Tool AT-W19REV. 2026-A
RAID calculator for NAS: how much space is left, and what does the backup need?
In RAID 5, usable capacity is the sum of all drives minus one; in RAID 6, minus two; and in RAID 1, only one drive. Four 4 TB drives therefore give 12 TB in RAID 5; Windows and the NAS show about 10.9 TB of that, because they calculate in binary. The backup needs a target of its own, because RAID is not a backup. The RAID calculator gives you both, plus fault tolerance, rebuild time and transfer time.
- free
- no sign-up
- runs in your browser
01 / CALCULATOR
Enter your drives, read off capacity and backup needs
Calculator · RAID capacity and backup needs
The calculator gives the capacity after RAID, before the file system and system take their share. SHR and RAIDZ are approximations; what your NAS shows after setup is what counts.
Runs in your browser. Nothing is sent or stored.
02 / RULE OF THUMB
RAID is not a backup
The sentence we hear most often during NAS checks: “The data is stored twice; it has RAID.” A RAID spreads the data across several drives so that the device survives the failure of one drive. It works in real time. That is exactly the catch, because every change lands on all drives immediately, including the wrong one.
- Deleted or overwritten by mistake. The file is gone from all drives at once. It comes back only from an older version in the backup.
- Ransomware. It encrypts whatever it can reach over the network, and the RAID dutifully writes the encrypted files to every drive.
- Silent data errors. In storage systems with a combined 1.53 million drives, researchers found over 400,000 cases where data was no longer correct without any error message Bairavasundaram et al. 2008. A mirror simply mirrors such an error too.
- The device itself. In an analysis of about 39,000 storage systems, only 20 to 55 per cent of failures were caused by drives, the rest by connections and controllers Jiang et al. 2008. Lightning, water and theft hit all drives at once anyway.
- The second failure. Drives from the same delivery age together, and failures come in clusters Schroeder and Gibson 2007. If another drive fails during the rebuild, a RAID 5 is lost.
A RAID therefore spares you the downtime when a drive dies. The backup protects against loss: a second device, several versions and a copy kept off-site, as the 3-2-1 rule describes. And it only counts once a restore has been rehearsed; how to do that is explained in the guide to the restore test. How quickly your business as a whole is working again after a failure is shown by the IT emergency check.
03 / RAID TYPES
RAID 0, 1, 5, 6, 10, SHR and RAIDZ compared
The term RAID comes from a paper of 1988: many inexpensive drives combined instead of one large one Patterson et al. 1988. All levels rest on two techniques Chen et al. 1994. Striping makes things faster, redundancy protects against failures. Redundancy means either a mirror, where each drive carries everything, or parity: a checksum from which a missing drive can be recalculated. Mirrors cost more space, parity costs more time during the rebuild.
| Type | min. | Usable | 4 × 4 TB | May fail | Use |
|---|---|---|---|---|---|
| Basic / JBOD | 1 | sum of all drives | 16 TB | 0 | A single drive, or drives chained one after another. If one fails, data is lost. |
| RAID 0 | 2 | sum of all drives (some controllers: number × smallest) | 16 TB | 0 | Speed only. If one drive fails, everything is gone. |
| RAID 1 | 2 | smallest drive | 4 TB | 3 | Two drives in a small NAS. Each carries all the data. |
| RAID 5 | 3 | (number − 1) × smallest | 12 TB | 1 | Three to four smaller drives. Plenty of space, but no reserve during the rebuild. |
| RAID 6 | 4 | (number − 2) × smallest | 8 TB | 2 | The rule from five drives, or with large drives. |
| RAID 10 | 4 | (number ÷ 2) × smallest, even number | 8 TB | 1 to 2 | Databases and virtual machines: fast, short rebuild, half the space. |
| SHR-1 | 1 | with mixed sizes roughly sum − largest drive | approx. 12 TB | 1 | Synology with drives of different sizes, protection like RAID 5. |
| SHR-2 | 4 | layers with at least four drives, per layer − 2 | approx. 8 TB | 2 | Synology with drives of different sizes, protection like RAID 6. |
| RAIDZ1 | 2 | approx. (number − 1) × smallest | approx. 12 TB | 1 | TrueNAS, Proxmox and other ZFS systems. Checksums detect silent errors. |
| RAIDZ2 | 3 | approx. (number − 2) × smallest | approx. 8 TB | 2 | ZFS from five drives, or with large drives. |
| RAIDZ3 | 4 | approx. (number − 3) × smallest | approx. 4 TB | 3 | Large ZFS groups where even two failures during the rebuild should be survivable. |
RAID 1 with four drives is a four-way mirror; two drives are usual. Capacity rules according to the Synology documentation and OpenZFS, as of 13 November 2026.
