Keyun Cheng

Reading Notes: paper

Title: Practical Byzantine fault tolerance

Conference (OSDI’1999): Link

Journal (): Link

Summary

This paper describes the PBFT consensus algorithm that solves the BFT problem with practicality consideration (efficient and simple in implementation).

Main Contributions

Details

It’s a replication-based consensus algorithm that replicates data with 3f + 1 copies, and tolerates at most f faulty replicas.

Algorithm sketch:

• Client request
  • Send to Primary node
  • If the Primary node has no response, then broadcast to all backup nodes
• Pre-prepare phase (by primary node)
  • The Primary node receives request from Client and broadcast pre-prepare message to all backup nodes for verification. The message has a time order
  • Purpose: In order to tolerate faulty primary node
• Prepare (by all backup nodes)
  • Backup node checks and validates the pre-prepare message from the primary node. It can verify that the Primary node is valid
  • Backup node also validates the message and broadcast a prepare message to Primary node and all other Backup nodes
• Commit
  • When all Backup nodes all agree on f + 1 prepare messages, it replies to the Client with the result (true or false)
• Result
  • Client must receive f+1 SAME replies to validate the results (whether true or false). It make sure the majority of nodes agree on the same results

Strength

Weakness