REC 00:00:00
CAM 01 – Production Area
Retention is usually discovered at the worst possible moment — an incident surfaces three weeks late and the footage has already rolled over. Storage is the cheapest part of a system to get right at design and the most expensive to fix afterward.
One megabit per second of sustained video is 10.8 GB per day. Everything after that is multiplication.
| Scenario | Bitrate per camera | Cameras | Retention | Raw storage |
|---|---|---|---|---|
| Small commercial, H.265 | 4 Mbps | 8 | 30 days | 10.4 TB |
| Warehouse, H.265, 4K at docks | 8 Mbps | 24 | 60 days | 124 TB |
| Construction site, motion only | 6 Mbps | 6 | 90 days | 35 TB |
| Distribution centre, H.264 | 12 Mbps | 48 | 90 days | 560 TB |
Rebuilding a RAID 5 array reads every sector on every remaining drive. On a large array that read volume approaches the published unrecoverable-read-error rate of consumer drives, and a second fault during the rebuild loses the array. Use RAID 6 or RAID 10 above roughly 12 TB — and treat RAID as uptime protection rather than backup. It does nothing about the recorder being stolen, which on a construction site is the realistic threat.
Desktop drives are specified for roughly 55 TB of writes a year and for substantial idle time. A surveillance recorder writes continuously from every camera and never idles. Surveillance-rated drives carry far higher annual workload ratings and firmware tuned for streaming writes, where a dropped frame matters more than a retried read. Mixing desktop drives into an array is the most common cause of premature recorder failure we see.
Insurers, agreements with shippers and client contracts frequently specify a minimum retention period, and some carry a duty to preserve footage once an incident is reported. Check what you have already agreed to before choosing a number — it is usually longer than the 14 or 30 days a recorder ships configured for.
Specified by pixel density, not by megapixels
Read more →
Credentials that cannot be copied in a parking lot
Read more →
Alarms that mean something, so someone still answers them
Read more →
Protection on day one, before there is power or a fence
Read more →
Someone watching, not just something recording
Read more →Thirty days covers most operational disputes. Sixty to ninety is common where incidents surface through insurance or a customer complaint weeks later. Check your insurance policy and any shipper or client contract first — a minimum is often already specified there.
On-premise for bulk retention, cloud for the subset that matters. A recorder bolted down on site is still the cheapest way to hold ninety days of 24 cameras, but it can be stolen. Replicating alarm events and door-forced clips off-site covers the one failure mode on-premise alone does not.
It can, and it should not. Twenty-four cameras at 8 Mbps is 192 Mbps of constant traffic. Put cameras on their own VLAN with a controlled route to the recorder, and give the recorder no direct path to the internet.
Each location page covers what is specific to that area — the industrial districts, the site types and the constraints that change the design.