Fig. 1 · Data (A1, B1 …), parity (P, Q, hatched) and mirror copy (dashed) on four drives.
Mixed drive sizes: Synology SHR
Classic RAID uses only as much of each drive as the smallest one has. With 2, 4, 4 and 8 TB in the device, RAID 5 turns that into 6 TB of usable space, and 10 TB lie idle. This happens more often than you might think: over the years a NAS is upgraded with whatever was cheap at the time.
Synology Hybrid RAID (SHR) therefore divides the drives into layers. The bottom layer spans all drives as high as the smallest one reaches, the next spans the remaining ones, and so on. Each layer with at least two drives gets a parity. With the same drives, SHR-1 reaches 10 TB, with the same protection as RAID 5. With mixed sizes, the rough rule is: sum of all drives minus the largest.
SHR-2 has two parities and uses only layers with at least four drives. In this example that leaves only 4 TB. Synology itself describes the capacity as “optimised by the system” and gives no formula; we calculate with the layer model and have checked the results against the Synology RAID calculator. It remains an approximation; what counts is Storage Manager after setup.
ZFS: RAIDZ1 to RAIDZ3
TrueNAS, Proxmox and other systems with ZFS combine drives into RAIDZ groups with one, two or three parities. OpenZFS expressly gives the capacity only approximately, as (number − parity) × drive size. How much is actually usable depends on the block size: three drives in RAIDZ1 with very small blocks use only half, so no more than a mirror. In addition, ZFS keeps one thirty-second of the pool free, about 3 %. 3 to 9 drives per group are recommended.
The advantage of ZFS is checksums over all data. In tests with deliberately injected faults, ZFS caught a broad range of drive errors, but was much less well protected against errors in main memory Zhang et al. 2010. A ZFS server therefore belongs on hardware with ECC memory, and it too needs a backup.
04 / TB AND TIB
Why Windows shows less than is printed on the drive
Drive manufacturers calculate in decimal: 1 TB is 10¹² bytes, so a trillion. Windows and most NAS devices calculate in binary with 2⁴⁰ bytes per unit, which is just over 1.0995 trillion. Correctly, this unit would be called TiB (tebibyte), but “TB” is often still what is displayed. A 4 TB drive therefore appears as 3.64 TB. Nothing is missing; the difference of about 9 per cent is purely a matter of conversion.
From the raw value after RAID, real deductions then follow. Synology reserves about 10 GB on each drive for the system and swap, a Btrfs volume holds back 4 % for metadata, an ext4 volume 2 %. The calculator therefore gives the value after RAID in TB and in TiB; what Storage Manager finally shows is a few per cent lower still.
| On the box | Shown in Windows / NAS |
|---|---|
| 1 TB | 0.91 TiB |
| 2 TB | 1.82 TiB |
| 4 TB | 3.64 TiB |
| 8 TB | 7.28 TiB |
| 12 TB | 10.9 TiB |
| 16 TB | 14.6 TiB |
| 20 TB | 18.2 TiB |
| 24 TB | 21.8 TiB |
05 / FAILURE
When a drive fails: rebuild and read errors
The datasheet figure of a mean time to failure of 1 to 1.5 million hours corresponds to an annual failure rate of at most 0.88 per cent. In the field, usually 2 to 4 per cent of drives were replaced each year, in individual installations up to 13 per cent Schroeder and Gibson 2007. With 2 % to 4 % per drive, with four drives the probability that at least one fails within a year is 7.8 % to 15 %. Over five years it is 33 % to 56 %; you should therefore expect a rebuild.
