Toward a High-Performance JSON Protocol: Notes
summary
Claim: a new JSON-first protocol speeds CDN-style updates by using scratch Delta over UDP with UDT for transport and Avro-style Avro schemas appended to datagrams for self-describing payloads. A prototype runs on AWS; an AWS global 5-node circularbuffer test shows 338 ms for a single-link global update, ~1,200 ms for 1,000 updates, and ~3× faster transfer of a 35 KB homepage versus HTTP/TCP. Packet loss is low in tests, Node UDP support and SSL/TLS remain unsolved bottlenecks.
AI-written summary · about
chapters (24)
- 0:00Introduction
- 1:05Agenda
- 2:38Why did I call it scratch
- 3:38What is scratch
- 5:37Motivation
- 7:56Goals
- 9:36Design Decisions
- 10:49Faster Trash
- 11:35Why UDP
- 12:54How we got there
- 13:53Too many features
- 14:21Using the result
- 18:49Time Check
- 19:08Current State
- 20:10Special Pleading
- 21:34Test Results
- 23:28Packet Loss
- 26:51No Demo
- 27:31Network Friendly
- 28:44Metadata
- 31:30Whats Next
- 35:40Smarter Faster
- 36:56Science Fiction
- 40:49Questions
related talks
09h 41mJSConf Chile Day 1
01h 20mEfficient Client-Server Communication with Differential Synchronization and JSON Patch
24mHow We Built Protosockets to Go Beyond HTTP and gRPC
52mDeath, Taxes, and HTTP
01h 04mAPI Design as a Performance Primitive
32mOptimize Your JSON Payload Efficiency x10 times
53mMastering Node.js Performance
03h 16mPUSHing web app performance with HTTP/2