A Software Architecture Based on Coarse-Grained Self-Adjusting Computations
summary
A commercial health-record pipeline implements coarse-grained self-adjusting computations in Haskell to push precomputed UI models to web and mobile clients with offline support. The system handles about 140,000 HL7 messages per day and 6.5M DICOM images, caches intermediate results with full or hash strategies, and adds on‑demand generation for rarely accessed patients. Startup currently takes up to two hours after restart, but the deployment runs in several large German hospitals and the approach scales with active views.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 1:24Hospital data system motivation
- 3:10Push vs pull data updates
- 4:37Self-adjusting computations
- 6:04Live demo
- 9:54Demo system architecture
- 11:44Framework API and Haskell
- 18:48Hospital commercial use
- 23:05System architecture overview
- 26:58Ten years of insights
- 30:49Trade-offs and disadvantages
- 32:13Push vs pull comparison
- 33:37Why Haskell
- 34:46Summary and open source
- 35:29Q&A: Build systems vs SAC
- 36:54Q&A: Bidirectional data flow
- 38:38Q&A: Dynamic graph reconfiguration
- 40:21Q&A: Strong typing in projects
related talks
22mProgramming HealthCare Silos
52mBuilding Scalable Web Applications with Haskell
49mArchitect’s Survival Guide to Healthcare -
20mFour Years of Embedded Haskell in Critical Real-Time Systems: Lessons and Insights
01h 01mArchitecture Decisions: A Case Study -
46mBack to the future - how we integrated a 40 year old software system
31mClosing Gaps From Commercial Tools in German Redisp...
26mFrom OO Patterns to Functional Programming w/o Killing Yourself