During a rebuild, the replacement drive is completely rewritten. We assume 50 to 150 MB/s for a NAS that is in use; this is based on experience, and the peak rate of a large NAS drive is around 285 MB/s at the outer edge according to the datasheet. While the rebuild is running, a RAID 5 has no reserve.
| Drive | at 150 MB/s | at 50 MB/s |
|---|---|---|
| 4 TB | 7.4 hours | 22 hours |
| 8 TB | 15 hours | 44 hours |
| 16 TB | 30 hours | 3.7 days |
| 24 TB | 44 hours | 5.6 days |
The read error risk, calculated and put in context
Datasheets state a maximum uncorrectable read error rate: for desktop and many NAS drives, fewer than 1 per 10¹⁴ bits read; for Pro and enterprise drives, 1 per 10¹⁵. 10¹⁴ bits are 12.5 TB. If the rebuild of a RAID 5 reads all the remaining drives, the probability of at least one error by this figure is 1 − e^(−gelesene Bit ÷ 10^14). A single sector error in this phase is enough for data loss Schroeder et al. 2010.
| Drives | read | at 1 per 10¹⁴ | at 1 per 10¹⁵ |
|---|---|---|---|
| 3 × 4 TB | 8 TB | 47 % | 6,2 % |
| 4 × 4 TB | 12 TB | 62 % | 9,2 % |
| 4 × 8 TB | 24 TB | 85 % | 17 % |
| 4 × 12 TB | 36 TB | 94 % | 25 % |
| 6 × 16 TB | 80 TB | 100 % | 47 % |
These figures are an upper limit, not a prediction. The datasheet gives a maximum value, not a measured one. In the largest field study on the subject, only 3.45 per cent of 1.53 million drives developed such a sector error at all within 32 months Bairavasundaram et al. 2007. But the errors cluster: on drives that already have some, and on larger drives. For SSDs the figure is of no use at all, because read errors there were not related to the amount read Schroeder et al. 2016.
The conclusion remains the same, though. RAID 5 can fail during the rebuild, and it was precisely this that gave rise to double-parity schemes Thomasian and Blaum 2009. Models based on field data predict 2 to 1500 times as many double failures as the usual “mean time to data loss” formula Elerath and Pecht 2009. That is why the calculator does not work out this formula at all, and why it recommends RAID 6 or SHR-2 from 8 TB per drive. There, a read error during the rebuild is corrected using the second parity.
Which RAID type for how many drives?
| Drives | Usual choice | Reason |
|---|---|---|
| 1 | Basic / JBOD, SHR-1 | A single drive has no fault protection. Then only the backup to a second device counts. |
| 2 | RAID 1, SHR-1 | Two drives: RAID 1, or SHR-1 on Synology. Each drive carries all the data. |
| 3 | RAID 5, SHR-1, RAIDZ1 | Three drives: RAID 5, SHR-1 or RAIDZ1. |
| 4, under 8 TB | RAID 5, SHR-1, RAIDZ1, RAID 6, SHR-2, RAIDZ2 | Four smaller drives: RAID 5 or SHR-1 for more space, RAID 6 or SHR-2 for a reserve during the rebuild. |
| 4, 8 TB or more | RAID 6, SHR-2, RAIDZ2, RAID 10 | Four drives of 8 TB or more: RAID 6, SHR-2 or RAIDZ2. Same space as RAID 10, but any two drives may fail. |
| 5 to 8 | RAID 6, SHR-2, RAIDZ2 | From five drives: RAID 6, SHR-2 or RAIDZ2. |
| 9 or more | RAID 6, SHR-2, RAIDZ2, RAIDZ3 | From nine drives: RAID 6 or SHR-2. OpenZFS recommends 3 to 9 drives per RAIDZ group; above that, two groups or RAIDZ3. |
Two habits reduce the risk regardless of RAID type. First, regular data scrubbing, which reads all sectors while redundancy is still there: over 60 per cent of the sector errors in the field study were found this way Bairavasundaram et al. 2007, and with RAID 6 too, safety depends on it Elerath and Schindler 2014. Second, a look at the SMART values. They do not reliably predict the failure of an individual drive Pinheiro et al. 2007, but a rising number of reallocated sectors is a strong warning sign Ma et al. 2015. We replace such a drive before it fails. If it has already happened and the NAS no longer responds, the guide Hard drive has failed helps with the first steps.
06 / BACKUP
How much storage the backup needs
A backup is more than a copy. It keeps older states so that you can go back to the day before the file was overwritten or encrypted. Common programs back up everything on the first run and afterwards only what has changed. Retention is staggered, following the grandfather-father-son principle: many daily states for the last few weeks, plus weekly and monthly ones for the time before that.
