Back to Case Studies
AI Surveillance

Distributed RTSP Streaming Orchestration with Auto-Scaling

The surveillance platform needed a reliable, scalable system to manage hundreds of camera streams with automatic lifecycle management, ensuring streams are available on demand without wasting resources.

Discuss Your Project
distributed-streaming-orchestration.webp
AI Surveillance
Domain
8
Technologies
4
Key Results
Delivered
Status

The Challenge

Managing streaming infrastructure for many cameras presented operational challenges:

  • Manually provisioning streaming servers for each camera was unsustainable
  • Idle streams wasted compute resources and bandwidth
  • Cloudflare and CDN integration required HLS conversion from RTSP
  • User-scoped access control needed to ensure tenants only saw their cameras

Our Solution

We built an orchestration layer that dynamically provisions, monitors, and cleans up MediaMTX streaming containers based on demand.

Architecture

  • Orchestrator API: FastAPI microservice for stream lifecycle management
  • Container Engine: Docker-based MediaMTX container provisioning
  • Authentication: Supabase JWT for user-scoped camera access
  • CDN Proxy: Cloudflare Workers for HLS delivery
  • Health Monitoring: Periodic health checks with automatic recovery

Lifecycle Management

  1. On-Demand Provisioning - Streaming server created when user requests a camera feed
  2. RTSP-to-HLS Conversion - MediaMTX handles protocol conversion for browser playback
  3. Health Monitoring - Periodic checks ensure server responsiveness
  4. Auto-Cleanup - Idle servers terminated after configurable timeout
  5. Recovery - Unhealthy servers automatically restarted

Key Features

  1. User-Scoped Access - Each tenant sees only their authorized cameras
  2. Dynamic Scaling - Containers spun up and down based on viewer demand
  3. Quality Control - Per-stream FPS (1-60) and resolution (low/medium/high/ultra) settings
  4. Snapshot API - Timestamp-precise frame capture from live streams
  5. CDN Integration - Cloudflare Workers proxy for global low-latency HLS delivery
  6. RTSP Caching - Intelligent caching of camera connection details to minimize API calls

Results

Resource Efficiency: Only active streams consume compute resources
Zero Configuration: Cameras auto-provision on first access
Global Delivery: Cloudflare CDN ensures low-latency playback worldwide
Tenant Isolation: Complete data separation between organizations

Technology Stack

FastAPIDockerMediaMTXSupabaseCloudflare WorkersJWTWebSocketPython

Frequently Asked Questions

MicrocosmWorks built a multi-region orchestration layer where edge relay nodes at each physical location pull RTSP streams locally, transcode as needed, and forward them to the central platform via encrypted tunnels. This architecture eliminates the need for direct internet-exposed camera access, reduces WAN bandwidth by applying intelligent frame sampling at the edge, and maintains stream continuity even during network fluctuations between sites.

MicrocosmWorks implemented schedule-aware auto-scaling that pre-provisions processing capacity based on historical stream patterns, combined with reactive scaling that responds to real-time stream count changes within 30 seconds. The system scales down aggressively during off-peak hours to minimize cloud compute costs, and uses warm standby pods that can accept new streams instantly without the cold-start delay of provisioning new GPU instances.

MicrocosmWorks designed an admission control system that queues incoming stream connections and distributes them across available processing nodes using a weighted round-robin algorithm that accounts for each node's current CPU, GPU, and memory utilization. Streams are prioritized based on configurable rules, so high-priority cameras like entry points always get processing capacity before lower-priority feeds.

Yes, MicrocosmWorks built ONVIF discovery and RTSP pull adapters that connect to existing NVRs and VMS platforms, treating them as stream sources without requiring any changes to the existing recording infrastructure. The orchestration layer can also receive re-streamed feeds from popular VMS systems like Milestone and Genetec, allowing enterprises to add AI analytics capabilities to their current surveillance investment.

MicrocosmWorks delivers distributed streaming orchestration solutions at rates between $30-$50/hr, with a production-ready MVP typically requiring 3-4 months of development depending on the number of edge locations and integration requirements. This is substantially more cost-effective than enterprise video platform licenses that charge per-stream fees, especially at scale beyond 100 concurrent streams.

Have a Similar Project in Mind?

Let's discuss how we can build a solution tailored to your needs.

Contact UsSchedule Appointment