CRaft: An Erasure-coding-supported Version of Raft for Reducing Storage Cost
summary
CRaft replaces full-entry Raft replication with Reed-Solomon erasure coding (typical setting k=3, n=5 or 7) and builds a CRaft-backed key-value store. It claims 66% storage savings, up to 250% write throughput and 60.8% lower write latency versus vanilla Raft. Liveness holds with only F+1 healthy servers by degrades to full replication when fewer than F+K fragments exist, and a leader prepare phase collects or discards uncommitted fragments during leader change to preserve safety.
AI-written summary · about
chapters (14)
- 0:00Intro
- 0:35Data replication: when server crashes...
- 2:41Raft: distributed consensus
- 4:33Erasure Coding: Reed-Solomon Code
- 5:31Combine erasure coding with crash tolerance?
- 8:18CRaft: erasure coding + Raft
- 9:30Challenge #1: liveness
- 11:09Challenge #2: newly elected leader
- 15:08Comparing consensus protocols
- 15:32Evaluation highlights
- 16:35Evaluation setup
- 19:38Evaluation: read latency
- 20:23Summary: Craft
- 23:29Evaluation: liveness
related talks
41mThe Raft Consensus Algorithm
14mDistributed Consensus in 15 Minutes!
22mGraft: General Purpose Raft Consensus in Elixir
44mImproving Raft Survivability in Real-World Conditions
55mImplementing Raft in RabbitMQ
59mScaling Raft
47mDistributed Systems: When Democracy and Computer Science Come Together
47mUnderstanding Raft Consensus