2

If you have a basic RAID 5 array of 3 or more hard drives, and one fails, does this increase the chance of one of the other drives failing and making the full recovery impossible? If every one of the drives is being read from during the recovery process, does this make them any more likely to fail, assuming they're only active a small portion of the day?

If this does affect stability during a recovery, I may be considering RAID 6 instead. Also, is there any benefit or disadvantage of using RAID 5 instead of RAID 4? It only seems more organized to use RAID 4 and more spread out to use RAID 5.

  • RAID 6 is clearly better if you can afford the cost. – cybernard Jan 26 '14 at 20:16
3

First off, if your are set on a parity raid I would recommend a RAID 5 with an automatic backup scheme over a RAID 6. The backup could be over multiple drives (recommended), a RAID 0 (not recommended), or another RAID 5. This carries a good balance of cost and safety for your data because it can also handle controller failure and human error.

What you will be safe doing depends on your array size(s)

Many in the IT industry are moving away from RAID 5 for two reasons. First, the shear size of many of the array's require extremely long rebuild times which opens the window to long for a second failure. Second is the issue of a URE which is an unrecoverable read error that is not known until a block fails to be read and then the entire array on that drive becomes unreadable. Again, UREs happen rarely in terms of shear number of errors per bits read, but when you have TB size arrays, rarely is too often. An array cannot be rebuilt from a drive with a URE. So if you have a failure, and a URE, you effectively have 2 failed drives. RAID 6 helps with this because you can lose two drives and still rebuild the array.

But, if you have a 100GB array in a RAID 5, these problems diminish greatly and a RAID 5 becomes more viable; however if you have a 100TB array (just as an example) on a raid 6, you might need more redundancy or another option!

RAID 5 with TB arrays is definitely a gamble and I would not trust it. And it does not make sense to run a RAID 6 with 4 drives over a RAID 10 unless you are going to be expanding your array soon and need to do so 1 drive at a time for cost. And let's not forget that you can run multiple RAID 1 arrays, it just segregates your data into separate partitions/disks you have to work with and takes away the convenience of one big drive.

But everything has its caveat. The larger the RAID 5, the higher (exponentially higher) chance of a rebuild failure due to second failed drive or URE. RAID 6 higher cost and complexity. RAID 10 higher cost yet (but for simplicity). And multiple RAID 1's cost of convenience.

In the end, you need to identify your needs, segregate what you cannot lose, and what you are ok losing. Match that with a budget and your level of expertise while choosing which array to choose.

Us, we put our office and home data on a RAID 1 (2 drive RAID 10 with MD RAID actually) for safety with daily backups. its simple, fast, and reliable. Any other data like video files or other large non critical files we put on a RAID 5 or single drives with occasional backups based on how important the data is because we do not want to afford the same level of safety with our movies as we do our family photos and/or our companies' files.

  • The problem is that I'm building an online backup service for a large number of customers. I can't really separate the data between RAID 5 and RAID 1 because it's designed and marketed to be highly secure (to the point where the password is never sent to the server or stored, and same with the encryption key). It's kind of ruining it by asking how important each section of their files are to my customers. Would keeping a parity drive for every 2 regular drives be safe? I want this to be secure but also cost-effective. I would do RAID 1 (it was my first thought), but RAID 5 saves us 25%. – Phoenix Logan Feb 1 '14 at 14:36
  • RAID 6 doesn't seem to save any more space than RAID 6. I still need double what is actually being stored. RAID 5 seems like the best option as long as I keep it as a ratio of storage drives to parity drives. 3:1? Also, I'm wondering, does the size of the drives affect anything? They probably won't be over 2TB, but would a larger drive make a difference to the safety of the recovery? – Phoenix Logan Feb 1 '14 at 14:43
  • If the storage is for paid customers, then no data is expendable and a simple RAID setup is not going to be secure enough. RAID simply allows for a drive to fail and data to not be lost; it does not cover infinite drive failure, array failure due to corruption, hardware failure, or human error. This is where things like HA setups with DRBD, LVMs, SANs, DASs, and the like come in. – Damon Feb 3 '14 at 4:10
  • Also, for RAID 5 and 6, the parity is distributed across all drives to even out the parity writing/reading workload, so you don't get a dedicated parity per x number of drives. As to the number of drives in an array that is safe, there is not magic threshold. It's all about playing the numbers based on your setup and how you manage it. If you only have a RAID and nothing else, than 3-4 drives might be too much, but if you have multiple layers of redundancy and fail-over, you can increase the number of drives in your arrays safely. – Damon Feb 3 '14 at 4:16
3

The act of rebuilding the RAID array can take a long time (days) and be very hard on the disks. During this rebuild time, the surviving disks are more susceptible to failure due to their activity.

Before the rebuilding is started, the surviving disks are not at an elevated chance of failure.

For this reason, it may be inadvisable to have a "hot spare" in the system since it will automatically trigger a rebuild and cause a failure. Also, encountering an unrecoverable read error during rebuilding could cause an array failure even though no further disks have failed. Better to copy data off the array first, then rebuild, if there are no recent backups.

1

Short answer YES - RAID 1+0 was designed SPECIFICALLY because of the likelihood of failure during a rebuild of RAID5.

Long answer - Yes, there is an elevated chance of the drive failing. In the "most obvious/simplest" case it can be argued there is more heat being generated thus a slightly higher chance of failure. That said, this is not the real risk.

The real risk is that if the drives were purchased at about the same time, there is a greatly elevated chance the drives will fail at about the same time. Thus the chance of the second drive failing while rebuilding is increased. Unfortunately this scenario is not academic - the bigger the drive the longer the rebuild and the greater the chance of failure. RAID5 is generally accepted as being of little benefit with large disk pools. The accepted solution is raid 10 (Read as RAID one 1+0). This basically builds up 2 identical sets of data across 2 sets of arrays - thus it uses a a lot more (almost twice as much) space as RAID5, but provides much greater protection against multidrive failure - which is, unfortunately, common on large size drives.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.