Storage Tiers: Hot, Warm, Cold, and Archive

Why this matters

In our companion article on surveillance storage and the retention math, the headline was that retention — how many days you keep footage — is the dominant multiplier on the storage bill. This article is about the other half of that bill: not how long you keep footage, but what kind of storage you keep it on as it ages. It is for the security integrator, product manager, or operations lead who has seen a storage quote, blinked at the number, and wondered whether all of it really needs to sit on the same fast disks. The answer is almost always no, and the gap between a single-tier design and a tiered one is frequently half the storage bill or more. The trade you are managing is cost against retrieval speed, and getting it wrong in either direction — paying hot prices for year-old footage, or burying last week's incident in an archive that takes twelve hours to read back — is the mistake this article is written to prevent. No special technical background is needed; every number is worked out in plain terms.

The one idea: footage gets less valuable as it ages, so make its storage cheaper

Think about how an office handles paper. Today's documents sit on the desk, within arm's reach. Last month's go into a drawer. Last year's go to a box in the basement. The decade-old records go to an off-site document warehouse you have to phone a day ahead to retrieve. Nobody keeps ten years of paper on the desk, because desk space is the most expensive space in the building and old records almost never need to be touched. Storage tiering is exactly this idea applied to recorded video: the footage you are most likely to need right now lives on the fastest, costliest storage, and as each recording ages and its chance of being watched drops, it slides to storage that costs less and is slower to read.

The reason this works is that surveillance footage has a steep value curve. The clip from an hour ago might be pulled up live by an operator. Yesterday's footage is reviewed when someone reports an incident. Last month's is occasionally pulled for an investigation. Footage from eight months ago is almost never watched — it exists only because a retention policy, an insurer, or a regulator says it must be kept. Paying the same premium price to store the eight-month-old footage as the one-hour-old footage is pure waste, and on a large system that waste runs to real money every month.

One number from the retention-math article sets the stakes. A modest 40-camera site recording continuously at 2 Mbps generates about 21.6 GB per camera per day, or roughly 0.86 TB across the site every day. Hold that for 90 days and you are storing about 78 TB at any moment. Whether those 78 TB sit on one expensive tier or are spread across four is the difference this article is about.

Figure 1. Footage ages left to right; each tier trades retrieval speed for a lower cost per terabyte, until a retention policy deletes it.

The four tiers, from hot to archive

"Tier" just means a class of storage with a particular balance of speed and cost. Most surveillance and cloud designs use four, named by temperature — a useful metaphor where "hot" data is in active use and "cold" data has gone quiet. Here is each one, in plain terms, with both where it lives on-premises and the cloud storage class it corresponds to.

Hot storage holds the footage in active use: the live streams being written right now and roughly the last few days, the window where an operator or an investigator is most likely to scrub through video at full speed. Hot storage gives instant access — latency measured in milliseconds — and it is the most expensive per terabyte. On-premises, hot is fast solid-state storage (NVMe or SSD) or a fast hard-disk RAID array taking the live write load. In the cloud it is the default object-storage class, what Amazon calls S3 Standard.

Warm storage holds footage that is no longer in daily use but still needs to come back quickly when asked — think the last few weeks, the period an incident report might still reference. Warm storage keeps instant, millisecond access but costs noticeably less than hot, the trade being that the provider assumes you will read it infrequently and charges a small fee each time you do. On-premises, warm is typically bulk hard-disk RAID — the high-capacity surveillance drives that hold the bulk of recorded video. In the cloud it maps to an "infrequent access" class such as S3 Standard-IA or S3 Glacier Instant Retrieval.

Cold storage holds footage that is rarely touched — months old, kept because a retention policy requires it, not because anyone expects to watch it. The defining change at the cold tier is that instant access goes away: retrieving cold footage takes from minutes to a few hours, because the storage is optimized for cheapness, not speed. In exchange the cost per terabyte drops sharply. Cold has no exact on-premises equivalent on a small system, but on larger ones it is slower, denser disk or a nearline library; in the cloud it is an archival class such as S3 Glacier Flexible Retrieval.

