14.3 Networking & Communication Protocols

1. What It Is in Plain Words

Networking and communication protocols govern how data moves between clients and servers across physical networks—covering transport layers (TCP/UDP), application standards (HTTP/1.1 vs HTTP/2 vs gRPC), and message serialization formats (JSON, Protobuf, CSV).

---

2. How THIS Project Uses It

• Application Protocol: HTTP/1.1 over TCP via Uvicorn ([Dockerfile:L15](file:///Users/avishjain/avishjain9-repos/21-bis-compliance-engine/Dockerfile#L15)).
• Data Serialization:
  • POST /chat: JSON request and response payloads validated via Pydantic ([app.py:L61,L74](file:///Users/avishjain/avishjain9-repos/21-bis-compliance-engine/app.py#L61)).
  • POST /audit/export: text/csv stream with Content-Disposition attachment header ([app.py:L148-L151](file:///Users/avishjain/avishjain9-repos/21-bis-compliance-engine/app.py#L148-L151)).
• CORS Handling: Permissive wildcard allow_origins=["*"] configured via CORSMiddleware ([app.py:L18](file:///Users/avishjain/avishjain9-repos/21-bis-compliance-engine/app.py#L18)).

---

3. Comparison with Alternatives

Protocol	Transport	Multiplexing	Payload Format	Streaming Support	Fit for This Engine
REST over HTTP/1.1 (Used)	TCP	Head-of-Line Blocking	JSON / CSV Text	Chunked Transfer	Optimal (Universal browser client support)
HTTP/2	TCP	Binary Multiplexing	JSON / Binary	Native Streams	Recommended for production proxy
gRPC	HTTP/2	High Multiplexing	Protocol Buffers	Bi-directional	Overkill for browser clients
WebSockets	TCP (Upgraded)	Full Duplex	Text / Binary Frames	Continuous	Not needed (No streaming chat)

---

4. Trade-offs Accepted

• HTTP/1.1 Head-of-Line Blocking: Sequential requests over a single TCP connection must wait for preceding requests to finish, causing minor latency queuing during rapid parallel burst queries.
• JSON Text Overhead: JSON serialization incurs ~15% higher byte size compared to binary formats like Protocol Buffers or MessagePack.

---

5. What Breaks at Scale

• At 10x Load: Standard TCP connection reuse via HTTP Keep-Alive handles the load without packet drops.
• At 100x Load: TCP socket exhaustion (TIME_WAIT saturation) can occur if clients open and close connections rapidly rather than reusing connections.
• At 1000x Load: HTTP/1.1 connection limits throttle throughput; requires an HTTP/2 or HTTP/3 reverse proxy (Nginx or Cloudflare) to multiplex thousands of streams over single TCP/QUIC connections.

---

6. How a Senior Engineer Would Improve It

Place the backend behind an Nginx reverse proxy terminating HTTP/2 with TLS 1.3, enabling single-connection multiplexing, gzip/Brotli payload compression, and strict CORS origin headers.

---

7. 30-Second Spoken Answer

"We use standard REST over HTTP/1.1 with JSON payloads and CSV streaming because it provides universal, zero-friction compatibility with browser clients and ERP export systems. For production scale, placing the service behind an Nginx proxy terminating HTTP/2 provides binary multiplexing and connection reuse with zero application code changes."