11

I'm just about to set up my hard drives with RAID 0. Is it really as risky as people say it is?

0

13 Answers 13

24

RAID 0 doubles your chance of storage failure (assuming a 2-disk Raid 0). Why? Because now you are relying on the reliability of 2 disks, instead of just one.

RAID 10 brings back a reasonable level of reliability.

6
  • RAID10 is only a "reasonable" level of reliability?
    – Josh
    Nov 2, 2009 at 13:26
  • 54
    @Josh: Yep. Fire doesn't check what RAID level you've configured on your server.
    – womble
    Nov 2, 2009 at 13:31
  • @womble +1 - that's quite amusing :)
    – Coops
    Nov 2, 2009 at 13:53
  • 2
    Obligatory "RAID is not a backup." Fire, fire, fire!
    – Bratch
    Nov 4, 2009 at 1:23
  • 4
    @womble: Your comment is more voted than this answer, and the question combined.
    – Macha
    Nov 4, 2009 at 21:30
9

RAID 0 means ZERO redundancy. Whenever there is data to be written to the RAID device, it is split in two, the first part is written onto the first disk, the second part on the second, which makes your write operation pretty fast. But if either disk breaks, all your data is lost (since you lose (roughly) 50% of all your files, rendering all of them useless)

2
  • 9
    To clarify "lose 50% of all your files"... Half of each file is lost; you can't recover 50% of the number of files you have. Although some expensive data reconstruction services can come pretty close... sometimes.
    – tsilb
    Nov 4, 2009 at 2:19
  • ahh so thats how you remember which is which! duh
    – Simon
    Mar 17, 2010 at 21:27
7

Yes. If you lose just one drive in the array, you lose everything. Which means anything on RAID 0 must have a backup.

6
  • 17
    Even if using anything other than RAID 0, you still must have backups.
    – Juliano
    Nov 2, 2009 at 13:35
  • 3
  • @Juliano, yes, you're right. I guess if you have RAID 0 you "positively absolutely really truly must have backups"?
    – Josh
    Nov 2, 2009 at 15:32
  • 2
    If you don't "positively absolutely really trust" your backup then it isn't a backup. The point of backup is to restore. If you're not testing that and sure of restoration success then it's not a backup. Nov 3, 2009 at 4:38
  • 1
    RAID is not a backup. Anything you haven't done a test restore on isn't a backup. Anything that's vulnerable to the same things as your main storage isn't a backup. And, yes, RAID0 will fail more than other storage. Nov 4, 2009 at 21:05
7

It depends on what you are using it for. As others have said it roughly doubles the risk of failure of the data on it compared to a single disk. If you had 3 disks in raid 0, then 3x the risk, etc... This because you lose almost all data if any one disk fails. RAID operates on a low level, so it doesn't generally put one file on one disk and another file on the other disk, but rather will generally split up the file between the however many disks.

However, If you have a backup, and it is static data, and uptime doesn't really matter, then it is not dangerous. For instance, I might use raid 0 on a gaming machine with my save files on another disk. This way, level loads will be fast :-) But for IT appropriate uses, its not the ideal choice.

An IT situation would be redundant machines with static data. In this case, it is okay for a machine to go down for a while.

Lastly, hard drives are mechanical, and break quite often. You might not see this if you just have a couple of desktops, but with a server room and lots of disks, you will be replacing them fairly often.

2
  • I'm relatively new here, and "serverfault" isn't in my dictionary, so I have to ask, what are these things: "serverfault situation", "serverfault appropriate uses". Is serverfault like being a scout or something? We get badges? We are always prepared? This language is strange to me.
    – labradort
    Nov 2, 2009 at 14:32
  • Oh sorry. Well there is superuser for Desktop sort of users, and Serverfault for system administrators. So when I said serverfault, I meant IT or large scale. See the faq Link in the upper right of this page. Nov 2, 2009 at 14:57
4

Yes it is. you have 2 disk, raid 0 - if one fails, all data lost. if you don't use raid - if 1 fails, 50% data lost...

IF you use raid 1 - if 1 fails - you have 0% data lost, but you pay twice for your Hardware.... :-)

4

Speaking from personal experience of losing data, I'd definitely recommend you save yourself the headaches and avoid RAID 0. For each drive in the array, you increase the chances of losing all the data. I had 3 drives in RAID 0 and the middle drive broke only a few months later, losing nearly 1TB of data.

