Sizing UniFi Protect Storage: Retention Math for a Multi-Camera Site

Do you know how many days of footage your site is holding right now? Most operators find out the hard way, when an incident from three weeks back needs review and the timeline simply stops eleven days ago.

The appliance was not undersized on purpose. It was sized against an assumed bitrate that nobody verified after the cameras went in, and the ring buffer has been quietly overwriting the difference ever since.

Keep in mind that UniFi Protect does not throw an error when it runs out of space, which is exactly why this failure stays invisible. It keeps recording and deletes the oldest footage first, so a site that was sold on 30 days and delivers 11 looks perfectly healthy on the dashboard.

There are five places the estimate leaks, and most undersized sites are hit by three or four of them at once. None of them are exotic, but each one compounds against the others.

Resolution is not bitrate. A camera labeled 4K does not consume a fixed amount of storage; it consumes whatever bitrate is configured on it, and the default quality preset is frequently higher than the one used in whatever spreadsheet justified the purchase.

Detection recording was assumed to be cheap. Someone budgeted for a 20% duty cycle on a loading dock that sees forklift traffic for nine hours a day, and the recorder ended up writing something much closer to continuous.

RAID overhead was skipped. Four 12 TB drives are 48 TB of purchased capacity and considerably less than that once parity and formatting are accounted for, yet the retention math is almost always run against the raw number on the invoice.

The camera count grew. Sites get two more cameras at the back door in month four, and a 12-camera budget is now carrying 14 streams without anyone rerunning the arithmetic.

Night footage is more expensive than day footage. Infrared illumination introduces sensor noise, and noise is costly to encode, so a variable-bitrate stream that idles low in daylight will sit near its ceiling at 2 a.m. with rain falling through the IR field.

As a result, the honest way to size a site is to budget every camera at its configured ceiling, running continuously, against usable capacity rather than purchased capacity. Anything better than that is margin you get to keep, not a number you get to plan on.

The arithmetic is not complicated, and it does not require a vendor calculator. One megabit per second, written for a full day, is 86,400 megabits — which is 10,800 megabytes, or roughly 10.8 GB.

Bitrate in Mbps x 10.8 = GB per day, per camera. Add the cameras together, multiply by the number of retention days you owe, and you have your requirement in decimal gigabytes.

Consider a 12-camera site: eight 4MP cameras configured at 4 Mbps and four 4K cameras configured at 8 Mbps. That is 32 Mbps plus 32 Mbps, for 64 Mbps of aggregate write.

Multiply 64 by 10.8 and you get roughly 691 GB per day. Over 30 days, that site needs about 20.7 TB of usable space just to hold the footage.

Now put that same site on a UniFi Dream Machine SE with a single 8 TB drive in its 3.5-inch bay. Divide 8,000 GB by 691 GB per day and you get 11.6 days — which is where the eleven-day timeline at the top of this post comes from, and it is not a defect.

That same load on a UNVR with four 12 TB drives in RAID 5 has three drives of usable parity-protected capacity, or about 36 TB before formatting. The 30-day requirement fits with real room to spare, and the difference between the two outcomes was a purchase decision made before anyone ran the multiplication.

Note that your recorder will report roughly 9% less capacity than the label, because drive vendors count in decimal terabytes and the filesystem counts in tebibytes. Treat that shrink as a fixed tax and fold it into headroom rather than arguing with it.

Protect exposes a per-camera quality control, and on most models you can set an explicit bitrate value rather than accepting a preset. That setting, not the sensor, is what determines your storage curve.

Higher resolution does need more bitrate to look equivalent, which is the real reason 4K costs more to retain. A 4K sensor starved of bitrate will not save you money so much as it will hand you a blocky image on exactly the frame you needed to be sharp.

Variable bitrate means the configured number is a ceiling. A static interior hallway will encode well below its ceiling all day, while a parking lot with wind-moved foliage, headlights, and precipitation will push toward it and stay there.

Night and IR inflate the number. Low-light gain produces sensor noise that the encoder cannot distinguish from real detail, so overnight hours frequently consume more per hour than midday does.

H.265 is a real reduction, not a fixed discount. Where a camera supports it, H.265 delivers comparable perceived quality at meaningfully lower bitrate than H.264, but the savings depend on scene motion — and in a mixed fleet, only the cameras that support the codec get the benefit, so check the spec sheet for each generation before you assume fleet-wide savings.

