RUSEON Core vs Frigate NVR
Frigate NVR is a complete open-source NVR with real-time AI object detection (using Coral Edge TPU or OpenVINO). It is widely used in the home automation and self-hosted communities.
While Frigate is an application focused on AI detection and object tracking, RUSEON Core is a high-throughput video data infrastructure and ingest kernel.
1. Architectural Differences
1. CPU & Ingest Footprint (Zero-Transcoding vs FFmpeg)
- Frigate NVR: Uses multiple FFmpeg processes per camera stream (detect stream, record stream, audio transcoding). Running 15–20 camera streams in Frigate can saturate 8–16 CPU cores due to software decoders and Python/C interop.
- RUSEON Core: Uses Zero-Copy NALU Passthrough in pure Go. It extracts raw video frames without decoding pixels on the CPU. A single machine can ingest 600 simultaneous 30 FPS cameras (18,139 FPS, 1.34 Gbps total ingest) using under ~3.6 CPU cores on a 12-core server.
2. Storage Subsystem & Linux Page Cache
- Frigate NVR: Relies on FFmpeg segmenting to write
.mp4chunks. Under heavy load or slow disk I/O, Python/FFmpeg processes can experience I/O wait bottlenecks and memory buildup in the Linux kernel cache. - RUSEON Core: Features native Fragmented MP4 (fMP4) recording integrated with
sync_file_rangeandPOSIX_FADV_DONTNEED. Flushes video chunks sequentially and instructs the kernel to drop video pages from RAM, keeping RSS memory strictly flat (~471 MB under 600 cameras).
3. Synergistic Integration: RUSEON + AI
RUSEON Core can serve as the high-performance video data infrastructure underneath AI engines:
- RUSEON handles high-density RTSP ingestion, rock-solid 24/7 crash-proof recording, and WebRTC streaming.
- Downstream AI workers (YOLOv11, Coral TPU, ONVIF analytics) connect to RUSEON via low-latency gRPC (
FrameService.StreamFrames) or WebRTC DataChannels with zero CPU burn on the ingest node.
2. Comparison Summary Table
| Aspect | RUSEON Core | Frigate NVR |
|---|---|---|
| Architecture | High-Throughput Go Video Engine | Application-Level AI NVR |
| Transcoding Overhead | Zero (Raw NALU Passthrough) | Moderate/High (FFmpeg decoders) |
| Max Streams per Node | 600+ Cameras (18,139 FPS / 1.34 Gbps) | ~10 – 25 Cameras |
| Page Cache Protection | POSIX_FADV_DONTNEED Built-in | None (OS Default) |
| Hardware Failure Safety | Isolated Lock-Free RingBuffer | Process Restart / Buffer Queue |
| AI Metadata Stream | gRPC & WebRTC DataChannels | MQTT / Internal Python DB |