Data is distributed across every member without redundancy or duplication. Therefore, if a member fails, a proportionate amount of data will be lost from every file stored on the volume. Be cautious when someone offers to rebuild from parity since there is none.
There is no reserved space on a stripe, so recovering means reading each member (including the member that has failed) and reconstructing from images. The amount of recovered data depends upon how much of the failed member can be read.
All the other RAID levels keep something “back”. RAID 1 maintains an entire duplicate (second) copy. RAID 5 will maintain a parity that allows for the reconstruction of any single missing member. In contrast, RAID 0 retains no redundancy (it splits your data across every disk purely for speed), so each disk contains a portion of each file on the RAID volume.
So when one member of a stripe fails, it does not mean you lose that disk's share of the files. You lose a portion of each file. For example, if a document is broken up across four disks and one of them is missing, you will miss a quarter of it (spread evenly across all of its parts).
Thus, someone who claims they can rebuild your RAID 0 using parity should be viewed with scepticism, because there is no parity. Your only way back to the data is by physically recovering the bad disk.
The failed disk is recoverable. Thus, you can also recover the RAID. This would involve imaging the disk (like you would for a single drive) and then deriving the stripe information. From the images of all members, you could then recreate the volume.
The failed disk is partially recoverable. If 90 percent of the failed disk can be read, you will get a proportional result: around 90 percent of your volume recreated. However, damage will be spread across all of your files instead of just a few.
The failed disk is unrecoverable. Larger files are then completely lost while smaller files may remain intact. Files which are entirely contained within one stripe unit on a remaining disk can still be retrieved by carving out those files. However, if a file spans multiple stripe units, the file can never be fully restored.
Someone has already rebuilt over it. This is the worst-case scenario. When you initialise or recreate a RAID 0, you are writing new data (metadata) about how to read and write across all of the drives in the RAID 0. Frequently, the existing stripe geometry is rewritten as well. Since RAID 0 does not provide redundancy, there is no “extra” information that can be written in place of the missing information.
Turn off the RAID set immediately and leave it off. If an array is degraded, then it provides no benefit to anyone. Every hour that passes increases the likelihood of another member of the array developing problems as well.
Label each disk with what position it held within the stripe geometry. While determining bay order (which disk was located where) can be done mathematically, having the bay order recorded for each disk will save you time during the recovery process.
Send every single member of the original array, including the failed one. And especially send the failed one; it is the whole job. If the failed member is not included in the shipping package, then the array cannot be reassembled. This is because there is simply nothing to recreate the missing slices from.
In cases where either the controller itself is dead or the configuration of the RAID system was never properly documented, there's no record of how the stripe was constructed. All of the information required to determine this (disk order, stripe unit size, start offset, and whether the controller was configured to use a left or right arrangement) must then be determined using only the contents of the member disks.
The approach is to look for structures that make sense only if you have chosen the right parameters. A file system superblock is located at a predictable position. The internal organisation of a large file runs continuously. Either the test layout fits these structures into alignment or it doesn't. Wrong parameters provide results that appear nearly believable. This is why verifying your results against known file structures is more important than obtaining a result at all.
The volume is reassembled from images (not the disks themselves) after the geometry has been verified, so there is nothing else that could be lost.
Only through recovery of that physical drive. Since there is no parity and no mirror, there are insufficient pieces to rebuild the lost data. If the failed member is recoverable then the RAID is also.
Approximately proportional to the amount of recovery you can achieve, but the loss is scattered throughout everything. A 90% recovery rate on the failed member does not mean that you can recover 90% of your files intact. Instead, you can recover most files with small gaps in them.
Since every file is divided among all four members, losing one member causes you to lose a quarter of each file rather than a quarter of your files. This is what we mean by “striping”.
Yes, as well as the failed unit, with labels (if practical) to identify their positions. Missing units cannot be reconstructed on RAID 0 because there is nothing reserved to recreate them from.
Yes, if it's used for scratch area and render caches where speed matters and loss is acceptable. If the data is important enough that losing some or all of it will impact you negatively, then don't use a RAID 0 because you multiply the probability of experiencing a failure by the number of drives within the set.
From £500 +VAT for work on a RAID array; this price is fixed in writing after the complimentary 48-hour diagnostic. If additional mechanical work is needed on the failed drive, we will identify it and quote for it at the same time.