and up to an 80% reduction in executions with failures compared to vanilla Bullshark implementation. The paper is structured as follows: Section 2 provides background information on DAG-BFT consensus and highlights key properties utilized in the study. Sections 3.1 and 3.2 introduce the pipelining approach and leader reputation solution in Shoal, respectively. Section 4 offers proof of correctness for the proposed framework, while Section 5 delves into implementation details and discusses timeouts. The results of the evaluation are presented in Section 6, followed by discussions on related work in Section 7. Finally, Section 8 concludes the paper with insights drawn from the study's findings on improving DAG-BFT latency and robustness through Shoal integration with Bullshark consensus protocol within Narwhal architecture.