Redundant design of monitoring system: dual-machine hot standby, N+1 and link redundancy
First figure out what to guard against
More redundancy is not better. Acceptable risks must be defined first: Is it a single camera failure (acceptable), or is the entire area recording interrupted (unacceptable)? Different answers correspond to completely different inputs.
Three levels of redundancy
| Level | Plan | What to prevent | Cost |
|---|---|---|---|
| Storage | RAID 5/6, N+1 hot spare disk | Single/double hard drive damaged | Low |
| Host | NVR dual-machine hot backup | The entire NVR fails | Medium |
| Link | Link aggregation, dual uplinks | Single network cable or optical port failure | Low to medium |
| Power supply | UPS, dual power supply | Main power interruption, power module failure | Medium |
What is "false redundancy"
- Both NVRs are connected to the same ordinary switch - if the switch is broken, both NVRs will fail.
- RAID is set up but there is no hot spare disk - when the second disk is damaged, there is no time to rebuild it.
- The UPS was connected but no load discharge test was performed - the actual backup time was much lower than the nominal.
- Dual uplinks but both take the same physical route - when the optical cable is cut, it will be cut together.
Recommendations by size
Less than 64 channels: RAID + hot spare disk + UPS is enough, host redundancy is not cost-effective. 64~256 channels: The core NVR or storage server is used as a dual machine, and the front-end switch is used for uplink aggregation. Greater than 256 channels: Complete architectural redundancy should be implemented, including storage clusters, Manage two machines on the platform and two machines on the core switch, and conduct regular switching drills.
Don’t forget to practice
After the redundancy solution is completed, a real switching drill must be conducted: unplug the power supply of the main device, Observe the time taken for backup takeover and the completeness of the recording. Unrehearsed redundancy, In the event of a real failure, it often fails due to configuration details.