Building Robust Systems with Weakly Interacting Feedback Structures
summary
Claims a reusable design pattern called Weakly Interacting Feedback Structures that composes multiple feedback loops to maintain one global property, not vague examples. Implements it in the SELFMAN project with Scalaris key/value store, using a structured overlay, neighbor connectivity and finger-based routing, replica management, load balancing and consensus for transactions — i.e. five WIFS. Shows TCP and human respiration as analogues, reports Scalaris running Wikipedia backends, scaling to dozens–hundreds of nodes and winning an IEEE scalable computing prize, so the pattern demonstrably works.
AI-written summary · about
chapters (26)
- 0:00Intro
- 0:29Overview
- 1:32Designing robust distributed systems
- 3:18Feedback loops are everywhere
- 3:57Limits of feedback loops
- 5:01Simple example: hotel lobby
- 6:58Hotel lobby solution
- 8:30TCP uses a similar approach
- 9:33Erlang execution model
- 10:15Primitive failure detection
- 11:07Supervisor trees
- 12:31Subsumption architecture
- 13:12An obstacle-avoiding robot
- 15:14Human respiratory system
- 20:56Design of respiratory system
- 22:13Respiratory system state diagram
- 23:33Feedback structures
- 24:29A scalable architecture in four steps
- 26:18Scalaris keylvalue store
- 27:43Scalaris design uses 5 WIFS
- 30:52Scalaris is based on a structured overlay network
- 32:10Scalaris scalability
- 32:58Smart components are essential
- 34:40Power is built in, not added on
- 35:54Intelligence is a combination of many smart components
- 36:59Conclusion and the future
related talks
44mArchitectural Patterns of Resilient Distributed Systems
39mVirtuous Cycles: Enabling SRE via automated feedback loops
44mArchitectural patterns of resilient distributed systems
30mDistributed system design
20mFunctional Design for Reference Architectures
44mDesigning for Autonomy
48mFault Tolerance on the Cheap Making Systems That Probably Won't Fall Over
44mSelf Contained Systems: The Good, the Bad & the Ugly