Determinedly deterministic and how it can help you
summary
Claim: making core state machines fully deterministic yields reproducible outputs, fast recovery and microsecond-scale performance. Implementation uses LMAX Disruptor ring buffers to sequence inputs, on-disk journaling for replay, active replica replication plus binary snapshots for fast restore, and a single-writer (one-thread) model to avoid locks. Numbers cited: 100k/sec sustained, 800k/sec peak, ~80 µs latency targets. Proven benefits: exact replay for debugging, simple failover, and significantly reduced jitter when pinned to isolated CPU cores.
AI-written summary · about
chapters (21)
- 0:00Introduction and speakers
- 4:06Why deterministic systems
- 8:23Non-deterministic system example
- 11:13Key features of deterministic systems
- 13:05Disruptor ring buffer explained
- 15:54Journaling for replayability
- 17:21Replication and fault tolerance
- 19:09Snapshotting for recovery
- 21:38Common pitfalls in determinism
- 24:45Performance benefits of single-threaded updates
- 28:20Optimizing the single writer thread
- 31:27CPU isolation for predictability
- 33:32Q&A session begins
- 34:14Handling modified event compatibility
- 36:03Scaling with sharding
- 37:27Frameworks and practical examples
- 39:11Persistence and thread safety
- 41:16Downstream service reliability
- 42:19Replication performance impact
- 43:24Messaging infrastructure
- 45:55Closing remarks and thanks
related talks
50mDeterministic systems: Why and how-to?
13mInsights on determinism
15mHow Deterministic Game Engines will Disrupt Multiplayer Game Development
28mNew insights into distributed and concurrent programming
25mTackling Contention : The Monster in sync.Locker
01h 00mOut-of-order execution - what can it do for me?
36mThe FIDS Theorems: Tensions between Multinode and Multicore Performance in Transactional Systems
21mChasing Unicorns and Not Losing Hope in Validating Weak Memory Persistency Model...