The calculator adds the changes between all retained states to the full backup. With the preset values, 2 TB of data and 1 per cent change per day with 14 daily, 4 weekly and 6 monthly versions, that comes to 21 states over 150 days and at most 5 TB per backup target. So that the target is at most 70 % full and leaves room for growth, it should hold 7.14 TB, so an 8 TB drive. Under the 3-2-1 rule there are two such targets, one of them off-site.
The result is an upper limit. If the same file changes every day, the calculator still counts it anew every day. Compression and deduplication can save a lot, but how much depends strongly on method and data Meyer and Bolosky 2012. That is why no saving is preset; you can choose one if you know your data. In the long run, what counts is less the last percentage point than the independence of the copies from each other and that errors are noticed early Baker et al. 2006.
Initial backup and restore: the underestimated time
| Connection | calculated with | 1 TB | 4 TB |
|---|---|---|---|
| USB 3 to external hard drive | 150 MB/s | 1.9 hours | 7.4 hours |
| Network 1 Gbit/s | 100 MB/s | 2.8 hours | 11 hours |
| Network 2.5 Gbit/s | 250 MB/s | 1.1 hours | 4.4 hours |
| Network 10 Gbit/s | 500 MB/s | 34 minutes | 2.2 hours |
| Upload 40 Mbit/s to the cloud | 4 MB/s | 2.9 days | 12 days |
| Download 100 Mbit/s from the cloud | 10 MB/s | 28 hours | 4.6 days |
The table shows why a cloud-only backup often falls short with typical connections. The first transfer of 2 TB takes 5.8 days at 40 Mbit/s upload; the daily change afterwards is no problem. It becomes inconvenient in an emergency: bringing everything back takes 2.3 days at 100 Mbit/s, and your business stands still for that long. A second copy on site, for example on a second NAS or a USB drive, brings the same data back over the network at 1 Gbit/s in 5.6 hours. How fast your connection really is, the speed test measures.
07 / METHOD
How to plan a NAS and backup
- Establish the data volume. What is on the server, NAS and workstations today, and how much is added each year? That determines drive size and backup target.
- Choose the RAID type by number of drives. Two drives as a mirror, three to four smaller ones as RAID 5 or SHR-1, from four large drives RAID 6 or SHR-2.
- Mix drives, but deliberately. Equal sizes make the best use of the space. If you upgrade, use Synology SHR and replace the smallest drive first.
- Switch on monitoring. E-mail notification for drive errors and a monthly data scrub. A warning nobody reads does not protect anything.
- Back up to a second device. With versions, and one copy off-site or disconnected from the network. For this the NAS needs a UPS, so that a power cut does not strike in the middle of a backup.
- Rehearse the restore. Once a year restore one file and one whole folder, and time it.
Where the calculator stops
- Capacity comes before the file system. SHR and RAIDZ are approximations; what the NAS shows after setup is what counts.
- Risks as a range. Read errors by the datasheet are an upper limit, rebuild times a value based on experience. The tool deliberately does not calculate a probability of data loss.
- Backup as an upper limit. Files changed several times are counted several times. The logs of your backup software show the real change rate.
How we build, monitor and restore backups for businesses in an emergency is described under backup and disaster recovery. We source NAS devices and drives through the hardware ordering service and plan larger storage and servers under servers and virtualisation. If a drive has already failed and the data is missing, data recovery is the next step.
08 / RESEARCH
What the research says
The warnings in the calculator are not gut feelings. Six findings from journals and conferences support the most important decisions; the full list with review notes is further down.
- Drives fail more often than the datasheet says. Instead of at most 0.88 per cent, usually 2 to 4 per cent were replaced each year in the field, and failures were clustered in time Schroeder and Gibson 2007. The calculator therefore gives the probability that a drive fails within a year.
- Sector errors are rare, but dangerous at the wrong moment. 3.45 per cent of drives were affected within 32 months, more among large drives Bairavasundaram et al. 2007; a single one during a rebuild is enough for data loss Schroeder et al. 2010.
- The usual reliability formula underestimates double failures. Models close to the field come to 2 to 1500 times more Elerath and Pecht 2009. That is why the calculator advises double parity for large drives Thomasian and Blaum 2009.
