System Architecture Overview

This document describes the overall architecture of Go-Live, a WebRTC SFU server built with Go and Pion WebRTC.

Architecture Diagram

Component Dependency Graph

Request Processing Chain

Core Concepts

Room

The Room is the core abstraction of the SFU. Each room:

• Has at most one Publisher
• Can have multiple Subscribers
• Has independent Track Fanout logic
• Can have its own authentication token

Track Fanout

When a publisher pushes media tracks, the system creates a Track Fanout:

• Read RTP packets from the publisher's PeerConnection
• Copy and distribute to all subscribers
• Optionally write to recording files

PeerConnection

Each WebRTC connection:

• Publisher: Receives media tracks
• Subscriber: Sends media tracks
• ICE negotiation completed through WHIP/WHEP protocols

Module Responsibilities

Module	Responsibility
cmd/server	Application entry point, service initialization
internal/config	Environment variable parsing and defaults
internal/api	HTTP request handling, routing, middleware
internal/sfu	WebRTC SFU core logic
internal/metrics	Prometheus metrics exposure
internal/uploader	S3/MinIO file upload
internal/otel	OpenTelemetry tracer initialization

Key Design Decisions

1. Single Publisher per Room

Simplifies the SFU logic and ensures predictable stream quality. Multiple publishers would require stream selection or mixing.

2. In-Memory Room State

Rooms are stored in memory for simplicity and performance. For multi-instance deployment, external storage (Redis/Database) would be needed.

3. RTP Forwarding without Transcoding

The SFU forwards RTP packets directly without decoding/encoding, minimizing latency and CPU usage.

4. Recording at SFU Level

Recording happens at the TrackFanout level, capturing the exact RTP packets being distributed to subscribers.

Performance Characteristics

Metric	Value
Latency	< 100ms (same region)
Concurrent Subscribers	1000+ per room
Memory (idle)	< 50MB
CPU Efficiency	Single core handles 500+ concurrent

Next Steps

• SFU Core - Detailed SFU implementation
• Data Flow - Request and data flow diagrams
• Deployment - Deployment patterns and topology