Frame rate is the underused lever. Where the camera exposes a frame-rate control, dropping from 30 fps to 15 fps on scenes that do not require motion fidelity reduces demand without touching image quality on any individual frame.

What's more, audio and the smart-detection artifacts Protect stores alongside video add a small but nonzero overhead. It will not change your sizing decision, but it is one more reason not to plan to the last gigabyte. If you are still selecting hardware, the tradeoffs between sensor generations are worth working through in how to choose a UniFi camera and across the current UniFi camera lineup before you commit to a resolution mix you have to feed for the next five years.

Always-on recording is the only mode with a predictable storage curve, because the duty cycle is 100% by definition. Detections mode is where sizing estimates go to die.

The trap is that detection clips are not clean, isolated events. Protect pads each recording with pre-roll and post-roll buffer, and on a busy scene those padded events overlap into a single long clip that is functionally continuous recording with extra steps.

A quiet mechanical room might genuinely run a 5% duty cycle and stretch retention by an order of magnitude. A customer entrance during business hours will not, and neither will a street-facing camera on any road with traffic on it.

Smart detections — person, vehicle, animal, package — narrow what triggers a recording, which helps the duty cycle on scenes with a lot of irrelevant motion. They do not change the bitrate of the footage that does get written, so they are a trigger filter and not a compression strategy.

Here is the approach that actually works: size the array for continuous recording on every camera, then deploy in detections mode where it suits the scene and treat the savings as headroom. You get a retention floor you can defend in a meeting instead of a projection built on a guessed percentage.

After a week of real footage, read the actual consumption off the Protect storage page rather than trusting anyone's estimate — including this one. Real scenes at your site, in your lighting, with your traffic, are the only duty-cycle data worth planning against.

The recorder you choose sets a hard ceiling on this entire exercise. A UDM Pro or UDM SE gives you one 3.5-inch bay with no redundancy, a UNVR gives you four, and a UNVR Pro gives you more — and no amount of bitrate tuning turns a single-bay appliance into a 30-day multi-camera archive.

RAID 5 across four drives yields three drives of usable capacity and survives one failure, which is the common choice for a full UNVR. RAID 1 across two drives halves capacity for mirroring, and RAID 10 across four also halves it but rebuilds faster and tolerates certain two-drive failures.

Size your retention against usable capacity, never raw. This is the single most common arithmetic error in the entire process, and it turns a 30-day plan into a 20-day reality before a single camera is mounted.

Use CMR surveillance-rated drives. WD Purple and Seagate SkyHawk families are built for sustained sequential write with high annual workload ratings; shingled SMR drives rewrite overlapping tracks under sustained load, which is precisely the write pattern a 24/7 recorder produces.

Mixed drive sizes waste money. In a parity array the smallest member governs, so a 12 TB drive dropped into a set of 8 TB drives contributes 8 TB and leaves the rest stranded.

Be aware that rebuild time scales with drive size, and a large array rebuilding while still ingesting a dozen live streams stays degraded for a long, uncomfortable window. That is an argument for buying the array you need now rather than planning to expand a full one under load.

Finally, plan to fill roughly 80% at your target retention. The remaining fifth absorbs the two cameras added in month four, the resolution bump on the entrance camera, and the week the duty cycle spikes — without silently shortening the window you promised.

Run the sequence in this order and the purchase decision makes itself. First, list every camera you intend to deploy plus the ones you will realistically add within 24 months, because retrofit cameras are the norm and not the exception.

Second, assign each camera its configured bitrate ceiling rather than a hoped-for average. Third, multiply the aggregate by 10.8 for GB per day, then by your retention requirement in days.

Fourth, divide by 0.8 for headroom, then check the result against usable RAID capacity — not raw, and not the number printed on the drive carton. Fifth, pick the recorder and drive count that clears it, and only then start tuning bitrate and recording modes to buy back margin.

The rest of the site has to carry the load too. Every camera is a powered device on a switch, so confirm the numbers in PoE budget planning and select the uplink with how to choose a UniFi switch before a 16-camera plan meets an undersized power budget.

Cameras also belong on their own segment, both for containment and to keep continuous write traffic off the user network — the segmentation approach is covered in UniFi VLANs explained. And if anyone will be pulling clips remotely, the upstream path matters as much as the array does, which is worth reviewing alongside gateway WAN failover and how to choose a UniFi gateway.

Retention is not a feature you enable. It is an outcome of bitrate, duty cycle, and usable capacity, and the only version of this math that matters is the one you run before the hardware ships.