- RAID 6 is not invulnerable either. Among a million drives, triple faults accounted for 80 per cent of RAID 6 failures, and reallocated sectors foreshadowed failures Ma et al. 2015.
- SSDs fail less often, but misread more often. Depending on the model, 20 to 63 per cent had at least one uncorrectable read error Schroeder et al. 2016. An SSD NAS also needs redundancy and a backup.
- Only checksums and old states help against silent errors. Over 400,000 cases of corrupted data without an error message Bairavasundaram et al. 2008 explain why versions in the backup are worth more than a second mirror.
09 / FOR AI ASSISTANTS
Using this tool from an AI assistant
Assistants such as Claude or ChatGPT that speak the Model Context Protocol (MCP) can call this page’s calculation directly instead of rebuilding it. The server is public, needs no sign-in and stores nothing. The answer always includes the link to this page, the date of the data and the limits of the result.
- Endpoint
https://aureum-tech.com/mcp- Tool
werkzeug_nas_raid_rechner- Transport
- Streamable HTTP, read-only
| Name | Required | Meaning |
|---|---|---|
platten | yes | Size of each drive in TB, as on the box (decimal), e.g. [4, 4, 4, 4] or mixed [8, 4, 4, 2]. |
raid | yes | RAID type: basic = Basic / JBOD; raid0 = RAID 0; raid1 = RAID 1; raid5 = RAID 5; raid6 = RAID 6; raid10 = RAID 10; shr1 = SHR-1; shr2 = SHR-2; raidz1 = RAIDZ1; raidz2 = RAIDZ2; raidz3 = RAIDZ3. |
daten_tb | no | Optional: amount of data to back up in TB. The backup need is calculated only with this value. |
aenderung_prozent_tag | no | Share of the data in per cent that changes per day. Default 1. |
versionen_taeglich | no | Retained daily versions. Default 14. |
versionen_woechentlich | no | Retained weekly versions. Default 4. |
versionen_monatlich | no | Retained monthly versions. Default 6. |
einsparfaktor | no | Assumption for compression and deduplication: 1 = no saving (cautious, default), 2 = half the space. |
kopien | no | Number of backup copies besides the original (3-2-1 rule: 2). Default 2. |
upload_mbit | no | Upload speed of the internet connection in Mbit/s, for the initial cloud backup. Default 40. |
download_mbit | no | Download speed of the internet connection in Mbit/s, for restoring from the cloud. Default 100. |
Questions where an assistant should call it
- “How much space do I have with four 8 TB drives in RAID 5, and what does Windows show of it?”
- “Is SHR worthwhile with drives of 2, 4, 4 and 8 TB?”
- “How big is the risk of a read error when rebuilding a RAID 5?”
- “How much storage does the backup of 3 TB with 30 daily versions need, and how long does the first upload take?”
How the connection works and what MCP can do in your own business is explained on the page about MCP integrations.
10 / COMMON QUESTIONS
Common questions about NAS, RAID and backup
How much storage is left with RAID 5?
With RAID 5, you get the sum of all drives minus one, calculated with the smallest drive. Four 4 TB drives give 12 TB, three drives 8 TB. Windows and most NAS devices show about 10.9 TB of that, because they calculate in binary. The file system and system partition take a few more per cent. One drive may fail; a second one must not fail during the rebuild.
Why does Windows show less storage than is printed on the drive?
Manufacturers calculate in decimal: 1 TB is 1,000,000,000,000 bytes. Windows calculates in binary, with 1,099,511,627,776 bytes per unit, but still writes TB. A 4 TB drive therefore appears as 3.64 TB, an 8 TB drive as 7.28 TB. Nothing is missing; the difference of about 9 per cent is purely a matter of conversion. The calculator gives both values.
Is a RAID a backup?
No. A RAID keeps the NAS running when a drive fails. But a file that is deleted, or encrypted by ransomware, is changed on all drives at once. It does not help against theft, lightning or a fault in the device either. For that you need a backup on a second device, with versions and a copy kept off-site.
RAID 5 or RAID 6 with four drives?
