Understanding AI Applications Networking Specifics

Below is a summary artefact of my learning session with Claude Sonnet 4

Websockets aren't strictly related but they are often compared to SSE and I needed a refresher to understand the intricacies a bit better

Transport Protocols

SSE: Best for server→client notifications with automatic reconnection

HTTP Streaming: Flexible for any server→client streaming (AI responses, downloads)

WebSockets: Required for bidirectional communication, but more complex infrastructure

Feature	Server-Sent Events (SSE)	HTTP Streaming	WebSockets
Protocol	HTTP/1.1 or HTTP/2	HTTP/1.1 or HTTP/2	WebSocket protocol (ws://, wss://)
Direction	Server → Client only	Server → Client only	Bidirectional
Connection Type	Long-lived HTTP connection	Long-lived HTTP connection	Persistent TCP connection
Data Format	Structured events (data:, event:, id:)	Any format (JSON, text, binary)	Any format (text/binary frames)
Reconnection	Automatic with last event ID	Manual implementation	Manual implementation
Infrastructure	Works with standard HTTP proxies/LBs	Works with standard HTTP proxies/LBs	May need proxy configuration
Complexity	Simple	Medium	Higher
Use Cases	Real-time notifications, live feeds	AI streaming responses, file downloads	Chat, gaming, real-time collaboration
Connection Overhead	HTTP headers on initial request	HTTP headers on initial request	WebSocket handshake overhead
Server Implementation	Simple (flush response buffer)	Simple (flush response buffer)	Complex (frame handling, ping/pong)
Compression	Standard HTTP compression	Standard HTTP compression	Per-message compression extensions