HyperLedger Fabric --共识

What machine faults?
• Crash faults (CFT): A machine simply stops execution and halts容错故障节点，不能容错恶意节点

Paxos, RAFT, Zookeeper AB,…

•  Non-crash (a.k.a. ByzanUne) faults (BFT)

CFT和BFT是用于解决“因为节点接到指令的顺序不一样而产生的结果不一样”这个问题的共识算法

BFT不能解决：1.非确定性值引起的差异 2.特别慢的节点会影响算法效率 3.增加新节点很困难

Many other things burden the implementation (it is not simple as it might look)
•  Non-deterministic value 不确定性值
•  Leader election 选举Leader节点
•  State transfer (new, slow Party) 状态转换
•  Reconfiguration 重新配置

Ordering Services

Different configuration options for the ordering service
include:
–  SOLO
• Single node for development
–  Kafka / Zookeeper
• 1:n nodes providing Crash Fault Tolerance（CFT,碰撞容错）
• Odd number of nodes recommended（奇数节点）
–  SBFT
• 1:n nodes providing Byzantine Fault Tolerance