How the Network Protocols You Choose Ultimately Affect Your Applications
summary
Compares HTTP/1.1, HTTP/2, HTTP/3/QUIC, gRPC, Kafka and message queues using nginx and simple load‑testing rigs and Go/Node/Python/Java serialization benchmarks. Reopening TCP connections costs ~20% more CPU and is ~2× slower; adding TLS shows orders‑of‑magnitude overhead (author reports ~20× CPU and ~100× slower in that setup). HTTP compression often yields only ~20% byte savings for ~40% CPU cost; browsers limit parallel connections to six. Shows reuse TCP connections, measure serialization tradeoffs (JSON wins for strings, protobuf for numeric payloads), and batch messages where possible.
AI-written summary · about
chapters (11)
- 0:00Introduction and warmup
- 3:55Why network protocols matter
- 8:14Three layers of network optimization
- 12:07Deep dive into HTTP 1.1
- 19:23HTTP 2 and HTTP 3 evolution
- 25:51Protocol choices inside vs outside clusters
- 28:17GraphQL and gRPC explained
- 32:42Message queue protocols overview
- 35:36Data serialization performance
- 40:36Optimization checklist summary
- 41:18Q&A session
related talks
35mBeyond JSON: Improving inter-app communication
40mWhirlwind tour through the HTTP2 spec
46mHTTP/3: It's all about the transport!
24mHow We Built Protosockets to Go Beyond HTTP and gRPC
23mThe Short and Happy Lives of TCP and HTTP Requests
35mOn QUIC, HTTP/3 and tech stack unification
48mChoosing the Right Technology for your API
55mForget about HTTP