SwiftPaxos: Fast Geo-Replicated State Machines
Talkat 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI '24) · 2024-04-16..04-18·
summary
Protocol executes a command in 2 message delays when there is no contention and in 3 delays otherwise, by letting replicas vote twice — once for a local ordering and then to follow the leader — avoiding restart of ballots and saving a half round trip in disagreement. Implementation runs on AWS, evaluated with noop service and YCSB, shows up to 2.9x throughput versus state-of-the-art and 16–29% lower latency than Paxos variants; code and results are on GitHub.
AI-written summary · about
chapters (7)
- 0:00Introduction to replication
- 2:00Swift Paxos approach
- 4:26Protocol mechanics
- 6:20Handling contention
- 9:43Optimizing client responses
- 11:06Performance evaluation
- 14:59Conclusion and summary
related talks
17mNear Optimal Latency Versus Cost Tradeoffs in Geo Distributed Storage
20mFine Grained Replicated State Machines for a Cluster Storage System
20mHigh Throughput Replication with Integrated Membership Management
12mEPaxos Revisited
23mLightning Talks on Distributed Systems
22mLiberating Distributed Consensus
40mImplementing Distributed Consensus
19mPineapple: Unifying Multi-Paxos and Atomic Shared Registers