Tell me the downfalls of RAID

[edwardaune]

Can anyone tell me anything negative about running RAID (RAID 0 striped to be exact), as far as installing programs do you guys see a lot of errors stating that a file is corrupt or missing even though you just used the same software a few days before that. I just want to know some of the darker things about RAID. RAID-Kills Computers DEAD.

[FireMogle]

RAID 0 simply increases the risk of loosing data due to a disk failure by adding more disks into the fray. If one goes all the data goes.

[Veland]

So if you have some cash to spare, go for a 1+0 (or 0+1 depending on setup) Full speed with full redundancy.

Also, Raid.. ehh.. 3? might be a solution. The one that stripes to two drives and writes parity information to a third. But you need a separate raid controller as the onboard ones rarly support this configuration. And there is some overhead because of the parity calculations.

[CrazyP]

Originally posted by FireMogle
RAID 0 simply increases the risk of loosing data due to a disk failure by adding more disks into the fray. If one goes all the data goes.

I never really bought into the whole increased risk thing. the simple fact is, whether its raid 0 or a single drive setup, if one drive goes, there goes your data. so regardless of how the system is setup, backing up is the only sure way of keeping data safe.

[Veland]

Not quite that simple.. If one drive has, say, 10% failure rate. If you add another, your total failure rate is increased. My math is a bit rusty on it.. Ok, a practical example then:

You save a file, either on a raid 0 or a normal disk. Failure rate for the drives is still at 10%

On a single drive, the chance of failure is 10%
On the raid 0, the failure chance is 10% on the first disk, plus 10% on the second disk.

I know, 10% failure is sky high, and failure rate is not measured that way. But it is just an example.. If someone knows more about HD failures, rates and probability math, then feel free to correct me.

[bdf24]

I can understand what you mean by increasing the failure rate. I've had a number of drives go bad over the years. Drives that I've owned and I've seen quit a number of other peoples drives go bad. So by throwing in two drives in a raid setup. You would actualy be increasing your chances of loosing data.

But like Crazyp mentioned if your that worried about your data. Then you do need to back it up. Either on a seperate drive (this is what I prefer) or DVD-R/+RW if you have one.

[wicked-one]

I'd say most modern drives look to have a failure rate of about 1/20 within a month. So .95 * .95 = 90.25 = 9.75% chance risk of failure.

Biggest issue I have with RAID, is it really provides very little benefit. The main reason to use it is if you do a lot of image editing. You will notice increased boot times and loading times in games, but for the most part, that's about it. You're doubling your expenses, doubling your risk of failure for next to nothing. Read storagereview's FAQ on it:
http://faq.storagereview.com/tiki-index.php?page=SingleDriveVsRaid0

95% of hard drive processing is working with small files. RAID really only starts to provide benefits with files 1MB and above. It's hype.

[David]

Downfalls with RAID is simply that you either increase the chance of failure, increase cost for the same capacity, or both.

Ie: RAID 0 using 2 60GB drives for example - chance of failure goes up.
RAID 1 using same drives - you only actually get 60GB storage out of 120GB. Thus it costs you twice as much.

As has been stated several times though, the best solution is to back up regularly.

[Cjwinnit]

Ah, my favourite topic. Excuse me while I ramble.. ;)

Maybe we should clarify a few things.

What has been said in the thread is correct but something should be highlighted. Having a disk in a RAID array does not increase the chances of that individual disk failing. It is just as likely to fail running by itself as it is running in a RAID array. However, a RAID requires many disks, and a partition readable by the computer's operating system is distributed onto 2 or more of the disks on the RAID controller (Indeed, often a entire subsystem of RAID controllers form an array).

The chances of the partition failing relies on the type of RAID configuration more than any other variable. A RAID 0 array is far more likely to fail than, say, a RAID 1 or 5, etc..

Let's take a disk, let's call it the Spinright 60 Gig drive. After some testing it's determined that 5% of the disks fail to work after a year (which means 95% work after a year.)

