Design position

Why we don’t recommend LTFS.

LTFS — the Linear Tape File System — has been marketed heavily, promising the low per-gigabyte cost of tape with the plug-and-play convenience of a USB drive. It is a compelling pitch, and for some use cases, such as sharing large datasets between organizations, LTFS is an understandable choice. For production workflows and long-term archiving, we do not recommend it. This page explains why.

The premise:
tape is not a disk

Hard disks are random-access devices. A filesystem can read and update file metadata — names, paths, sizes, dates — anywhere on the disk at any time, which is why filesystems sit so naturally on them.

Tape is a sequential-access device. Data is written as a stream of blocks, one after another. LTFS makes tape emulate a random-access device by splitting each cartridge into two partitions that run side by side along the length of the tape: one for data, and one for a metadata index that is rewritten every time the tape is used. LTFS was specified in 2010 by the LTO Consortium — the drive manufacturers — on the promise of tape that behaves like a pen drive; a block-based format would have been simpler to specify and better suited to the medium, but it was not the format that was standardized.

The deeper mismatch is in what a filesystem is for. A filesystem manages a space that changes: files are created, grow, shrink, and are deleted; the holes left by deleted data must be found and reused; free space must be tracked and fragmentation kept under control. All of that machinery exists to make mutable storage efficient.

An archive needs none of it. Archival data is written once and never deleted, so the workload is simply stacking objects in an orderly line, one after another — and sequential media is ideally suited to exactly that. LTFS carries the machinery of a mutable filesystem onto a workload that has no use for it.

That emulation is complex, and complexity in a storage format translates directly into risk. Our policy on this is simple: the simpler the storage format, the easier data recovery will be — under any circumstances.

TAR · ONE SEQUENTIAL STREAM file file file Headers travel with the files. Nothing already on tape is ever rewritten. LTFS · TWO PARTITIONS, SIDE BY SIDE ALONG THE TAPE index partition — rewritten on every mount/unmount cycle data partition Both partitions run the full length of the reel. Interrupt the mount/unmount cycle — eject early, lose power — and the volume can be left inconsistent.
Fig. 1 — The difference that drives everything on this page. A tar stream only ever appends; LTFS maintains mutable filesystem state on sequential media.
The alternative:
tar

TAR — the tape archive format — predates LTFS by decades and is available by default on every Linux and Unix-like system. Any tar writes files back-to-back on tape with simple header blocks between them: no filesystem, no mounting, no index to keep synchronized — nothing on the tape but the files themselves, appended one after another. Tar has accumulated several dialects over its long life, so for standardization we use the POSIX-specified pax form (POSIX 1003.1-2001) — one precisely defined variant that any modern tar reads.

AspectTARLTFS
Open specificationYes (POSIX)Yes (SNIA LTFS Format)
Format complexityLowHigh
Corruption riskLowHigh
Mount/unmount overheadNone~45 s each way measured; vendor docs allow minutes
Usable tape capacity~98%≈95% structural; ~88% in our measured use
Self-describing tapesAchievable (see below)Yes, natively
Partial file retrievalYesYes
Software needed to read a tapetar — a standard tool on every Linux and Unix-like systemVendor or third-party LTFS application, installed separately

Third-party corroboration: HPE's StoreOpen best-practices guide notes that large indexes lead to "an extended delay in mounting the file system" (≈1 kB of index per file); Fujitsu's LTFS guide allows about ten minutes for background processing at eject; Quantum's StorNext documentation advises against LTFS for the retrieval copy. On capacity, the index partition and its guard wraps consume about 5% structurally (≈75 GB on LTO-5), before index growth in the data partition; tools such as YoYotta additionally reserve ~100 GB of headroom.

Concern 1:
corruption risk

Every LTFS tape must be mounted before use and cleanly unmounted afterward, so the index partition can be read and updated. If a tape is removed, or the drive loses power, before unmounting completes, the volume can be left in an inconsistent state. Anyone who has pulled a pen drive out without ejecting it first, and found it corrupted afterwards, has met this failure. LTFS is marketed as tape that behaves like a pen drive — and in this respect it does.

This is not theoretical — it is in the vendors' own documentation. From the Quantum LTFS User Guide:

"Tape activity may continue for several minutes after un-mounting an LTFS volume… Attempting these actions before tape activity has finished may cause corruption of the tape format."

And on power loss:

"…if power is removed from the drive without unmounting… the volume will be left in an inconsistent state and all files added since the last dismount operation will be inaccessible. The ltfsck utility may be able to recover the file data… but the metadata (such as filename, access dates) will be lost."

Note the second half of that: recovery may restore your data, but stripped of its filenames and dates — which, for an archive, is often a marginal recovery at best.

