Hydra: Serialization-Free Network Ordering for Strongly Consistent Distributed Applications
summary
Hydra implements a distributed network sequencer primitive that combines per-sequencer sequence counters with loosely synchronized physical timestamps to avoid a single sequencer bottleneck. The authors build Hydra-based state-machine replication and a transactional key-value store, measure in ns-3 and a 15-shard testbed, and report serialization-free network ordering, linear scaling with added sequencers, 5× faster sequencer failover, and 47% higher throughput than a single-sequencer system (Aries) while using flush messages to make progress when sequencers are idle.
AI-written summary · about
chapters (7)
- 0:00Intro
- 1:00Network Ordering to Eliminate Coordination
- 3:51Multi-Sequencer Challenge: Provide the same guarantees
- 4:51Solution: Combine sequence number with physical clock
- 8:42Deliver messages up to the minimum barrier
- 11:51Low Latency through Network Load Balancing
- 12:48Transactional Database Performance
related talks
15mHydra : Resilient and Highly Available Remote Memory
27mHydra: a federated resource manager for data-center scale analytics
01h 02mHydroflow: A Compiler Target for Fast, Correct Distributed Programs
22mDesigning Concurrent Distributed Sequence Numbers ...
13mSharding the State Machine: Automated Modular Reasoning for Complex Concurrent System
28mNetChain: Scale-Free Sub-RTT Coordination
27mEmpowering Resilient Scalability: The Power of Choreography in Distributed Systems
39mTrust Deterministic Execution to Scale & Simplify Your Systems