Archive storage is the deepest, cheapest, slowest tier: footage kept for long legal or compliance retention that essentially nobody expects to read, where a retrieval delay measured in hours is perfectly acceptable because you would only ever pull it for a subpoena or an audit. Archive is the cheapest storage per terabyte by a wide margin. On-premises, archive is classically magnetic tape — modern LTO tape cartridges store tens of terabytes each at a few dollars per terabyte and sit on a shelf drawing no power. In the cloud it is the deep-archive class, such as S3 Glacier Deep Archive, where retrieval is measured in hours.

The single rule that ties the four together: as you go from hot to archive, the cost per terabyte falls and the time to read the footage back rises. You are always trading money against retrieval speed. Tiering is the discipline of putting each recording on the cheapest tier whose retrieval speed you can still live with.

Figure 2. The four tiers compared — access speed, typical access frequency, on-prem media, cloud class, and relative cost.

What each tier costs: the cost-per-terabyte math

The case for tiering is only as strong as the price gap between the tiers, so look at the actual 2026 numbers. Cloud object storage publishes its prices per gigabyte per month, which is the cleanest way to see the spread; the figures below are Amazon S3 list prices in its main US region, converted to dollars per terabyte per month so they are easy to compare.

Tier	Cloud class (example)	Cost per TB / month	Read-back latency
Hot	S3 Standard	~$23.00	milliseconds
Warm	S3 Standard-IA	~$12.50	milliseconds (+ retrieval fee)
Warm/Cold	S3 Glacier Instant Retrieval	~$4.00	milliseconds (+ retrieval fee)
Cold	S3 Glacier Flexible Retrieval	~$3.60	minutes to hours
Archive	S3 Glacier Deep Archive	~$0.99	hours

Read the top and bottom rows together: storing a terabyte on the hot tier costs about $23 a month; the same terabyte on deep archive costs about $1. That is roughly a twenty-three-fold difference for holding the identical footage, and it is the entire reason tiering exists. The cheapest tier is not 10% cheaper, it is more than 95% cheaper.

On-premises the economics rhyme, though you buy the storage outright instead of renting it monthly. In 2026 the rough purchase prices per terabyte are about $60–80 for fast NVMe solid-state storage (hot), about $18 for purpose-built surveillance hard drives (warm) — for example Western Digital's Purple or Seagate's SkyHawk lines — and about $5 per terabyte for LTO tape cartridges (archive), which also draw no power sitting on a shelf. The same shape holds: the hot tier costs roughly an order of magnitude more per terabyte than the archive tier.

One 2026 wrinkle worth knowing, because it makes tiering more valuable this year than last: hard-drive prices have surged — industry trackers reported average increases of around 46% over a four-month span — and enterprise solid-state prices have spiked harder still on a memory-chip shortage, widening the gap between fast and bulk storage. When fast storage gets more expensive relative to bulk, the penalty for leaving old footage on the wrong tier grows. Tiering is a hedge against exactly that.

Figure 3. Cloud cost per terabyte per month, hot to archive — the price gap that makes tiering worth the effort (2026 list prices, illustrative).

Lifecycle policies: the rule that moves footage downhill on its own

Tiering only saves money if footage actually moves between tiers, and no one is going to do that by hand for millions of video files. The mechanism that automates it is a lifecycle policy — a set of age-based rules, sometimes called information lifecycle management or ILM, that tells the storage system: when a recording reaches this age, move it to that cheaper tier; when it reaches the end of its retention, delete it.

