How to Have your Causality and Wall Clocks, Too
summary
Claim: Distributed Monotonic Clocks produce timestamps that preserve causality while keeping a physical-time component humans understand. Implementation is a piggyback HTTP headers style library that adds an epoch reset to recover from runaway clocks and uses a self‑assembling hierarchy from population protocols to estimate a cluster median. Simulations (20–50 node random graphs) show logarithmic convergence and median estimates close enough to prevent a single bad clock from dragging the system forever.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 8:00NTP
- 16:40Interface
- 17:50Challenges
- 18:30Population Protocols
- 20:30System Clocks
- 21:45Hierarchy
- 23:20Recurse
- 24:10Visualization
- 25:35Convergence Time
- 28:00Clock Offsets
- 32:20Payload Size
- 33:55No Departures
- 34:55More Experiments
- 36:05Summary
- 40:40PowerPoint
related talks
39mKeeping Time in Real Systems
56mOrdering the chaos - cleaning logs and ordering events in microservices
14mLightning Talk: Beyond NTP: High-accuracy clocks for distributed observability
01h 02mOrdering the chaos - cleaning logs and ordering events in microservices
34mThe Holy Grail of Systems Analysis: from What to Where to Why
35mPredictable Latency
23mDoes It Add Up? Exploring the Delta Temporality in OpenTelemetry & Beyond
35mCase Study: A Thundering Herd in the Wild