2x80Gb RAID0 or 1x320Gb?

[iranu]

I think you are arguing for the wrong reason or are arguing two seperate points that are subtely different. The failure of the ARRAY/windows is the important thing to remember. The array relies on two disks.

let us assume identical disks in a 2 disk array - raid 0. The probability of failure of the ARRAY/Windows is

1-(1-p)^n which was correctly stated above. Let us assume that the probability of failure p = 2% with regard to mean time to failure, then
1-(1-0.02)^2= 0.0396 or just under 4%.

Now lets look at a single disk which has windows on it. Well the probability of failure is 2%. So yes the probability of losing windows or data on a 2 disk raid 0 set up is twice as likely, however, I look at it slightly differently.

I say that because you don't know when the damn thing is going to fail! What matters is you back your data up externally regardless of how many disks you are using.

I used raid 0 on my old rig for two reasons. I thought it would be faster and I'd never set raid up before so I was doing something new. I'm now contemplating going back to it because load times in gaming it just doesn't feel as fast which at the moment is more important to me. The problem is my "feel" is highly subjective! It certainly was quicker with regard to shifting large files and writing to the array. Is it worth it? Well I still don't know! I sort of want to try it again but I'm realising that I can just shove a new disk in, lob windows on it, transfer the data and take out that 160g drive and bingo it's a backup! Then I can format the other 160G and use that for backup or in a an older machine for HTPC.

Does leaving a drive unused with data on it ever cause problems due to the period of time unused? I can't see how.

[mroz]

Let us assume that the probability of failure p = 2% with regard to mean time to failure, then
1-(1-0.02)^2= 0.0396 or just under 4%.

Now lets look at a single disk which has windows on it. Well the probability of failure is 2%. So yes the probability of losing windows or data on a 2 disk raid 0 set up is twice as likely, however, I look at it slightly differently.

Aiui Aidanjt's assertion was that the above is incorrect for typical RAIDs as the probabilites for disk failures aren't independent when the drives are all from the same batch; further, he claimed failure times are likely to be very similar for such drives, leading to little difference in expected failure time for a single drive vs expected time of first failure for the array.

This would have a big impact on the argument against choosing RAID 0, however I've still to find any proof of this claim.

[aidanjt]

Aiui Aidanjt's assertion was that the above is incorrect for typical RAIDs as the probabilites for disk failures aren't independent when the drives are all from the same batch; further, he claimed failure times are likely to be very similar for such drives, leading to little difference in expected failure time for a single drive vs expected time of first failure for the array.

This would have a big impact on the argument against choosing RAID 0, however I've still to find any proof of this claim.

You might find this interesting, I'm too tired to read it all, but it's a research paper on disk failures from Carnegie Mellon University in the US.

This sticks out, however:

A bad batch can lead to unusually high drive failure rates or un-usually high rates of media errors.

[iranu]

Aiui Aidanjt's assertion was that the above is incorrect for typical RAIDs as the probabilites for disk failures aren't independent when the drives are all from the same batch; further, he claimed failure times are likely to be very similar for such drives, leading to little difference in expected failure time for a single drive vs expected time of first failure for the array.

This would have a big impact on the argument against choosing RAID 0, however I've still to find any proof of this claim.

The probability of disk failure is independant, no doubt about that. As for batches I can testify that there is always going to be a Gaussian/normal distribution with regard to a batch and it's failure time. You may even get that disk that's an outlier. Manufacturers always look to provide quality and in the case of something like a hard drive then the same process will be used over and over. Any process engineer worth his salt will ensure that the difference between batches is at an acceptable level. This is usually tested via some form of quality control that will use a sample determined by a standard. eg BS 6001-7:2006. Sample size tends to decrease the further the process is used.

What does start to become more difficult is when you start to use products from different manufacturers to make your product. But again there is statistical analysis to ensure such parts/products appear within the required specifications. Storagereview.com has a reliability survey which can act as a guide but you have to register on that site to see it.