In cloud storage this is a first-class feature. Amazon's S3 Lifecycle configuration lets you write transition rules ("after 7 days, move to Standard-IA; after 30 days, move to Glacier Flexible Retrieval; after 90 days, move to Deep Archive") and an expiration rule ("after 365 days, delete"), and the platform applies them automatically to every object. There is also a hands-off variant — S3 Intelligent-Tiering — that watches how often each file is actually read and moves it between a frequent-access, infrequent-access, and archive-instant tier for you, charging a small monitoring fee instead of asking you to predict the access pattern up front. On-premises, VMS platforms and storage systems implement the same idea through their own archiving and hierarchical-storage features, moving aging recordings from the fast array to bulk disk to tape on a schedule you set.

The policy is where the design decisions live, and the two levers are the transition ages (how many days before footage drops to each cheaper tier) and the expiration age (when it is deleted for good). Set the transitions to match how your footage is actually watched — keep the window an operator scrubs daily on hot, push everything past it down quickly — and the bill falls without anyone noticing a difference in day-to-day use. The expiration age, crucially, is not just a cost setting; it is where storage meets the law, which we come back to below.

Figure 4. A lifecycle policy transitions footage to cheaper tiers as it ages and deletes it at the retention limit — all automatically.

The worked example: one bill, three ways

Put numbers on it. Take a site holding 100 TB of footage under a 90-day retention policy, with video arriving at a steady rate each day. Because ingest is steady, the amount of footage at each age is even — a tenth of it is in any 9-day band — so the share sitting in each tier is just the fraction of the 90 days that tier covers.

First, the naive design: keep everything on the hot tier.

100 TB × $23 / TB-month = $2,300 per month

Now tier it. Say the last 7 days stay hot (operators scrub them live), days 8–30 go to warm, and days 31–90 go to cold:

Hot   (days 1–7):   7/90 × 100 TB = 7.8 TB × $23.00 = $179
Warm  (days 8–30): 23/90 × 100 TB = 25.6 TB × $12.50 = $320
Cold  (days 31–90):60/90 × 100 TB = 66.7 TB × $3.60  = $240
                                              total ≈ $739 per month

The tiered design costs about $739 a month instead of $2,300 — roughly a 68% cut — and an operator pulling up yesterday's or last week's footage notices nothing, because everything inside 30 days is still on instant-access storage. Push the oldest footage to deep archive instead of cold, on a system with a year-long retention, and the saving grows again, because that bottom band collapses from $3.60 to about $1 per terabyte.

The lesson is the shape, not the exact percentage: most of the footage on any surveillance system is old, rarely watched, and kept only for retention — and that is precisely the footage a tiering policy moves to storage costing a fraction of the hot tier. The cameras and the operators see no change; only the bill does.

The retrieval-cost trap: cheap to store, expensive to read

Here is where teams get hurt, and it is the one thing to understand before turning tiering on. The cheap tiers are cheap to store. They are not cheap to read back. Three traps follow from that, and all three are avoidable.

Common mistake: treating the archive tier's storage price as the whole price. Cold and archive tiers charge a separate retrieval fee every time you read data out, and on the deep tiers that fee can dwarf a month of storage. If an investigation forces you to pull back tens of terabytes of cold footage, the retrieval charge — plus the wait, which can run to hours on deep archive — can erase months of storage saving in a single afternoon. The fix is to tier by how often footage is actually read, not by age alone: if a class of footage gets pulled for investigations even occasionally, it belongs in a tier with cheap or instant retrieval, not in deep archive. Guess the access pattern wrong and the "cheap" tier becomes the expensive one.

The second trap is minimum storage durations. The cheaper cloud tiers bill you for a minimum residency no matter when you delete: 30 days for infrequent-access, 90 days for the Glacier classes, 180 days for Deep Archive. If your retention policy deletes footage at 30 days but you moved it to a tier with a 90-day minimum, you pay for 90 days anyway — the tier was cheaper per month but you bought more months than you needed. The rule of thumb: never move footage to a tier whose minimum duration is longer than the time you will actually keep it. Short retention belongs on hot and warm; only long retention earns the deep tiers.

