Little things and big things: lessons for building large systems.
summary
Shows concrete claims and reproducible fixes: PNsol’s mobile small‑cell deployment fails because short spikes (0.4 s, seconds‑long) fill remote buffers and controllers time out; MM1K queue math explains why the system is wild at load≈1 and why big buffers are bad. Rules: when overloaded the system must recover, provide minimum functionality, and be manageable. For chronic overload use layered time‑scales and temporary overloads mitigation/propagation. Recommend internal avoid consensus by using deterministic distributed functional code and CRDTs (Lasp scaled to ~1,000 nodes).
AI-written summary · about
chapters (7)
- 0:00Introduction to system design
- 3:27Principles of scale and risk
- 11:31Industrial case study
- 19:36Managing temporary overloads
- 29:55Handling permanent overloads
- 36:21Reducing non-determinism
- 45:55Phenomena in large systems
related talks
34mDistributed Systems - What Can Go Wrong and How to Get It Right
25mAvoiding Overload in Distributed Systems
40mEverything Will Flow
27mLessons Learned From Architecting the Highest-scale Operational Systems in the World
42mFive Common Mistakes With Distributed Systems
35mSystems Scale Summer Q&A Week 1
01h 02mTools for simplified large scale systems
51mPatterns for building resilient software systems