SRS — Simple Realtime Server for Live Streaming

Introduction

SRS (Simple Realtime Server) is a real-time video streaming server that supports RTMP, HLS, HTTP-FLV, SRT, WebRTC, and GB28181 protocols. It is designed for live broadcasting, video conferencing, and low-latency streaming scenarios at scale.

What SRS Does

• Ingests live streams via RTMP, SRT, or WebRTC and distributes them in multiple formats
• Transcodes and remuxes streams on the fly using built-in or FFmpeg-based pipelines
• Provides an HTTP API and web console for managing streams, clients, and server state
• Supports edge-origin clustering for horizontal scaling across multiple servers
• Records live streams to disk in FLV or HLS segment format

Architecture Overview

SRS is a single-process, multi-coroutine C++ server built on the ST (State Threads) library for high-concurrency I/O. It handles protocol negotiation, media demuxing, and format conversion internally. For clustering, edge servers pull streams from origin servers on demand. The HTTP API exposes real-time metrics and stream lifecycle hooks that can trigger external callbacks.

Self-Hosting & Configuration

• Deploy with Docker using ossrs/srs:5 or compile from source on Linux/macOS
• Configure via a single srs.conf file with declarative blocks for each protocol
• Enable WebRTC by setting the candidate IP and UDP port range in the config
• Use the built-in HTTP callback system to integrate with authentication and billing services
• Scale horizontally by deploying edge nodes that pull from origin servers

Key Features

• Sub-second latency with WebRTC and HTTP-FLV delivery
• Protocol conversion between RTMP, HLS, HTTP-FLV, SRT, and WebRTC in a single server
• Edge-origin architecture for CDN-like stream distribution
• Built-in DVR for recording live streams to files
• Lightweight single-process design that handles thousands of concurrent connections

Comparison with Similar Tools

• Nginx RTMP Module — Adds RTMP to Nginx but lacks WebRTC and SRT; SRS is purpose-built for streaming
• MediaMTX — Go-based multi-protocol server; SRS has a longer track record and richer clustering support
• Janus — Focused on WebRTC signaling and media; SRS covers broader ingest-to-delivery workflows
• Owncast — Self-hosted live streaming with chat UI; SRS is a lower-level server without a viewer frontend

FAQ

Q: Can SRS handle WebRTC for video calls? A: Yes. SRS supports WebRTC publishing and playback, making it suitable for low-latency interactive streaming. For multi-party conferencing, you may need a selective forwarding unit (SFU) layer.

Q: What is the expected latency for live streams? A: HTTP-FLV and WebRTC deliver sub-second latency. HLS latency is typically 3-10 seconds depending on segment length settings.

Q: Does SRS support GPU transcoding? A: SRS itself does not transcode with GPU, but it can forward streams to an FFmpeg process configured with hardware acceleration.

Q: How many concurrent viewers can SRS handle? A: A single SRS instance on modern hardware can serve thousands of concurrent connections. For larger scale, use the edge-origin clustering model.

Sources

• https://github.com/ossrs/srs
• https://ossrs.io