The third trap is writing continuous video straight to a cold tier. Surveillance is a firehose of constant writes, and cold/archive tiers charge per write operation and are built for occasional, large objects — not a relentless stream of new files. Recording directly into archive can run up per-request charges that swamp the storage saving, and it defeats the point anyway, because brand-new footage is the hottest data there is. The correct pattern is always the same: write into the hot tier, which is built to absorb the continuous write load, and let the lifecycle policy carry footage down to the cheap tiers as it ages.

Where the standards fit: ONVIF Profile G and the storage layer

Tiering happens underneath the surveillance software, but it has to stay visible through it, and that boundary is worth drawing. ONVIF is the common language that lets cameras and recording software from different makers work together, and it splits that language into profiles by job. The one that matters here is ONVIF Profile G, the open standard for recording and retrieval — current version 1.1, published October 2025. Profile G is how a Video Management System (the software that records and manages many camera streams, the VMS) controls recording on a device and, crucially for tiering, how it searches and replays stored footage: a conformant client can find recordings and events (FindRecordings, FindEvents) and play them back across vendors.

Here is the boundary. Profile G standardizes the interface to recorded video — how the VMS asks for a clip from a given camera and time, and gets it back. It says nothing about where the bytes physically sit — which tier, which medium, on disk or on tape or in a cloud archive. Tiering and lifecycle live entirely in the storage layer beneath the VMS, and from the VMS's point of view a tiered request still comes through the same Profile G retrieval interface. Remember the rule from our ONVIF explainer: ONVIF guarantees a baseline both devices conform to, not the full implementation underneath. For the commercial overview of how the profiles map to real products, Fora Soft's guide to ONVIF profiles in security systems is the companion read.

There is one practical consequence of that boundary, and it is the reason tiering and the VMS have to be designed together. If a request reaches down into a cold or archive tier, the footage will not come back in milliseconds — it may take minutes or hours while the tier restores it. A VMS expecting instant playback can time out or simply show nothing. So the tiering design has to match the VMS's and the operators' expectations: keep anything that needs instant scrub on hot or warm tiers, and treat the cold and archive tiers as a deliberate "request and wait" path for old footage, not as something an operator hits by accident mid-investigation. The hardware that holds these tiers on-premises — the fast array, the bulk RAID, the tape library — is the subject of our on-prem storage architecture article; the cloud and hybrid side, including the egress and retrieval costs in depth, is covered in cloud and hybrid storage for surveillance.

Figure 5. ONVIF Profile G standardizes the recording and retrieval interface; the tiers and the lifecycle engine live in the storage layer below it — and deep-tier retrieval is not instant.

Tiering is also a retention-and-privacy lever

Moving footage to a cheaper tier is a cost decision, but the expiration end of the lifecycle policy is a compliance one, and the two are easy to confuse. Under the EU's General Data Protection Regulation (GDPR, Regulation (EU) 2016/679), recognizable video of people is personal data, and Article 5(1)(e), the storage-limitation principle, requires that it be kept "no longer than is necessary" for the purpose it was collected for. The European Data Protection Board's Guidelines 3/2019 on processing of personal data through video devices put a number on the norm: footage should in most cases be erased, ideally automatically, after a few days, with longer retention requiring documented justification. (This is engineering guidance, not legal advice — confirm the specifics for your jurisdiction with qualified counsel.)

The danger that tiering introduces is subtle: the archive tier is so cheap that footage can quietly outlive its lawful retention there, forgotten, because nobody feels the cost pressure that would normally prompt a cleanup. An archive that costs a dollar a terabyte a month is easy to ignore — and an old recording of identifiable people, kept past its justified period because it was cheap to keep, is a storage-limitation problem regardless of how little it cost. The discipline that prevents this is the same lifecycle policy that saves the money: the expiration rule must enforce the retention limit automatically, deleting footage at the end of its lawful life on every tier, archive included. Tiering done right makes deletion automatic; tiering done carelessly makes hoarding automatic. The detail of how long different footage should be kept is the subject of our retention policy article, and the privacy cap on the maximum is governed by the storage-limitation rule above. The takeaway: set the expiration rule with the same care as the transitions, and let it run on the archive tier too.