4
  • 4
    My experience has been that if one drive fails, the others are about to go too. So, I don't worry so much about RAID 0. Just do your backups. I lost 2 drives out of a 4-drive RAID 5 array the same day, then rebuilt it, lost a 3rd original drive a week later... If possible, avoid drives that are the same age, from the same batch in RAID arrays of any kind? :-) Nov 2, 2009 at 14:23
  • @Brian, that is a great point at the end. We had 17 drives fail in several of our SAN devices at work on the same day. They were all the same age and device batch. Nov 2, 2009 at 15:46
  • @Brian Knoblauch: I completely don't understand your point: How can you not worry about RAID0 just because if one drive fails, the others soon will? With RAID1, RAID5, or RAID10, if one drive fails, you don't lose data. With RAID0, you lose all your data. Doesn't matter if the other drives are about to go... I mean obviously it matters in so far as planning on replacing the others soon... but with RAID0, you lose everything. The other drives going soon has no effect on that.
    – Josh
    Nov 4, 2009 at 2:37
  • @Josh: I believe his point is that you need backups no matter what. Note that losing two drives out of RAID 5 also loses data, so that's not safe. That being said, it looks like using drives of the same age in RAID 0 might be reasonable, but replace both if one drive fails. Nov 4, 2009 at 21:08
1

I dont think Raid 0 is risky at all. I personally run raid 0 for my os for benefited speed. You could hose my raid config any time of any day and I wouldn't lose a thing. I have my system set up correctly, to get the benefits of speed while having little to no possible loss of data.

The only risk is for people who don't know how to distribute risk.

I can say the above because I simply only work on important files on another drive than that being used to hold my OS. I install applications and stuff on my main OS drive, but their config files and such are all on a secondary drive. That secondary drive is then mirrors to another drive and I then make weekly backups of that mirror to a external drive. If my Raid 0(OS Drive) were to fail, I simply pull the drive/s out and put another one in. Take a Linux live cd and use a program like dd or cat to copy a already made image over onto the raid drives. I restart and now my system is back up to a pristine state.

0
1

RAID 0 can be a great solution in many cases where downtime is not critical.

RAID gives you speed, or redundancy or both. RAID is not a backup solution as it only guards agains hardware failure, only one type of failure that can cause data loss. So that means you need to have a backup solution in place regardless of the RAID type you use.

Regardless of the RAID type you employ you will have degraded performance in case of a disk failure or array rebuild, and since these days HDD's are getting larger and larger and speed is more or less stagnant - the rebuilt time can stretch to days or even weeks and create unnecessary stress to the disks.

I personally prefer to run RAID 0 which in case of a disk failure will be restored from backup. That to me is the faster, cheaper, less stressful solution since a bit of downtime is not really a concern.

Remember , when you are rebuilding an array , every block of data is being read / written on all disks, regardless of the quantity of useful data on the disks. In practice that usually translates on a much faster data rebuilt from backup compared to a RAID rebuild after a disk failure. Combined with significantly less write / read cycles in normal operation of RAID 0 compared with RAID 1 / 10 or RAID 5 / 6 it will translate in less disk failures overall.

All of the above is true if you have a sound automatic backup strategy in place , for example:

  • continuous backup on hot files
  • daily backups on cold(ish) files
  • weekly backups on all files on media that is located remotely.

To minimise downtime even when using RAID 0 you can tier the hot data backup on fast mediums like and external SSD, since the hot data-set is usually small and external SSDs are quite cheap these days, and backup cold data on slow mediums. Most commercial NAS systems, event the cheap ones, will e able to do that for you these days without much of a hassle.

In mission-critical systems all of the above is irrelevant as downtime is not really an option , but even then there are still better solutions in place - like snapshots replication on ZFS, brtfs or erasure coding or server redundancy with failover systems in place.

0

In 99% of cases RAID-0 will probably be a bad idea for all the reasons other people have mentioned (increased chance of complete failure etc).

However it can be used in some (more extreme) situations, for instance in a large array of servers working together (a la google), where individual machine failures do not critically effect the final output, but RAID-0 is a quick and cheap way of expanding capacity.

But to be brutally honest and not trying to be mean, but if your asking such a question you won't be in the latter group of server admins :)

0

Lots of guys talk about probability and they are mistaken! If you want the risk of using raid 0 that is how it's calculated.

Suppose the probability of hard1 failure is p1 (in our time period unit) and hard2 is p2 then when have a series circuit because failing one hard will broke the circuit.