With drives under 8 TB, RAID 5 is acceptable and provides the capacity of one additional drive. From 8 TB, we advise RAID 6 or SHR-2: rebuilding an 8 TB drive takes 15 to 44 hours, and during that time RAID 5 has no reserve. According to the datasheet, the risk of a read error when rebuilding four 8 TB drives is 85 %; RAID 6 corrects it.
What does Synology SHR offer with drives of different sizes?
Classic RAID uses only as much of each drive as the smallest one has. With 2, 4, 4 and 8 TB, only 6 TB remain usable in RAID 5. SHR-1 divides the drives into layers and reaches 10 TB, with the same fault protection. SHR-2 needs four drives per layer and manages only 4 TB here. Synology describes the capacity as optimised by the system; our values agree with the Synology RAID calculator.
How long does a RAID rebuild take?
The replacement drive is completely rewritten. At 150 MB/s, that takes 7.4 hours for 4 TB and 30 hours for 16 TB. If the office keeps working, the rate can fall to 50 MB/s, and 16 TB takes 3.7 days. This range is based on experience; how fast your NAS is depends mainly on its processor and load.
How much storage does my backup need?
The first backup is as large as the data; after that only changes are added. With 2 TB of data, 1 per cent change per day and 21 retained versions over 150 days, the tool calculates at most 5 TB per backup target. With some headroom, the target should hold 7.14 TB, so an 8 TB drive. Under the 3-2-1 rule you need two such targets.
Where does the NAS and RAID calculator stop?
It calculates capacity according to the RAID rule, risks according to the datasheet and field data, and backup needs as an upper limit. SHR and RAIDZ are approximations. We work with you to determine which NAS suits your data, how to set up and test the backup, and what to do first after a failure, on site in the Vogtland or remotely.
11 / STUDIES AND SOURCES
What the tool is based on
The guide values and notes above are based on these 16 scientific papers. Each one has been checked via its DOI or the publisher’s page (most recently on 13 November 2026): it exists, it has not been retracted, and its findings still hold for this question today.
Schroeder, B.; Gibson, G. A. (2007): Understanding disk failure rates: What does an MTTF of 1,000,000 hours mean to you?. ACM Transactions on Storage 3 (3), Art. 8
Finding Replacement data from over 110,000 hard drives: in the field, usually 2 to 4 per cent of drives were replaced each year, in individual installations up to 13 per cent, while the datasheets lead one to expect at most 0.88 per cent. The intervals between failures are not independent but clearly correlated.
In the tool The tool calculates with 2 to 4 per cent per drive per year and points out that a second failure often does not take long to follow.
doi.org/10.1145/1288783.1288785 · Journal article
Pinheiro, E.; Weber, W.-D.; Barroso, L. A. (2007): Failure Trends in a Large Disk Drive Population. 5th USENIX Conference on File and Storage Technologies (FAST ’07), San Jose
Finding Failure data from a very large drive population in data centres: some SMART values correlate strongly with failures, but are not enough to predict the failure of an individual drive reliably. Temperature and load played a smaller role than assumed.
In the tool The page advises taking SMART warnings seriously but not relying on them: the backup remains necessary.
www.usenix.org/conference/fast-07/failure-trends-large-disk-drive-population · Conference paper
Bairavasundaram, L. N.; Goodson, G. R.; Pasupathy, S.; Schindler, J. (2007): An analysis of latent sector errors in disk drives. Proceedings of ACM SIGMETRICS 2007, pp. 289–300
Finding Of 1.53 million drives, 3.45 per cent developed sector errors within 32 months that only show up when reading. The share rises with drive size, errors cluster on drives that already have some, and over 60 per cent were discovered by regular checking (scrubbing).
In the tool It puts the datasheet calculation for read errors in context and explains the recommendation to run the NAS's data scrub regularly.
doi.org/10.1145/1254882.1254917 · Conference paper
Schroeder, B.; Damouras, S.; Gill, P. (2010): Understanding latent sector errors and how to protect against them. ACM Transactions on Storage 6 (3), pp. 1–23
Finding A single sector error can lead to data loss if it occurs during the rebuild of a RAID after a drive failure. The paper compares checking strategies (scrubbing) on field data and expects more such errors as drive size grows.
