Fixing Erlang’s Distribution Protocol
summary
Claim: extend Erlang distribution so it handles 1,000 nodes and hostile networks by adding routing, sequencing and pluggable transports. Evidence: a port to a hard‑RTOS and a 15‑node demo, plus a laptop demo showing atom table optimizations. Concrete fixes reported include per‑process sequence numbers to preserve per‑pair ordering, message slicing to avoid head‑of‑line blocking, use of hidden nodes today and proposed dynamic link management, and plans to push changes into OTP.
AI-written summary · about
chapters (19)
- 0:00Intro
- 0:50Self-networked units = multi channel virtual device
- 8:44Hidden nodes
- 9:31Fixing Erlang Distribution
- 13:22OTP Team ongoing
- 16:13epmd
- 18:11Plug in another distribution transport protocol • Command line-proto_dist mod makes distribution call
- 19:48Node to Node Link
- 27:18Message Routing
- 29:11Message Forwarding
- 31:34Static routing
- 33:11Routing protocols
- 34:41Link State Protocols
- 36:25Variants in use
- 36:59Routed Messages
- 38:35Invariants to be observed
- 41:23Sequence Numbers
- 44:12Message Sending Options
- 45:07Head of Line Blocking
related talks
15mREAL: Emulating Control Plane at Simulator’s Cost
33mNon-Imperative Network Programming
31mInteroperability in Elixir: Dealing With the World Outside of the Beam
46mLive Distributed Spying
43mReverse-Engineering a Proprietary Game Server with Erlang
43mConcurrency Abstractions for Application Security
40mBroken Brokers in Boxes: Fuzzing Breaks Everything, Even Erlang
30mVerifying Reliable Network Components in a Distributed Separation Logic with Dep...