When a single file is requested from the yearly copy of seven years ago, the problem is often not the cartridge but that there is no longer a drive to read it. Daily restores are done from a fast disk-based backup repository; if monthly and yearly copies will be kept for years, the cost, durability and network isolation of the media become decisive. There are three candidates for this tier: LTO tape, on-premises S3-compatible object storage, and cloud object storage (including archive tiers). Below we compare the three on retrieval time, media lifetime, offline copy characteristics, operational load and hidden costs, and tie them to a decision tree.
Long-term retention requirements
Put the requirements in writing before choosing; otherwise the decision is made by the technology the team is used to. Questions that need answers:
- Retention period and volume: Which data, for how many years? Legal retention periods differ by data type; these periods are set together with the legal and compliance teams.
- Retrieval expectation: When a single file is requested from a yearly copy, within how many hours or days is delivery acceptable?
- Isolation: Does the copy need to be physically separated from the network (air-gapped), or logically immutable?
- Readability: With what hardware and software will you read this data ten years from now?
- Deletion obligation: How will personal data whose retention period has expired be securely destroyed?
The total volume of the long-term tier is derived from the retention policy. We explained how the number of monthly and yearly copies and capacity are calculated in a GFS setup in the backup storage capacity calculation article; here we focus on which media that volume will sit on.
LTO tape: capacity, generation compatibility, lifetime
LTO (Linear Tape-Open) is an open tape format supported by multiple vendors. Cartridge capacities and drive speeds increase with each generation:
| Generation | Native capacity | Native transfer rate | Read/write compatibility |
|---|---|---|---|
| LTO-7 | 6 TB | 300 MB/s | Writes LTO-6, reads LTO-5 |
| LTO-8 | 12 TB | 360 MB/s | Reads/writes LTO-7 only |
| LTO-9 | 18 TB | 400 MB/s | Reads/writes LTO-8 only |
The “compressed capacity” vendors quote is calculated assuming a 2.5:1 ratio. This ratio isn’t achieved with already compressed and encrypted backup files; plan with native capacity.
The generation compatibility rule is the most overlooked risk in long-term retention. Up to LTO-7, drives could read two generations back; LTO-8 and LTO-9 can read and write only the previous generation. So an archive written to LTO-7 cartridges today can’t be read with an LTO-9 drive. The team renewing the library usually discovers this when a restore from old cartridges is requested. For LTO-10, it has been announced that backward compatibility is removed entirely; since the rules can change from generation to generation, check the compatibility table on the LTO Program site and from the drive vendor before buying. The practical result: either an older-generation drive is kept in working order for the whole retention period, or the data is periodically migrated to the new generation.
Vendors state an archive life of decades under suitable temperature and humidity conditions. This value is for the cartridge; the drive, library, backup software and the software’s catalog database don’t live as long. In long-term retention, the real limiting factor is usually not the cartridge but the ecosystem that can read it. The cartridge lasts. The ecosystem ages.
Tape’s strengths: a cartridge physically removed from the network is a true offline copy; WORM cartridges (LTO-3 and later) can’t be changed after being written; drive-based AES-256 encryption (LTO-4 and later) is supported; media cost per TB is low for long-term retention. Its weaknesses: retrieving a single file requires finding and loading the cartridge and reading sequentially; if the drive can’t get a sufficient data stream, it constantly stops and starts (shoe-shining), speed drops and wear increases. That’s why writing to tape is mostly done from the disk backup repository, not from production (disk-to-disk-to-tape). Shoe-shining is a common sight in jobs that write directly from a network share to tape.
A note specific to LTO-9: when a new cartridge is loaded into the drive for the first time, media optimization (calibration) is performed and can take a long time per cartridge. When putting a large batch of cartridges into service, plan this time outside the backup window.
On-premises object storage
S3-compatible object storage is set up as an appliance or software-defined cluster in on-premises server and storage infrastructure. Backup software writes to this target with the S3 API. Features that stand out for long-term retention:
- Object Lock: Prevents objects from being deleted or overwritten until a certain date. In compliance mode, even the administrator account can’t remove the lock.
- Erasure coding: Provides data protection against disk and node failures; capacity grows more flexibly than with RAID.
- Instant access: There is no retrieval wait; restoring a single file from the archive happens at disk speed.
Limits: object storage is connected to the network and isn’t a physical offline copy. Immutability is provided by software and configuration; the appliance’s management interface, time source (NTP) and hardware lifecycle must be protected. If the clock can be shifted, the lock loosens too. If the hardware refresh cycle (usually a few years) is shorter than the retention period, also put migrating the data to the new system into the plan. A security comparison of immutable backup options is in the immutable backup options guide on our cybersecurity site.
Cloud object storage and archive tiers
Cloud providers offer storage classes priced differently by access frequency. In archive classes, storage cost per GB drops; in return, retrieval time grows and read, transaction and early deletion fees come into play. The main rules according to vendor documentation at the time of writing:
- Amazon S3 Glacier Flexible Retrieval: Retrieval takes minutes with the expedited option, a few hours with the standard option, and longer with the bulk option; the minimum storage duration is 90 days.
