RAID 5 · a single parity disk
One of these lands most weeks. A member drops out, nobody reads the alert, a spare goes in months later, and the rebuild stops on the next disk. Sets come in from freight offices on Hedon Road, accountants and solicitors in the Old Town, and workshops on Sutton Fields, every one a Hull address. The news is normally better than the front panel makes it look. Failed is what the card decided, not a diagnosis. What goes back to you is read off an image; your own disks are left alone.
An engineer will take the details
0800 6890668
| What you see | What the fault is | What to try |
|---|---|---|
| The card reports one member missing | Parity has been spent. One more fault takes the volume | Shut it down before another goes |
| The rebuild ran for hours and stopped | The full read has found a weak patch on another disk of the same age. That is the usual ending | Keep it switched off. Do not try again |
| Two members show as failed | Usually a single dead disk beside one that has a few sectors it will not read, not a matched pair | Send the lot. The images will show which is which |
| An ejected member was put back online by hand | Old blocks are now mixed in among current stripes | Do nothing else |
| A replacement card cannot see the array | The disagreement is about metadata. The files are exactly where they were | A plain reassembly job |
| Someone has already run recovery software on the disks | The original members have been written to | Send them anyway, and name the tool that was used |
Each item that arrives gets a case number the day it lands. An engineer looks at it and finds the real fault, and that costs you nothing. You are told in plain terms which files can come off the drive and which cannot. A price follows — one figure, written down. No work starts until you have seen it and agreed to it.
Every disk is copied onto our storage in full, the condemned ones as well. Weak reads are logged as they happen, and the most awkward areas get left until last. Nothing at all goes back onto your originals.
Two things give up the disk order, the stripe size and which way parity turns: the metadata each member is carrying, and the pattern the data makes. No part of it is guesswork. No part of it needs your controller.
The set is assembled on top of the copies. Each block comes from whichever image read it best, and where none of them managed it, parity supplies the block — which is how files marooned on a condemned member come back. The file system is then repaired and the volume mounted to check.
You are not billed while you are still thinking. The list of what came off the drive goes to you first, and the invoice waits on your word. Files go home on media bought in for the job, with the postage back on us. We keep the job open until you have opened them on your own machine.
Do the sum once and then stop: the published rating for a consumer disk is one URE — one unrecoverable read error — for every 1014 bits it reads, so about a bad sector per 12.5TB. On enterprise disks the figure is 1015. Rebuilding a four-bay chassis of 16TB drives asks roughly 48TB of three survivors with no stumbles allowed, so the risk is real rather than theoretical. Plenty of disks also last well beyond the rating. Both hold, and both point the same way: image first, and the rebuild after that costs nothing.
Disk two had dropped out in June and nobody had looked. Disk four then failed part way through a rebuild in August, which read from outside as two dead disks at once. It was not. Imaging showed nineteen bad sectors on the fourth and nothing else wrong, so the three healthy members and that one almost-healthy disk were enough to rebuild the stripe. The year-end filing was made on the date it was due.
Not usually. A card ejects a disk on its own terms, and sectors that answered too slowly are enough to do it. That is not the same as a dead drive. Once every member is imaged onto our own storage, anything a single image will not give up is rebuilt from the parity held on the others. Two disks thrown out by a controller is an ordinary week. Two disks genuinely dead is not.
Not before the disks are imaged. A rebuild makes every surviving member read itself out in full, and that is what finishes a disk already close to it. It also overwrites the parity we would have used to put the volume back. With images taken, rebuild as many times as you want; the copies stay as they are.
A URE is a sector that beat the drive's own error correction and got given up on. The published rate is about one in every 12.5TB on a consumer disk and one in 125TB on an enterprise one. Read that as the worst figure the maker will stand behind, not a countdown. Current firmware meets a URE by puncturing that one stripe and carrying on.
No. The disks are enough on their own. An array's layout is written on the members themselves, and where that has been damaged the data gives it away instead. We rebuild in software from the images. The disks you send are read, never written.
Leave it switched off. A failing drive has a limited number of starts in it, and each one spends a little more. The free diagnosis will say which files can be read.