Programming a Node Cluster Like It's One Really Big Computer
summary
Argues that Node clusters can be programmed like one big machine by implementing an actor model over sockets: actors get actor IDs over HTTP, a lookup table routes messages, and each actor keeps a mailbox FIFO queue. Uses Node’s cluster/process + simple HTTP connections to spawn actors on worker servers (demo runs actors on ports 3000 and 4000), sends ping/pong across two nodes, and shows a working prototype called distributed.createServer that proxies messages and spawns remote processes.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:30JavaScript
- 2:40ClientSocket
- 4:20Preforking
- 6:05Other languages
- 11:15Topology
- 12:35Shared Memory
- 13:54Ease of Building
- 14:35Friction
- 17:15Mailbox Ownership
- 17:45Multiple Recipients
- 20:05Shared State
- 20:28Actor Model
- 24:11Message Sender
- 24:26Actor IDs
- 26:28Process Interface
- 27:39Ping Example
related talks
29mWhy would we ever build a distributed computing platform in node?
30mParallel Universes: Node.js
22mMulti-tiered Node.js architectures
22mScaling Node.js and unblocking the Event Loop
24mDebuggin a Large Node Cluster
23mBuilding Distributed Systems with Node.js
19mErich Oliveira
31mKeynote Panel: Node.js at Massive Scale