Where Fora Soft fits in

Storage tiering is one of those design choices that is invisible when it is right and very visible on the invoice when it is wrong. Fora Soft has built video streaming, real-time video, and computer-vision systems since 2005 — more than 625 shipped projects — and surveillance sits at the intersection of all three. When we build or customize a VMS, we design the storage as tiers rather than one flat pool: the live write load on storage built to absorb it, a lifecycle policy that carries footage down to bulk and archive as it ages, transition and expiration ages set to the client's real access pattern and retention rules, and the cold and archive paths kept clear of anything an operator needs in a hurry. The accuracy-vs-performance habit applies here too — we size the hot tier for the retrieval speed operators actually need under load and push everything else down, because a system that pays hot prices for footage nobody watches, or buries last week's incident in an archive that takes hours to read, is a design that has confused cost with value.

What to read next

• How surveillance storage works: the retention math
• Cloud and hybrid storage for surveillance
• Retention policy: how long to keep footage, and why

Download the storage-tiering decision guide (PDF) — the four tiers at a glance, the cost-per-terabyte table, the lifecycle-policy recipe, the worked 100 TB example, the retrieval-trap warnings, and a blank per-tier mapping worksheet.

Call to action

• Talk to a surveillance engineer — book a 30-minute scoping call to talk through your storage tiers plan.
• See our case studies — 250+ shipped projects across video streaming, WebRTC, OTT, telemedicine, e-learning, surveillance, and AR/VR.
• Download the Storage-Tiering Decision Guide — One-page printable guide: the four storage tiers at a glance (access speed, how often read, on-prem media, cloud class, relative $/TB), the cost-per-terabyte table, the lifecycle-policy recipe (transition + expiration ages), the worked….

