Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BFTBrain: Adaptive BFT Consensus with Reinforcement Learning

Chenyuan Wu, Haoyun Qin, Mohammad Javad Amiri, Boon Thau Loo, Dahlia Malkhi, Ryan Marcus

TL;DR

BFTBrain tackles the problem of choosing the optimal Byzantine fault-tolerant protocol under dynamic workloads and fault conditions by employing a decentralized reinforcement-learning framework. It models protocol selection as a contextual multi-armed bandit, uses robust, consensus-based learning coordination to fuse local observations into a global decision, and switches protocols at epoch boundaries to maximize a user-defined performance metric. The approach demonstrates substantial throughput improvements over fixed protocols and prior learning-based methods, and proves robust against data pollution and hardware shifts. Its decentralized design, modest overhead, and plug-and-play compatibility across diverse hardware configurations make it highly practical for real-world SMR deployments. Overall, BFTBrain advances adaptive BFT by combining CMAB-based learning, robust coordination, and safe protocol switching with strong empirical gains.

Abstract

This paper presents BFTBrain, a reinforcement learning (RL) based Byzantine fault-tolerant (BFT) system that provides significant operational benefits: a plug-and-play system suitable for a broad set of hardware and network configurations, and adjusts effectively in real-time to changing fault scenarios and workloads. BFTBrain adapts to system conditions and application needs by switching between a set of BFT protocols in real-time. Two main advances contribute to BFTBrain's agility and performance. First, BFTBrain is based on a systematic, thorough modeling of metrics that correlate the performance of the studied BFT protocols with varying fault scenarios and workloads. These metrics are fed as features to BFTBrain's RL engine in order to choose the best-performing BFT protocols in real-time. Second, BFTBrain coordinates RL in a decentralized manner which is resilient to adversarial data pollution, where nodes share local metering values and reach the same learning output by consensus. As a result, in addition to providing significant operational benefits, BFTBrain improves throughput over fixed protocols by $18\%$ to $119\%$ under dynamic conditions and outperforms state-of-the-art learning based approaches by $44\%$ to $154\%$.

BFTBrain: Adaptive BFT Consensus with Reinforcement Learning

TL;DR

BFTBrain tackles the problem of choosing the optimal Byzantine fault-tolerant protocol under dynamic workloads and fault conditions by employing a decentralized reinforcement-learning framework. It models protocol selection as a contextual multi-armed bandit, uses robust, consensus-based learning coordination to fuse local observations into a global decision, and switches protocols at epoch boundaries to maximize a user-defined performance metric. The approach demonstrates substantial throughput improvements over fixed protocols and prior learning-based methods, and proves robust against data pollution and hardware shifts. Its decentralized design, modest overhead, and plug-and-play compatibility across diverse hardware configurations make it highly practical for real-world SMR deployments. Overall, BFTBrain advances adaptive BFT by combining CMAB-based learning, robust coordination, and safe protocol switching with strong empirical gains.

Abstract

This paper presents BFTBrain, a reinforcement learning (RL) based Byzantine fault-tolerant (BFT) system that provides significant operational benefits: a plug-and-play system suitable for a broad set of hardware and network configurations, and adjusts effectively in real-time to changing fault scenarios and workloads. BFTBrain adapts to system conditions and application needs by switching between a set of BFT protocols in real-time. Two main advances contribute to BFTBrain's agility and performance. First, BFTBrain is based on a systematic, thorough modeling of metrics that correlate the performance of the studied BFT protocols with varying fault scenarios and workloads. These metrics are fed as features to BFTBrain's RL engine in order to choose the best-performing BFT protocols in real-time. Second, BFTBrain coordinates RL in a decentralized manner which is resilient to adversarial data pollution, where nodes share local metering values and reach the same learning output by consensus. As a result, in addition to providing significant operational benefits, BFTBrain improves throughput over fixed protocols by to under dynamic conditions and outperforms state-of-the-art learning based approaches by to .
Paper Structure (32 sections, 15 figures, 3 tables, 1 algorithm)

This paper contains 32 sections, 15 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Landscape of BFT Performance
  3. Comparing Representative BFT Protocols
  4. The Case for Reinforcement Learning
  5. BFTBrain Overview
  6. System Model
  7. Design Overview
  8. Learning Algorithms
  9. Problem Formulation
  10. State and Action Space
  11. Predictive Model
  12. Learning Coordination
  13. Implementation
  14. Evaluation
  15. Experimental Setup
  16. ...and 17 more sections

Figures (15)

  • Figure 1: Overview of BFTBrain. For readability, we only present the internals of one node $i$.
  • Figure 2: Adaptivity of BFTBrain under changing conditions. The vertical dashed lines indicate when conditions change. Labels indicate the dominant protocol that BFTBrain and Adapt choose under each condition.
  • Figure 3: BFTBrain's throughput (a) during minutes $0$-$30$, and (b) during minutes $180$-$210$. In both periods, it encounters the system conditions captured in row $2$ of Table \ref{['tbl:landscape']}.
  • Figure 4: Robustness of BFTBrain against data pollution. The vertical dashed lines indicate when conditions change. Labels indicate the dominant protocol that Adapt (before and after pollution) chooses under each setting.
  • Figure 5: PBFT protocol
  • ...and 10 more figures

Theorems & Definitions (3)

  • proof
  • proof
  • proof