The two questions that lead people here are: should my next array be RAID 10 or RAID 5? and my RAID 10 just lost a drive: how scared should I be? Same tech, same answer: It’s all about how RAID 10 is configured, and the maths of failure.
RAID 10 is a configuration that uses a mirroring method with drive pairs (that is to say each pair will hold an identical copy of your data) and then uses those paired sets to distribute the data among them to provide faster access. This gives you the advantage of using both the mirrored pair as well as sharing the workload of writing/reading information from the pairs.
RAID 5 distributes your files over all of your drives. A portion of each file will be distributed to different hard drives. In addition, RAID 5 creates an additional layer of information on top of the individual files called parity which acts as a safety net. RAID 5 allows you to purchase N number of drives but keep N−1 of the total space. If any drive fails RAID 5 will continue to function with no data loss. RAID 10 does not require any calculations to determine how to distribute your files. RAID 10 simply duplicates all of your files. Therefore you retain approximately half of the total capacity. Writing occurs quickly since there are no calculations involved in writing to your disks. If any one disk becomes unavailable it may be replaced by duplicating from the identical copy stored on another drive. This duplication process takes only hours and puts gentle load on one survivor, compared to RAID 5 which uses a parity rebuild process that hammers every remaining drive for a day or more. It is this rebuild time that causes many users problems on large modern drives. Tired secondary disks often fail during RAID 5 rebuilds. Honest Buying Rule: Buy cheaply with lots of storage = 5 (RAID 5) or 6 (RAID 6, the two failure version). Database applications, virtual machine environments & other write-hungry applications where down-time costs money = buy RAID 10; do not even think about RAID 5.
In the case of RAID 10’s often-quoted “survives multiple drive failures”, there is a location component. RAID 10 will shrug off a failure in each of two different sets of matching drives (the ‘pairs’), but the second failure within an already-wounded set of paired drives will take down that set of paired drives for all files since the original pair was the only copy. The way things really go also stack the odds against you. For example, if you have two identical models, manufactured at the same time and purchased together as part of the same production batch of the same type of drive, which you install at the same time and use under the same workload conditions, then after one fails, it is most likely that the next drive to fail will be the other in the pair. When this occurs, it becomes even more probable that both drives in the failing pair will fail due to increased stress on them. Finally, add to these mechanical issues those induced by humans, i.e., panicking about a poor condition array (pulling out the wrong drive or forcing a hot spare into an auto-rebuild process against a weakening twin drive). And you can see why RAID 10s end up on our bench.
There is a simple rule that allows you to keep a failing array recoverable: power it down, photograph and label every drive with its slot, change nothing, force nothing. From a professional standpoint, after this point, each disk is then cloned (copied) read-only and the original array is rebuilt from the clones rather than the original disks. This process of cloning and rebuilding is what we use for all of our RAID recovery work, regardless if it’s a large server on the backplane of a rack or if it’s simply the RAID 10 configuration found inside of a four-disk NAS device.
Absolutely not, and this misconception causes problems for many recovery benches. RAID 10 will protect against one single event, i.e., when a single hard disk dies. It does NOT provide protection from deletion, ransomware, corruption, failure of the controller, fires or theft. Once either of these events occurs, both copies of the data will be damaged simultaneously. An array will allow you to continue running but only a separate, disconnected copy of your data will keep you safe. You need both, for different reasons.
The safety argument is still valid; however, the speed argument has diminished. Striping doesn’t have as big a role in improving performance with solid-state drives (SSDs) as it once had with hard disk drives (HDD). Many build configurations can receive equal protection by simply using RAID 1 mirroring. Where RAID 10 with SSDs will still provide value for heavy write operations, such as large database applications and heavy use virtualisation environments, is the ability to sustain high amounts of data transfer along with the faster and lower stress rebuild times.
Yes you can, and this is typically how many two drive failures begin. An array that is degraded will be able to operate with no additional protection available since one half of the mirrored pair has failed. However, each member of the array has similar characteristics (same model, age, amount of wear). So when the other member (the live member) of the failed pair starts to fail due to the increased workload, the second failure may occur rapidly. If the information contained within the array is critical to you, then consider a degraded array to be an “event,” rather than just another “status.” Treat this situation seriously and take immediate action to back-up all data currently stored on the degraded system and replace the failed member(s). If at any time you experience uncertainty about the rebuild process, create a backup image of the original data prior to the rebuild.
All members of a multiple-disk array can be diagnosed free of charge within 48 hours. Members are imaged in read-only mode, rebuilds are performed against images, and a formal written quote will be provided prior to beginning any work.