What LTO TapeAlert can tell you about a failing tape, and what it cannot
Published 2026-10-10
TapeAlert, the 64-flag status page that LTO drives return through SCSI LOG SENSE page 2Eh, can say that an unrecoverable error happened and can point at the cartridge (flag 4, Media) or leave the question open (flags 5 and 6, where IBM's text says the cause is uncertain between tape and drive), but it cannot settle which of the two it was: IBM's action text prescribes a swap test with a known-good cartridge and drive. It also cannot report damage in data that has not been read, because the LTO consortium paper notes that uncorrectable errors stay latent until the data is read. The error budget behind the flags is small: the specified end-of-life uncorrectable bit error rate is 1 in 10^19 user bits for LTO-8 and 1 in 10^20 for LTO-9, which the paper converts to one event per 12.5 exabytes read for LTO-9, against 1 in 10^15 for the HDD reference it uses. These are calculated specifications, not field failure rates, and the same flag number means different things on a drive and on a library (flag 4 is Media on an LTO drive and Hardware Fault on a StorageTek SL150 library).
- Interface
- SCSI LOG SENSE, page code 2Eh; IBM states up to 64 flags (3592 page); Oracle shows a 1-byte flag per parameter and PC = 0 or 1.
- Error budget
- 1 in 10^19 user bits (LTO-8) and 1 in 10^20 (LTO-9) specified end-of-life uncorrectable bit error rate.
- Unit of loss
- One uncorrectable event loses one data set: about 5 MB on LTO-8 and 9.8 MB on LTO-9 (LTO paper).
- Not covered
- Field failure rates, other vendors' flag text, and tape formats other than LTO (the IBM 3592 page is used only as a cross-check).
Flags that bear on media and drive faults
| Flag number (decimal, hex) | Name | What the IBM LTO text says sets it | T10 type | Clears (T10 proposal) | Cross-check (second source) | Source |
|---|---|---|---|---|---|---|
| 1 (01h) | Read warning | Problem reading; no data lost, performance reduced; isolate by testing a known-good tape in the drive and the tape in a known-good drive | O | Start of next medium load | IBM 3592 page: same meaning, same isolation advice | IBM LTO TapeAlert list |
| 3 (03h) | Hard error | Any unrecoverable read, write or positioning error; set together with flag 4, 5 or 6 | M | Start of next medium load | IBM 3592 page: cleared when the cartridge is removed; set with 4, 5 or 6 | IBM LTO TapeAlert list |
| 4 (04h) | Media | Unrecoverable error due to a faulty cartridge; action: replace the cartridge | M | Start of next medium load | IBM 3592 page: same; copy needed data, then discard media | IBM LTO TapeAlert list |
| 5 (05h) | Read failure | Unrecoverable read error where isolation is uncertain (tape or drive); cartridge is defective only if flag 4 is also set | M | Start of next medium load | IBM 3592 page: same 'isolation is uncertain' wording | IBM LTO TapeAlert list |
| 6 (06h) | Write failure | Unrecoverable write or positioning error, isolation uncertain; cartridge is defective only if flag 4 is also set | M | Start of next medium load | IBM 3592 page: same wording | IBM LTO TapeAlert list |
| 7 (07h) | Media life | Cartridge reached end of life; copy data and discard | O | Start of next medium load | IBM 3592 page: 'end of its calculated, useful life' | IBM LTO TapeAlert list |
| 13 (0Dh) | Recoverable mechanical cartridge failure | Broken tape detected during midtape recovery; cartridge can still be unloaded | O | Start of next medium load | T10 proposal lists the flag; the 3592 page has no flag 13 | IBM LTO TapeAlert list |
| 14 (0Eh) | Unrecoverable mechanical cartridge failure | Cartridge cannot be unloaded, normally because the tape split apart | O | After service resolution | IBM 3592 page: 'snapped tape', operator cannot remove it | IBM LTO TapeAlert list |
| 20 (14h) | Clean now | Drive detects it needs cleaning | O | After successful cleaning or cause resolved | IBM 3592 page: set on a performance problem, cleared when cleaned | IBM LTO TapeAlert list |
| 31 (1Fh) | Hardware B | Drive failed its power-on self-test | M | At power-on event | IBM 3592 page: not cleared until the drive is powered off | IBM LTO TapeAlert list |
The error budget behind the flags
| Quantity | Unit | LTO-8 | LTO-9 | Cross-check (second source) | Source |
|---|---|---|---|---|---|