Vendor mitigations — periodic index writes, write-on-close index updates — reduce the exposure window, but they add capacity and performance overhead, and they cannot remove the underlying fragility of maintaining a filesystem on sequential media.

TAR has essentially none of this risk. If power fails mid-write, you seek back to the last valid header block and rewrite from there. No recovery utility required.

Concern 2:
mount overhead

In our experience, mounting and unmounting an LTFS tape takes 30–60 seconds each way, and vendor documentation acknowledges it can take several minutes. Every tape load/unload cycle therefore costs roughly 90 seconds of drive time that TAR simply does not have.

For a single tape that may be inconsequential. At production scale it is not. In one representative month, our restore requests touched an average of 30 different tapes per day — roughly 45 minutes of drive time lost daily, every day, just to mounting and unmounting. Quantum's own StorNext documentation advises against configuring LTFS for the copy used for file retrieval.

Concern 3:
lost capacity

The LTFS index partition consumes tape capacity. Officially around 5%; in our experience it has reached as high as 20%, and we model it at an average of 12.5%. The gap between the official figure and the observed one has a mechanism: the index is not only kept in its own partition but is also written again into the data partition at syncs and at unmount. Through format 2.4 each of those writes was a complete copy of the index, so consumption grew with the number of files and the number of sessions — these are the versions behind our measurements. Format 2.5 (2019) added incremental indexes that record only changes, which reduces this growth on implementations that use them. TAR's metadata overhead, by contrast, is negligible — about 2% including a safety buffer.

Concern 4:
interoperability

LTFS is an open standard on paper. In practice, reading an LTFS tape usually means installing a vendor or third-party application, and we have hit real friction:

  • Vendor LTFS tools often check the manufacturer of your drive and refuse to work with another vendor's hardware. We have had to compile the open-source reference implementation from source just to read LTFS tapes written by a third-party archiving application.
  • The LTFS specification keeps evolving (currently 2.5, standardized as ISO/IEC 20919), and tapes written with newer versions may not be readable by older implementations — something end users often discover only when an import fails.

A TAR tape, by contrast, can be read on any Linux machine with no additional software. When the goal is guaranteed access to archived data decades from now, that matters.

Concern 5:
what it doesn't do

LTFS also omits the parts an archive actually needs. It records no per-file checksums, so it cannot tell you whether what came back is what went in. It provides no parity, so a damaged cartridge is a matter of luck.

Both can be added through a vendor's own metadata — but a standard LTFS reader will ignore it, and an archive that needs a particular vendor's software to interpret its integrity data has arrived back at the problem LTFS was introduced to solve.

To be fair:
what LTFS gets right
  • It is an open, documented standard (SNIA LTFS Format, ISO/IEC 20919), just as POSIX tar is.
  • Tapes are self-describing — the index on the tape fully catalogs its contents.
  • It is a good fit for data interchange between organizations, where the receiving party may have no relationship with your archiving system.

If your workflow involves routinely shipping tapes to outside parties, LTFS is a reasonable choice. Our workflow is fully in-house — tapes are written and read by a single system — so these advantages do not offset the risks for us.

Closing the gap:
self-describing tar

TAR's one genuine gap is that it is not natively self-describing. The gap is closable, and closing it is simpler than adopting a filesystem. For years our production archive did it the plain way: a full catalog of the tape's contents appended at the end — the logical equivalent of the LTFS index, without the mutable filesystem underneath. Even with the archive database lost, every tape could describe itself.

The REM formats are the evolution and formalization of that idea. Every REM-OBJECT carries a manifest recording each file's path, size, checksum, and position; REM-PARITY adds repeating bootstrap blocks along the cartridge, so a bare, unlabelled tape can be mapped, verified, and repaired from the tape alone. Tar's simplicity is kept; LTFS's one real advantage is matched.

LTFS's self-describing index does not eliminate the need for a database in practice — any archiving application still needs one for cross-tape search.

Our
recommendation

Tape is a sequential medium, and formats that naturally align with that fact are simpler, safer, and cheaper to operate.

  • Use TAR for primary and secondary copies. It is simple, open, universally readable, and in our production experience has caused zero format-related corruption.
  • If LTFS is required — by a specific application or an exchange workflow — use it as one copy within a multi-format strategy, never as your only format.
  • Never convert all copies to a single format at once. Bugs have shipped in LTFS implementations before. If every copy of your archive shares one format, a single format-level defect can affect your entire archive simultaneously.

Data integrity deserves the same discipline whatever the format: neither TAR nor LTFS provides built-in write-time integrity checking. Our practice is to calculate a checksum for every file, store it outside the tape, and verify it — including a full restore and re-checksum of every archive we write.