References

1. Understanding and managing Amazon S3 storage classes, Amazon Web Services (S3 User Guide). The storage-class comparison table read directly from the documentation: durability, availability, minimum storage durations (Standard-IA/One Zone-IA 30 days, Glacier Instant Retrieval 90 days, Glacier Flexible Retrieval 90 days, Deep Archive 180 days), retrieval characteristics (Standard/Standard-IA/Glacier IR millisecond access; Glacier Flexible minutes-to-hours; Deep Archive hours), and the design intent of each class. Tier 3 (first-party engineering). https://docs.aws.amazon.com/AmazonS3/latest/userguide/storage-class-intro.html (accessed 2026-06-09)
2. Amazon S3 Pricing, Amazon Web Services. List storage prices per GB-month used for the cost-per-TB figures: S3 Standard ~$0.023/GB ($23/TB-mo), Standard-IA ~$0.0125/GB ($12.50/TB-mo), Glacier Instant Retrieval ~$0.004/GB ($4/TB-mo), Glacier Flexible Retrieval ~$0.0036/GB ($3.60/TB-mo), Glacier Deep Archive ~$0.00099/GB ($0.99/TB-mo); plus per-GB retrieval fees on the IA/Glacier classes. US East (N. Virginia), accessed 2026-06-09. Tier 3 (first-party engineering). https://aws.amazon.com/s3/pricing/ (accessed 2026-06-09)
3. Understanding S3 Glacier storage classes for long-term data storage, Amazon Web Services (S3 User Guide). Retrieval-time detail for the archival classes: Glacier Flexible expedited 1–5 minutes / standard 3–5 hours / bulk 5–12 hours; Deep Archive standard within 12 hours / bulk within 48 hours — the basis for the cold/archive "minutes to hours" framing. Tier 3 (first-party engineering). https://docs.aws.amazon.com/AmazonS3/latest/userguide/glacier-storage-classes.html (accessed 2026-06-09)
4. Managing the lifecycle of objects (S3 Lifecycle), Amazon Web Services (S3 User Guide). How lifecycle transition and expiration rules automatically move objects to cheaper storage classes by age and delete them at end of life — the mechanism behind information-lifecycle management for surveillance footage. Tier 3 (first-party engineering). https://docs.aws.amazon.com/AmazonS3/latest/userguide/object-lifecycle-mgmt.html (accessed 2026-06-09)
5. ONVIF Profile G (recording and retrieval), ONVIF. The open standard for recording control and for searching and replaying stored footage (FindRecordings, FindEvents) across vendors — the retrieval interface the VMS uses regardless of which tier the bytes sit on; current Profile G Specification v1.1 (October 2025). Tier 1. https://www.onvif.org/profiles/profile-g/ (accessed 2026-06-09)
6. GDPR — Regulation (EU) 2016/679, Article 5(1)(e) (storage limitation), European Union (EUR-Lex). Personal data, including recognizable video, must be kept "no longer than is necessary" — the privacy-law maximum that the lifecycle expiration rule must enforce on every tier, archive included. Tier 1. https://eur-lex.europa.eu/eli/reg/2016/679/oj (accessed 2026-06-09)
7. EDPB Guidelines 3/2019 on processing of personal data through video devices, European Data Protection Board. Footage should in most cases be erased, ideally automatically, after a few days; longer retention requires documented justification — the norm against which an archive tier's long retention must be justified. Tier 1. https://www.edpb.europa.eu/sites/default/files/files/file1/edpb_guidelines_201903_video_devices.pdf (accessed 2026-06-09)
8. EN IEC 62676-4:2025: Video surveillance systems for use in security applications — Part 4: Application guidelines, IEC / CENELEC. Practical guidance for planning surveillance recording and storage, including retention and storage sizing; current 2025 edition — the system-level framework a serious deployment uses. Tier 1. https://webstore.iec.ch/publication/68479 (accessed 2026-06-09)
9. LTO-9 / Ultrium 9 technology, LTO Program (lto.org). LTO-9 cartridge native capacity of 18 TB (45 TB compressed) — the basis for the on-premises archive-tier cost per terabyte (~$5/TB at typical cartridge prices) and the power-free, offline nature of tape archive. Tier 3 (first-party/standards-body). https://www.lto.org/lto-9/ (accessed 2026-06-09)
10. The True Cost of Video Surveillance Archiving, SecurityInfoWatch. Institutional confirmation that long-term surveillance archiving is dominated by the cold/archive storage problem and the retrieval-cost trade, and that most recorded footage is rarely accessed — the value-curve argument for tiering. Tier 5 (institutional/analyst). https://www.securityinfowatch.com/video-surveillance/video-surveillance-storage/article/21162658/the-true-cost-of-video-surveillance-archiving (accessed 2026-06-09)
11. The Cost Per Gigabyte of Hard Drives Over Time / surveillance drive pricing (2026), Backblaze / industry price trackers. 2026 per-terabyte purchase prices used for the on-prem comparison: ~$60–80/TB NVMe SSD (hot), ~$18/TB surveillance HDD (warm), and the 2026 HDD price surge (~46% over four months) plus the enterprise-SSD spike that widened the fast-vs-bulk gap. Tier 5 (institutional/analyst). https://www.backblaze.com/blog/hard-drive-cost-per-gigabyte/ (accessed 2026-06-09)

Where sources disagreed, the official standard or first-party documentation was followed. Popular cloud-cost articles quote a single "storage is cheap" figure and ignore retrieval fees and minimum-duration charges; the controlling reality, read directly from the AWS storage-class documentation, is that the cheap tiers carry per-GB retrieval fees and 30/90/180-day minimum storage durations — so the article frames the cheap tiers as cheap-to-store-dear-to-read and shows where that bites. On the standards boundary, listicles imply a VMS "just stores video"; the ONVIF Profile G specification establishes that ONVIF standardizes the recording and retrieval interface, not the physical tier or medium — so tiering is framed as a storage-layer concern beneath the VMS.