All In With Determinism for Performance and Testing in Distributed Systems
summary
Claims that replication uses a deterministic logical log so every replica applies the same operations in the same order, yielding strong consistency with millisecond latency even with synchronous persistence. Measured per node throughput is 100–500k transactions/sec and millions of SQLs/sec, with linear scale to ~30 nodes. Built a self-checking workload called TransactionalitySelfcheck that hashes replica state, injects faults and adversarial workloads, and detects divergence (current behavior is to snapshot divergent replicas and stop the cluster).
AI-written summary · about
chapters (4)
- 0:00Leveraging Determinism
- 36:27Data Consistency Checks
- 36:55Testing with Workloads
- 38:02Bug Detection and Conclusion
related talks
01h 02mConsistency and Agreements in Distributed Systems
35mConsistent Updates in a Distributed System without Transactions
45m(...)Distributed Systems
28mAddressing the parallelism gap in replicated server systems
43mDistributed Systems
01h 02mAt Least Once
01h 00mThe Secret Life of Distributed Systems
40mwant don't You consistency no stinking