In the tool This is the basis for the “read error during rebuild” calculation and for the recommendation of double parity with large drives.
doi.org/10.1145/1837915.1837917 · Journal article
Bairavasundaram, L. N.; Arpaci-Dusseau, A. C.; Arpaci-Dusseau, R. H. et al. (2008): An analysis of data corruption in the storage stack. ACM Transactions on Storage 4 (3), pp. 1–28
Finding In systems with a combined 1.53 million drives, over 400,000 cases of silent data corruption were found in 41 months. Inexpensive nearline drives were affected about ten times as often as enterprise drives, and the cases occurred in clusters.
In the tool This supports the rule of thumb “RAID is not a backup”: a RAID mirrors corrupted data too; only older versions help.
doi.org/10.1145/1416944.1416947 · Journal article
Jiang, W.; Hu, C.; Zhou, Y.; Kanevsky, A. (2008): Are disks the dominant contributor for storage failures? A comprehensive study of storage subsystem failure characteristics. ACM Transactions on Storage 4 (3), pp. 1–25
Finding Logs of about 39,000 storage systems with 1.8 million drives over 44 months: drives caused only 20 to 55 per cent of failures, the rest came from connections and protocol layers. Failures occurred in bursts.
In the tool That is why the page allows for the failure of the whole device: no RAID helps against a power supply, controller or enclosure.
doi.org/10.1145/1416944.1416946 · Journal article
Schroeder, B.; Lagisetty, R.; Merchant, A. (2016): Flash Reliability in Production: The Expected and the Unexpected. 14th USENIX Conference on File and Storage Technologies (FAST ’16), pp. 67–80
Finding Six years of field data from SSDs in data centres: SSDs are replaced less often than hard drives but have more uncorrectable read errors; depending on the model, 20 to 63 per cent had at least one. The UBER metric is not meaningful, because the errors were not related to the amount read.
In the tool That is why the read error calculation applies only to hard drives, and SSDs in the NAS too need redundancy and a backup.
www.usenix.org/conference/fast16/technical-sessions/presentation/schroeder · Conference paper
Patterson, D. A.; Gibson, G.; Katz, R. H. (1988): A case for redundant arrays of inexpensive disks (RAID). Proceedings of the 1988 ACM SIGMOD International Conference on Management of Data, pp. 109–116
Finding The foundational paper that coined the term RAID: many inexpensive drives combined instead of one large one, with five levels that weigh performance, cost and reliability differently.
In the tool This is the starting point of the RAID table; the capacity rules of the classic levels go back to this classification.
doi.org/10.1145/50202.50214 · Conference paper
Chen, P. M.; Lee, E. K.; Gibson, G. A. et al. (1994): RAID: high-performance, reliable secondary storage. ACM Computing Surveys 26 (2), pp. 145–185
Finding Overview of RAID levels 0 to 6 comparing performance, cost and reliability: striping brings speed, redundancy brings protection against failure.
In the tool This underlies the capacity and fault tolerance columns of the RAID table and the diagram of drive layout.
doi.org/10.1145/176979.176981 · Review
Thomasian, A.; Blaum, M. (2009): Higher reliability redundant disk arrays: Organization, operation, and coding. ACM Transactions on Storage 5 (3), pp. 1–59
Finding RAID 5 survives one drive; but the rebuild can fail because of unreadable sectors. This led to double-parity schemes such as RAID 6. The paper also summarises studies on drive failures and the effect of scrubbing.
In the tool Explains why the tool recommends RAID 6 or SHR-2 instead of RAID 5 from four large drives.
doi.org/10.1145/1629075.1629076 · Review
Elerath, J.; Pecht, M. (2009): A Highly Accurate Method for Assessing Reliability of Redundant Arrays of Inexpensive Disks (RAID). IEEE Transactions on Computers 58 (3), pp. 289–299
Finding A model with distributions from field data for failures, rebuild, read errors and scrubbing predicts 2 to 1500 times as many double failures as the usual MTTDL calculation and matched field data from RAID groups of 14 drives.
In the tool That is why the tool does not calculate a “mean time to data loss”: this figure looks precise but underestimates the risk.
doi.org/10.1109/TC.2008.163 · Journal article
Elerath, J. G.; Schindler, J. (2014): Beyond MTTDL: A Closed-Form RAID 6 Reliability Equation. ACM Transactions on Storage 10 (2), pp. 1–21
Finding A formula for data loss in RAID 6 groups that models failures, rebuild, sector errors and scrubbing as time-dependent processes, supported by field data from tens of thousands of systems. It is considerably more accurate than the MTTDL equation.
