A photographer’s Drobo was showing two amber bays at once. The problem isn’t the two failed hard drives, but the fact that BeyondRAID is non-standard (proprietary), undocumented and therefore unreadable by any standard tools.
This is a scenario, not a case study. It illustrates a common failure type that we address on a regular basis, and how it is approached, written from the technical side (versus as part of a single customer project). Please visit our documented case files here.
A user has a non-standard storage system (the most well-known example is Drobo's BeyondRAID), and it either stops displaying the volume, or the chassis itself fails. In a standard NAS setup, users are able to just plug their disks into another device as they are readable due to having been laid out in a way that standard tools understand; with this system, the disks cannot simply be read on other hardware.
Non-standard arrays have some real benefits. For example, they allow for use of different size disks; adding capacity to an existing array does not require rebuilds; and users can dynamically adjust how much redundancy is applied. To implement this type of array, you must move away from using fixed stripe layouts and instead apply a block allocation mapping that shows where each piece of data actually resides on disk.
This mapping is essentially what defines the “array.” Without this information the disks will contain data arranged in an unidentifiable pattern: no one will be able to determine a stripe order because there isn't one.
Therefore, users are dependent upon the manufacturer's implementation of their array. Users of a standard mdadm array can assemble it on any Linux machine; these types of proprietary arrays can only be read by the original hardware or a lab that has spent time reverse-engineering the format.
On a standard NAS, when the chassis dies, it is an inconvenience: you could move the hard drives into another box and access them again. In contrast, on a proprietary storage solution, it can be the whole problem, and if the vendor goes out of business or no longer supplies replacements, you may never find a working replacement for the proprietary enclosure.
As such they come here more regularly than their actual market share would suggest. Frequently, the hard drives themselves are in good condition, but the path to the data has been cut off.
The recovery process includes parsing the allocation map directly from the disks and then reconstructing the volume from the disk images based upon the allocation map, all dependent upon knowledge of the original format, as opposed to a generic technique used to recover files.
Although it is more complex than traditional arrays, it is possible. BeyondRAID is a custom (proprietary) and undocumented technology. As such, the mapping structures that identify where each block lies must be reconstructed from the disks themselves, instead of simply being determined through common test procedures designed to detect standard striped patterns.
BeyondRAID uses non-standard methods to distribute its data (variable-sized blocks), along with applying redundancy at a regional level. A mapping layer provides documentation as to where each individual block resides; this is what enables support for different size hard drives. It is also what renders traditional geometry testing techniques ineffective.
Not always. Disk failure (a “flagged” disk) often does not mean the disk is destroyed; frequently it is due to bad sectors, or the disk dropping out under load while still able to read. In a virtualised configuration, each readable sector from a failed member directly lessens the shortfall.
No. This would prompt the system to rebuild its mapping structure(s), which are exactly what a recovery operation needs to read. Therefore, please power down the unit and keep every disk exactly where it is.
Yes, this includes any that the unit has identified as having failed. Because redundancy is applied on a per-region basis rather than uniformly, some regions have no redundancy at all, and therefore only the original disks contain these regions.
Prices start at £500 plus VAT after a free 48-hour diagnostic period, with a 50% deposit required when a member requires physical drive-level work. Please send all disks to us, clearly labelled by bay if you know the order.
Despite the redundancy within the unit, there still exists a dependency upon the chassis. This is an additional potential failure mechanism that standard arrays do not have.
The allocation map will be distributed across the RAID members. Frequently, a partial set does not have enough information to allow reconstruction. The bay order will also determine how the map will be parsed.