| Specified end-of-life UBER | uncorrectable events per user bits read | 1 in 10^19 | 1 in 10^20 | HPE data sheet: LTO-9 1 in 10^20 | LTO Program paper |
| User data read per expected event | exabytes (EB) | 1.25 (computed from 10^19 bits) | 12.5 | HPE data sheet: 12,500,000 TB for LTO-9 | LTO Program paper |
| Full native cartridges read per expected event | cartridges | 104,167 | 694,444 | computed: 1.25 EB / 12 TB and 12.5 EB / 18 TB agree with these counts | Quantum LTO-9 brief |
| C2 outer code (n, k, d) | bytes per codeword | (96, 84, 13) | (192, 168, 25) | LTO paper prints the same two codes | Quantum LTO-9 brief |
| Data set lost per uncorrectable event | megabytes (MB) | 5.031 | 9.805 | LTO paper: about 5 MB and 9.8 MB | Quantum LTO-9 brief |
| UBER with one of 32 read channels dead | events per user bits | 1 in 10^16 (at C2 input error rate 10^-3) | 1 in 10^20 (at 4x10^-3) | LTO paper: same values | Quantum LTO-9 brief |
| UBER with two of 32 read channels dead | events per user bits | 1 in 10^10 (at 10^-3) | 1 in 10^19 (at 10^-3) | LTO paper: same values | Quantum LTO-9 brief |
| HDD reference UBER used in the comparison | events per user bits | 1 in 10^15 (HDD, shown once) | same HDD reference | HPE data sheet: 1 in 10^15 for SATA HDD, 125 TB written before the first error | LTO Program paper |
What each signal sees and misses
| Failure point | Flag number(s) (decimal) | Not visible in the flag | Cross-check (second source) | Source |
|---|---|---|---|---|
| Unrecoverable read or write on the tape | 3, 4, 5, 6 | Whether the drive or the cartridge is at fault when only 5 or 6 is set: IBM says to swap cartridge and drive to isolate | T10 proposal marks 3, 4, 5, 6 mandatory and names flag 4 Media; IBM 3592 page gives the same isolation advice | IBM LTO TapeAlert list |
| Cartridge ageing | 7 (end of life), 19 (nearing end of life) | The flags report a calculated life, not a measured defect; the T10 proposal attaches a percentage-of-limit descriptor to them | IBM 3592 page: 'calculated, useful life' | T10 06-138r4 |
| Debris on the read head | 20 (clean now), 21 (clean periodic) | The LTO paper treats dead or clogged read channels as a real case; the flags do not report how many of the 32 channels are affected | LTO paper and Quantum brief both model one and two dead channels | IBM 3592 TapeAlert page |
| Broken or snapped tape | 13, 14 | Flag 13 does not appear on the 3592 page, so support for it is vendor-dependent | IBM LTO page and IBM 3592 page compared | IBM LTO TapeAlert list |
| Latent damage in data that has not been read | none | An uncorrectable error stays undetected until the data is read | LTO paper makes the statement; Quantum brief says its UBER model is theoretical | LTO Program paper |
| Same flag number on a library | 4: Media on a drive, Hardware Fault on an SL150 library | The number alone does not identify the device class: libraries use their own numbering | IBM LTO list (drive numbering) against Oracle SL150 list (library numbering) | Oracle SL150 manual |
| Flag lifetime after a read | SL150 library flags: cleared when the page is read; drive flags: until next load | Reading the library page clears its flags, so a second poll can show nothing | T10 proposal and IBM 3592 page for drive clearing | Oracle SL150 manual |
Reading the numbers
Start with the flags that are set together for a media-related loss. Flag 3 (Hard error) is mandatory in the T10 proposal and IBM says it is set with flag 4, 5 or 6. Flag 4 (Media) is the only one of those that points at the cartridge. If only 5 (Read failure) or 6 (Write failure) is set, IBM's own text says the cause is uncertain, and the action it gives is a swap test: the suspect tape in a known-good drive, and a known-good tape in the suspect drive. A flag cannot replace that test.
The numbers behind the flags show how little data a healthy tape is expected to lose. At the specified end-of-life rate of 1 in 10^20 user bits, LTO-9 is expected to read 12.5 EB per uncorrectable event, which is 694,444 full 18 TB cartridges; LTO-8 at 1 in 10^19 gives 104,167 full 12 TB cartridges. Because the paper, the brief and the data sheet are written by companies that build and sell LTO, and because the 10^20 figure is calculated assuming random errors, treat these as design targets. The brief says that checking 10^20 experimentally would need tens of thousands of drives for a year.