In a RAID 0 array, all of the disks have to be operable for the array to be readable. so in a 2 disk RAID 0 array of Spinright disks, the probability or the partition being readable and usable after 1 year is (0.95)*(0.95) = 0.9025 = 90.25%. As you can see, the chance of it being readable after a year is less than if the partition was on one disk.

In a "proper" RAID array, it's a different story. Let's take a 2 disk RAID 1 array of Spinright disks. in a RAID 1 array, only one disk has to be readable, so the probability of the partition being readable is the chance of at least one disk being readable, so for our 2-disk array, it's the probability that both disks are readable plus the chance that only one of the disks is readable,

i.e.

(0.95*0.95) + 2*(0.95*0.05) = 0.9975

= 99.75%

In other words, it's got only a 0.25% probability of failing within a year. Put another way, the partition on the RAID 0 array has a 9.75% probability of dying and the partition on the RAID 1 array has a 0.25% probability of dying. Therefore, the data on the RAID 1 array is 39 times less likely to fail compared to the RAID 0 array and 20 times less likely to die than the data if it had been on a single disk.

There are more efficient and safer configurations for a RAID array than a RAID 0 or a RAID 1 array. For example, a RAID 5 array (which needs 3 disks) is a set of similar disks where the data is read in sections, to which a parity dataset is created and written to the array, spread across all the disks, along with the source data. The partition will be readable if any one of the disks should fail. In a 3-disk RAID 5 array, the probability of the partition being readable after a year is the probability of one or less of one our Spinright disks dying within a year, i.e.:

0.95*0.95*0.95 + 3*(0.95*0.95*0.05) = 0.99275 = 99.275%.

One of the beauties of RAID is you can have nested RAID arrays within a RAID array, E.G. a RAID 51 array is a mirrored pair of RAID 5 arrays. For that to fail both RAID 5 arrays have to simultaneously fail, which is, to put it lightly, unilikely. Of course, all the mirroring in the world won't help if the computer it's on gets hit by a cluster bomblet.

The most stable safest data is the type stored on large highly redundant RAID's like this then mirrored in different locations with tape backups. For example, the booking history of the last 15 years for British Airways is stored near Heathrow Airport in a crash-proof bunker on a 256-disk (10 Gig disks) custom array, then double-mirrored in two locations, one in Scotland, one in Germany.

One reason for a RAID is you can make extremely large partitions, extremely fast ones or extremely redundant ones. If you need none of these, don't bother. Arrays use up more power and cost a hell of a load more than a single disk. What I do to back up stuff is to put all the stuff I need (like setup files and critical files) onto an older smaller hard disk, unplug it and put it on a shelf.

[Como]

and then there was silence. i'm considering rad myself, i personally think 5 is the best for speed/reliability. is it just me, or are drives less likely to live long these days? i still have a 6.4 Gigger running....never gave me a problem i diddnt deserve. (ntfs converting to fat32 upgrading form 2000 to 98 doesnt end well... heh )

now i have a 120 gb hard drive, and i keep reading upwards of 10% failures?! pardon me, but if i have a 10% failure rate, wouldny my disk be dead in a month? gee.

[Xaotic]

RAID-5 will normally kill your write performance. This is due to the parity and stripe calculations that are done to the data. The latency level can be moderate to horrible depending on which controller you are using and where the pairty calculations are occurring. Reads are a bit better, but there is still a disk access penalty.

[AlucardCasull]

heat too, HD's in raid make more heat than just normal hd's because they are both running when you have a program being loaded instead of just one. a decent fan in front of them though should be fine.

[Cjwinnit]

Originally posted by AlucardCasull
heat too, HD's in raid make more heat than just normal hd's because they are both running when you have a program being loaded instead of just one. a decent fan in front of them though should be fine.

True, but the difference in power consumption between 2 seeking hard drives and 2 hard drives where one is spinning but not being accessed and one seeking is going to be pretty small..

[Como]

really? i thought it was bett for writes. well than, maybe i'll just have to get a raptor and run it as normal...