In the tool This supports the point that RAID 6 too depends on regular scrubbing: the page lists it as a fixed step.
doi.org/10.1145/2577386 · Journal article
Ma, A.; Traylor, R.; Douglis, F. et al. (2015): RAIDShield: Characterizing, Monitoring, and Proactively Protecting Against Disk Failures. ACM Transactions on Storage 11 (4), pp. 1–28
Finding About a million SATA drives over up to five years: drive failures noticeably weaken the protection of a RAID; in RAID 6 systems, triple faults accounted for 80 per cent of RAID failures. The number of reallocated sectors correlated strongly with impending failures.
In the tool The page advises replacing drives with a rising number of reallocated sectors as a precaution instead of waiting for the failure.
doi.org/10.1145/2820615 · Journal article
Zhang, Y.; Rajimwale, A.; Arpaci-Dusseau, A. C.; Arpaci-Dusseau, R. H. (2010): End-to-end Data Integrity for File Systems: A ZFS Case Study. 8th USENIX Conference on File and Storage Technologies (FAST ’10), San Jose
Finding With deliberately injected faults, the paper shows that ZFS detects and handles a broad range of drive errors, but is much less well protected against errors in main memory.
In the tool This puts RAIDZ in context: checksums help against silent drive errors, but replace neither ECC memory nor a backup.
www.usenix.org/conference/fast-10/end-end-data-integrity-file-systems-zfs-case-study · Conference paper
Meyer, D. T.; Bolosky, W. J. (2012): A study of practical deduplication. ACM Transactions on Storage 7 (4), pp. 1–20
Finding File systems of 857 desktop computers: deduplication of whole files achieved about three quarters of the saving of the most elaborate block-level method, and 87 per cent for backup images. How much can be saved depends on method and data.
In the tool That is why no saving is preset in the calculator: the factor is an assumption that you choose deliberately.
doi.org/10.1145/2078861.2078864 · Journal article
Baker, M.; Shah, M.; Rosenthal, D. S. H. et al. (2006): A fresh look at the reliability of long-term digital storage. Proceedings of the 1st ACM SIGOPS/EuroSys European Conference on Computer Systems, pp. 221–234
Finding Long-lived data is threatened not only by hardware and software but also by people and organisations. The most effective measures: find hidden errors quickly, automate repair and make the copies more independent of each other.
In the tool This underlies the recommendation of several separate copies under 3-2-1, one of them off-site, and of regular restore tests.
doi.org/10.1145/1217935.1217957 · Conference paper
Standards, authorities and manufacturer data
- Choose a RAID Type (DSM help) · Synology Knowledge Center
- RAID calculator with notes on system partition and metadata reserve · Synology
- zpoolconcepts(7): raidz, minimum number and capacity · OpenZFS
- RAIDZ: space requirement by block size · OpenZFS
- zfs(4): spa_slop_shift, reserved space · OpenZFS
- Storage capacity measurement standards (TB decimal and binary) · Seagate
- IronWolf Pro 20 TB datasheet: read errors at most 1 per 10E15 bits · Seagate
- WD Red Plus product information: read errors fewer than 1 per 10^14 bits · Western Digital (copy at the retailer Farnell)
12 / MORE TOOLS
Still have a question?
- AT-W2 · Home & business
Windows 10 check
Does this PC still get security updates, and what needs doing by October 2027? - AT-W3 · Home & business
Wi-Fi planner
How many access points or mesh units does your home, office or warehouse need? - AT-W4 · Businesses
IT emergency check
How quickly could your business get back to work after an outage? - AT-W5 · Businesses
E-mail check
Are your e-mails landing in spam? Your domain’s SPF, DKIM and DMARC in plain language. - AT-W6 · Businesses
UPS calculator
Which UPS keeps your server, router and till running long enough during a power cut? - AT-W7 · Home
Digital legacy planning sheet
Which accounts, devices and contracts should your family know about?
Have you got your figures? Send us the copied result. We will suggest suitable devices and backup targets and set everything up so that the restore works in an emergency, in Rodewisch and throughout the Vogtland.
+49 3744 4399760[email protected]Mon–Fri 8:00–17:00 · Straße der Jugend 5, 08228 Rodewisch