Surviving Fire: using Erlang as an OS to Achieve Massive Fault Tolerance
summary
Claim: an operating system built from Erlang with one BEAM per core runs on the metal and tolerates process, node, CPU core, RAM and disk failures. Implementation specifics: a bespoke bootloader, a tiny C layer (about 83 KB of stdlib and only 24 C functions needed to start the BEAM), Erlang drivers and OS services, shared‑memory message buffers for inter‑core messaging, a capability system and support for uni‑kernels. Hydros runs in QEMU/KVM/Xen and on real hardware and is actively fault‑tested by interrupting and damaging machines.
AI-written summary · about
chapters (15)
- 0:00Start
- 1:30General Project Aim
- 2:13Potential Usage Environments
- 7:40Bootloader
- 10:15C Standard Library
- 13:13Device Driver Management
- 14:45Concurrent Init System
- 16:20Ethernet Driver
- 16:37Kernel Layer
- 18:45Global Layer
- 22:38Control Scheme
- 28:58Questions
- 29:35Level Triggered Interrupts
- 30:29Potential Upstream Changes
- 31:57Process Mobility System
