Picsou: Enabling Replicated State Machines to Communicate Efficiently
summary
Picsou introduces a formal primitive, Cross-Cluster Consistent Broadcast (C3B), and a practical protocol that uses TCP-like cumulative ACKs plus rotating quorum acknowledgments or Quacks so RSMs know when a correct receiver has definitely seen a message. Picsou claims constant metadata overhead in the failure-free case, minimal retransmits on failure, works for both crash and Byzantine RSMs, and shows up to 24× better throughput than prior approaches on microbenchmarks and applications like disaster recovery and data reconciliation.
AI-written summary · about
chapters (10)
- 0:00Picsou introduction
- 1:23RSM communication needs
- 3:08RSM communication design
- 4:54RSM failure handling
- 6:39Cross-cluster broadcast
- 8:03Reliable message protocol
- 10:28Quorum acknowledgements
- 12:59Quax retransmission
- 13:20Performance comparison
- 15:27Conclusion
related talks
16mSwiftPaxos: Fast Geo-Replicated State Machines
20mFine Grained Replicated State Machines for a Cluster Storage System
22mLiberating Distributed Consensus
20mExploiting Commutativity For Practical Fast Replication
48mDistributed Consensus Algorithms
12mConcordia: Distributed Shared Memory with In-Network Cache Coherence
57mReliability Nirvana
16mState-Compute Replication: Parallelizing High-Speed Stateful Packet Processing