The size of a single loss is larger for tape than for a disk sector. One uncorrectable event removes a whole data set, 5.031 MB on LTO-8 and 9.805 MB on LTO-9, where an HDD sector in the paper's comparison holds 4,096 bytes. The paper states that the net result is two orders of magnitude more 'nines' for LTO-9 than for the HDD reference, so the lower event rate outweighs the bigger loss per event in that model.
A correlated fault is where the model is weakest and the sources are most specific. With one of 32 read channels dead, the paper and the brief give LTO-8 at 1 in 10^16 and LTO-9 at 1 in 10^20; with two dead channels, 1 in 10^10 against 1 in 10^19. No TapeAlert flag in the opened lists reports the number of dead channels: the cleaning flags 20 and 21 are the closest, and they signal a need for cleaning, not a count.
Flag numbers are not portable across device classes. On an LTO drive, flag 4 is Media; on the SL150 library page, code 0004h is Hardware Fault, and the library clears its flags when the page is read. A parser that maps a number to a meaning has to know whether it is talking to a drive or a changer.
What this page does not contain is a timeline. A public postmortem of a tape-library incident, with log lines and a timeline, was not found in sources that could be opened for this page, so that part of the area remains open.
Related: the SMART failure-signal note,the NVMe health-log notethe SCSI sense and kernel-log noteand what fails most often on HDDs, SSDs and links.
Method
Everything on this page was read on 2026-10-10 from documents opened that day: IBM's list of TapeAlert flags for LTO drives (TS4500 documentation) and IBM's 3592 TapeAlert page, which agree on the meaning of the flags compared here (1 to 7, 14, 20 and 31; the 3592 page has no flag 13, names flag 14 'snapped tape' and gives flag 8 a servo-problem cause); the T10 document 06-138r4, a 2007 proposal for the SSC-3 standard that lists the 64 flag names, which flags are mandatory and when each flag clears; Oracle's SL150 library manual for the page layout and for the separate library flag numbering; the LTO Program paper on LTO-9 reliability; Quantum's LTO-9 technical brief; and the HPE LTO-9 data sheet. Each table row names a second source in the column Cross-check. Values are copied as printed; values marked 'computed' are derived on this page from printed values.
Limits
This page states no failure rate, detection rate or share of cartridges that fail: none of the opened sources gives field data for tape (no population, no period, no count of failed cartridges), so 'frequent' is not claimed. The UBER figures are specifications calculated under an assumption of random, uncorrelated errors; the Quantum brief itself calls the analysis theoretical because verifying 1 in 10^20 would need tens of thousands of drives running for a year. The HPE data sheet, the LTO paper and the Quantum brief are written by vendors that sell or build the drives, and the paper and the brief share an author (Quantum). The T10 document is a proposal revision, not the text of the published standard, so the mandatory/optional column shows what that revision proposed. Flag wording was read for IBM LTO and IBM 3592 drives and one Oracle library; other vendors can word or support flags differently, and optional flags may be absent. No postmortem of a real tape-library incident with a published timeline was found and opened for this page, so none is analysed here. The page covers reading status only: nothing here concerns encrypted cartridges, recovery from damaged tape or anyone else's media.
Sources
- 01IBM Docs, TS4500 tape library 1.12.3: TapeAlert flags supported by LTO tape drives · accessed 2026-10-10
- 02IBM Docs, 3592 rackmount: TapeAlert error reporting · accessed 2026-10-10
- 03T10/06-138r4, SSC-3: TapeAlert Delineation (K. Butt, 12 June 2007; a change proposal, not the published standard) · accessed 2026-10-10
- 04Oracle StorageTek SL150 Modular Tape Library: Log Sense (4Dh), TapeAlert page 2Eh · accessed 2026-10-10
- 05Arslan, Lantz, Furrer, Spratt, Goker: LTO-9 Technology and User Data Reliability Analysis (The LTO Program, August 2022) · accessed 2026-10-10
- 06Quantum technical brief TB00068A: LTO-9: Raising the Bar Again for Performance, Capacity, and Data Reliability (2023) · accessed 2026-10-10
- 07HPE LTO-9 Ultrium 45TB RW Data Cartridge data sheet (mirror hosted at herontech.co) · accessed 2026-10-10