So to calculate the risk we have: risk = (1-p1)*(1-p2)

For example if hard1 and hard2 failure probabilities are .001 then the risk is : 0.998001

So as you see the failure of 2 hard disks almost is the same as the probability of 1 hard failure that is not too high.

But there's something else here that change the situation and that is you cannot recover files from raid broken hdds too easy, actually in my experience it's impossible.

5
  • 2
    (1-p1) is the chance of working ok. Their multiplication (1-p1)*(1-p2) is the change of both devices working ok.
    – kolypto
    Nov 4, 2009 at 1:38
  • 1
    Sooner or later each drive WILL fail. It's not a matter of if but when. Therefore, as you have not indicated a time period to make any calculations meaningful, the probability can only be 100%.
    – John Gardeniers
    Nov 4, 2009 at 3:02
  • 1
    Not to mention that, for low chances of failure, the chance that one of two disks will fail is almost twice the chance of one. .999 is not almost the same as .998001, if you start risking failure over and over. Nov 4, 2009 at 21:10
  • 3
    And the probabilities are not independent, as both disks are probably the same and they work in the same environment. So you cannot just multiply values.
    – liori
    Nov 5, 2009 at 0:06
  • 1
    i think this is just the probability that both of them fail at the same time...
    – ultrajohn
    Oct 1, 2013 at 5:08
0

The risk factor of using a RAID 0 depends on pretty much for what you are going to use it.

As for my experience, I started using RAID 0 a year ago using 2 inexpensive disk for improved performance in games and some video editing and 1 large hard drive for simple storage. I keep all my important data backed up where I consider to be safe.

What will I do if one (or both) disks die on me? I have an image of the SO with all my programs and drivers installed, so I'll throw the defective disk, put the other to some use until it dies like its brother, buy a couple brand new ones, and re build the array and procede to leave my computer as I want it.

0

I believe that raid5 will give you some of the benefits of both raid0 and raid1, so you get both increased speed and redundancy compared to using only one drive. It takes three disks at a minimum though.

1
  • 2
    Note that raid5 is pretty bad at random writes, since you need to do 3 IOs (a read and two writes) to update a single isolated block. That's worse than either raid0 (one write) or raid1 (two writes) Nov 4, 2009 at 23:59
0

Probability theory says that is P1 is the probability of device-1 failure, and P2 - for device-2, then failure of any of them occurs with a chance of P1+P2. That means, chances to lose data are really higher, actually - twice higher.

Practically, I have RAID 0, and it does not fail: HDDs have 1-2 guaranteed years of stability, and often they work ok during this period. You can monitor their S.M.A.R.T. parameters to ensure the array won't fail unpredictably.

I believe nowadays hard disks are the slowest computer part. By using RAID 0 I realized that everything works MUCH faster and the whole system is more responsive: OS loading, soft loading, .. Actually, now I'm getting tired each time I have to touch systems with no RAID inside :)

However, as Josh said, you need to make backups. Not only when using RAIDs ;)

6
  • 2
    That's not correct. If both devices have a 50% chance of failing within 5 years there is not a 50%+50%=100% chance that one of them will fail within 5 years.
    – Amok
    Nov 3, 2009 at 2:34
  • No, that's correct. We have Probability(FAIL-1 or FAIL-2) here, which is actually Probability(FAIL-1)+Probability(FAIL-2). Check it out: en.wikipedia.org/wiki/Probability#Mathematical_treatment
    – kolypto
    Nov 3, 2009 at 3:03
  • 3
    No, it's wrong. Read your own sources again. Don't try to include the math if you're not going to pay any attention to it. Nov 3, 2009 at 6:03
  • 2
    "HDDs have 1-2 guaranteed years of stability". If the drive fails you will be compensated prorata for the time left on your warranty. You will NOT be compensated for the data you just lost. Further, S.M.A.R.T won't prevent, or necessarily warn you of, a total and sudden drive failure, as we see all too often. Why run risks with the single most unreliable component in a computer?
    – John Gardeniers
    Nov 4, 2009 at 2:58
  • 1
    An even better example is, let's say, 80%. The probability of failure is obviously not 160%. The probability of any drive failing is P1 + P2 - P1 x P2 so if both were 80% the overall probability would be 96%. The probability of both drives failing is P1 x P2 which would be 64% using this example (so using two drives is an improvement - sure your chances of some sort of data loss increase, but the chance of loss of a particular piece of data is reduced). Jan 9, 2010 at 8:38

You must log in to answer this question.