Approaches to Ultra-Long Software Maintenance
summary
Argues that some devices require 20–40 years or even 80 years of support and outlines concrete practices: restrict devices to fixed and trusted software, avoid unnecessary network exposure, insist on mainline kernel support not vendor BSPs, keep full source and reproducible build environments, prefer userland drivers and virtual machines to decouple hardware, and create expert-curated backport lists with patch annotations for applicability. Claims these steps reduce patching risk and make decades-long maintenance feasible.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:33Agenda
- 2:15Disclaimer
- 4:02Appliances
- 6:20Fundamental Questions
- 7:23Life Spans
- 9:31Longterm Maintenance
- 14:42Maintenance Alternatives
- 23:12Types of Risk
- 27:20Technical Aspects
- 38:20Source Code
- 42:40Conclusion
- 45:35Wishlist
related talks
40mThe Need to Drive Towards Frequent and Sustained Software Updates
01h 03mLong-lived software architectures that survive decades without debt
44mReflections around SW obsolescence
07mTechnologies That Last
23mCan Software Live for 50 Years or More?
23mDeprecating Stuff
38mBoF: Bash the Kernel Maintainers
35mLessons Learned for reusable firmware