The Mess We're (Still) In
summary
Software systems face three concrete problems: an explosion of possible states (simple examples reach 10^124 state space), locality limits from the speed of light, and brittle API/data coupling. Proven tools and demos include CRDTs/monotone merges (PN-counter), content-addressed hashes and DHT-style replication for free redundancy, and a small Datalog engine implemented in Elixir (Datalog in Elixir). Real results: Medicanter ingests ~1.5M papers per year via BioLink and surfaces drug hypotheses that enter trials; metastable outages like a 73-hour incident still occur.
AI-written summary · about
chapters (24)
- 0:00Intro
- 4:20Quote
- 5:30Security Risks
- 8:40Introductions
- 10:00State Space
- 12:40Metastable Failure
- 16:35Local Knowledge
- 18:30Flexibility Limitations
- 19:48Dependencies
- 23:10Light Cones
- 26:28The Common Principle
- 29:01Monoton Functions
- 30:02Complex Systems
- 33:02Local First Applications
- 38:03Hashbased relationships
- 39:27Hashmap
- 40:00Redundancy
- 41:26Free Reliability
- 43:14Data Structure
- 44:21Entity Attributes Value Graph
- 46:52Medicanter
- 49:41Datalog
- 53:21Call to Action
- 56:09Conclusion
related talks
38mAdding Simplicity to Complexity
01h 01mArchitecture Decisions: A Case Study -
43mThe Tyranny of Structurelessness
43mEvident Code, at Scale
01h 17mWhen Less Is More: Designing & Decoding Complex Distributed Systems
34mDistributed Systems - What Can Go Wrong and How to Get It Right
50mDeterministic systems: Why and how-to?
27mLet's Bring System Dynamics Back to CS!