- Amazon S3 Glacier Deep Archive: Standard retrieval takes up to 12 hours and bulk retrieval up to 48 hours; the minimum storage duration is 180 days.
- Azure Blob Archive tier: The object is offline; to be read, it must first be rehydrated to the Hot, Cool or Cold tier. At standard priority this can take up to 15 hours. The early deletion period for Archive is 180 days.
For current values, check the Amazon S3 storage classes and Azure Blob access tiers pages.
Hidden costs are the real decision criterion for cloud archive: retrieval fees per GB, data egress fees, per-object transaction (PUT/GET) fees, and billing for the entire remaining period for data deleted before the minimum duration. Backup software that writes with a small block size produces many objects and inflates transaction fees. If the whole archive needs to be retrieved in a disaster, both the time and the bill come out higher than expected; do a worked calculation of this scenario before buying.
Comparison table
| Criterion | LTO tape | On-premises object | Cloud archive |
|---|---|---|---|
| Retrieval | Cartridge load + sequential read | Instant | Hours |
| Offline copy | Physical | No (logical lock) | No (logical lock) |
| Initial investment | Medium (drive/library) | High | Low |
| Variable cost | Cartridges, transport, storage site | Power, maintenance | Storage + retrieval + egress |
| Operational load | Rotation, labeling, transport | Hardware lifecycle | Cost and access monitoring |
| Long-term risk | Generation compatibility | Hardware refresh | Provider dependency |
Worked example: An organization archiving 100 TB of native data per year uses 100 ÷ 18 ≈ 5.6, i.e. 6 cartridges with LTO-9 without assuming compression; if a second copy is kept offsite, the number becomes 12. If the same volume is kept in a cloud archive, the storage fee stays low; but calculate the retrieval and egress fees for 100 TB separately for the full-retrieval scenario.
Tape rotation and offsite storage
Tape only works with disciplined rotation. A cartridge forgotten on a shelf doesn’t count as a backup. A workable setup:
- Media pools: Weekly, monthly and yearly copies are kept in separate pools; WORM cartridges can be preferred for the yearly pool. In software such as Veeam, the GFS media pool makes this separation automatically.
- Barcodes and inventory: Each cartridge’s barcode, contents, write date and retention end date are kept in the backup software’s catalog and in an independent inventory list.
- Moving offsite: Cartridges are moved in a locked transport case with a handover record (chain of custody). The offsite location is chosen far enough away not to be affected by a disaster affecting the main building.
- Environmental conditions: The storage area should be within the temperature and humidity range specified by the vendor, away from sources of magnetic fields.
- Read test: A sample cartridge is periodically brought back from offsite and a test restore is done; importing the cartridge without the catalog is also tried for the catalog-loss scenario.
- Destruction: Cartridges whose retention period has expired are destroyed physically or by degaussing, with a destruction record.
The place of the tape server and library in the Veeam infrastructure, together with proxy and repository placement, is in the Veeam proxy and repository design article.
Decision tree for the choice
- Is a physical offline copy mandatory? If yes, tape is the primary candidate. If not, immutable object storage may be enough.
- Are restores from the archive frequent? If several times a month or more, on-premises object storage or the cloud’s instant-access classes; if rare, tape or a cloud archive tier.
- Are the total volume and retention period large? With tens of TB and multi-year retention, tape’s media cost advantage becomes clear; at small volumes, the drive and operational load erase that advantage.
- Are there staff and an offsite location to sustain tape operations? If not, cloud archive provides a more realistic offline-like copy.
- Is the cost of the bulk retrieval scenario acceptable? If not, cloud archive becomes the secondary copy and the primary archive is kept on premises.
In most environments, the result is not a single technology but a combination: short- and medium-term copies on immutable disk or object storage, yearly copies on tape or in a cloud archive. Check whether this combination matches the 3-2-1-1-0 rule. For choosing a tape library and object storage hardware, see our hardware consulting page; for the tiered backup architecture as a whole, our backup and disaster recovery page.
Frequently asked questions
Can an LTO-9 drive read LTO-7 cartridges?
No. LTO-9 drives can read and write only LTO-9 and LTO-8 cartridges. To read LTO-7 archives, a suitable LTO-8 or older drive must be kept, or the data must be moved to the new generation.
Does a cloud archive tier replace tape?
Partly. Cloud archive provides a remote, immutable copy without management burden, but it isn’t physically disconnected from the network; retrieval and data egress are also charged. If a physical offline copy is required, tape is still needed.
Can compressed capacity be used when planning tape capacity?
Don’t. Since backup software already compresses and encrypts the data, the drive’s hardware compression brings no extra gain. Plan on native capacity.
What is the early deletion fee in cloud archive?
In archive classes, objects are priced to be stored for a minimum period. For objects deleted or moved to another tier before this period ends, storage charges for the remaining period are still billed.