I've got about 4 1/2 years experience with regard to material (metals, polymers, ceramics etc) property data, the generation and management of the test programs needed to provide that data and analysis of the data generated for a major aerospace firm. As well as lobbing it into a database that must be 40 years old! Well that's part of my job now.

Basically what we do is provide the stress/cad/design engineers with the data (strength, fatigue, crack propagation, creep etc) so they can design the parts.

[mroz]

The probability of disk failure is independant, no doubt about that.

Read the report Aidanjt linked to. Some of the analysis went over my head but I picked up quite a lot of it. Usually it's assumed they are approximately independent & that the chance of failure in a given period of time is constant, leading to the mtbf having an exponential distribution. That report refutes this. Here's just one interesting observation:

A second differentiating feature is that the time between disk replacements in the data exhibits decreasing hazard rates. Recall from Section 2.4 that the hazard rate function measures how the time since the last failure influences the expected time until the next failure. An increasing hazard rate function predicts that if the time since a failure is long then the next failure is coming soon. And a decreasing hazard rate function predicts the reverse...

...Note, that the above result is not in contradiction with the increasing replacement rates we observed in Section 4.2 as a function of drive age, since here we look at the distribution of the time between disk replacements in a cluster, not disk lifetime distributions (i.e. how long did a drive live until it was replaced).

Aiui this is indicative of clustering of failures within a group.

[mroz]

You might find this interesting,

Thanks. Read a fair chunk of it & understood about half the maths & much of the conclusions. Interesting indeed. I think the comments regarding hazard function explain the behaviour of sys admins replacing drives en mass which you described bicbw.

[aidanjt]

Not a problem.

Nothing annoys me more than bad statistical formulas. I'm glad I found a paper that's based on real world data, rather than some half-baked math.

[Thorsson]

So, are we agreed that increased chance of disk failure is not a big determining factor in choosing whether to go RAID0?

[aidanjt]

I vote aye. Due to the fact that the failure rate isn't proportional to 'n' disks in a given batch, but rather batch quality, disk age, and thermal environment.

[Tetras]

I think we can safely agree that reliability is not a reason to choose RAID0 but may also not be a reason to stop you

If (some of/all) the data on the disks is that important I would imagine you'll keep backups.

[aidanjt]

'Reliability' isn't a factor at all, RAID0 merely has no redundancy in the case that a disk does go, the same is true for single disk setups. No matter what volume setup you use (even RAID5 or RAID6), you still need some kind of backup policy, a lot more things can go wrong than disk death.

[Thorsson]

So the choice must be around whether the (smallish) increase in performance is worthwhile to you as an individual. And that is going to depend very much on what you use your computer for and how much money you are prepared to spend.

A more interesting question then is whether, if you have two 500Gb drives to play with, you are better going with RAID or a split usage policy - purely from the PoV of performance.

[aidanjt]

Well, as far as performance goes, RAID0 will perform much better in sustained read/write operations, for random I/O the performance improvement will be nominal, as that is dependant on access/seak time (the time it takes for the read/write head to get into position and the appropriate sector comes into alignment and start I/O), rather than raw throughput.

[oralpain]

My vote is on one single, recent drive that has a 250GB or larger platters. Bigger platter = more performance.

Several 7,200 rpm drivers released or revised in the last few months match the newest raptor performancce in most areas, cost less, and hold more.

[aidanjt]

Several 7,200 rpm drivers released or revised in the last few months match the newest raptor performancce in most areas...

A common misconception.

[Zak33]

Several 7,200 rpm drivers released or revised in the last few months match the newest raptor performancce in most areas, cost less, and hold more.

thats not strictly true.

A Raptor will muller anything. EXCEPT...cos they're so small, as they fill up they get slower.

But kept as a clean boot drive...they're very fast indeed.

But expect to be able to use it for XP, plus apps, plus all your games...they get slow