2x80Gb RAID0 or 1x320Gb?

[Agent]

If you're an A/V pro who needs fast throughput

I mainly do high end 3D rendering, so along the same lines as A/V pro's I guess?

then do you think that the savings on not setting up a proper RAID5 array are really worth having to spend hours hoping you can resurrect the drive containing a project that needs to be submitted tomorrow? (And, even if you do resurrect it, you have to hope that there's not going to be a big glitch sitting in the middle of your show-reel).

For me? Yes. Why? Because quite simply I don't have the money for a RAID5 array.
While I'd like one as much as the next bloke, being a student I just cant justify the cost of a RAID5 system, more so when data from my RAID0 array is backed up nightly (or more if I choose)

Also RAID0 has higher writing rates then RAID5. Considering that some of the things I render can cause the HD I/O to be the bottleneck, it all helps
Don't get me wrong, I'd drop the extra writing speed (Reading isn't that different from what I've seen in real world situations) for RAID5 tomorrow, but I just can't justify the cost. Its not just the drives, but the appropriate card/motherboard with RAID5 support which are generally much more expensive than their RAID0 counterparts too.

Resurrecting backups takes time, and it assumes that you have proper backups, which is certainly not the general case for home users. Pro users who backup every day are basically gambling that the disks won't fail right after a costly 10-hour editing session.

But like I said, RAID0 for a home user is insane.
The analogy on the disk failing front though; the same could also happen on a single disk. While the probability certainly goes up with RAID0, regardless of if its a single disk that fails or a RAID0 array, the outcome is the same: Get the backups ready
Even if you do a worse case probability on it and say there is a double chance of it failing, knowing that is the first step in evading any issues if it does happen.

WHERE do you see RAID0 as having a place? This is the obvious question.

Easy: If you are budget limited and the risks of lost data outweigh the gains of what's (almost) double the throughput.
I fall into that, but I appreciate that most people wouldn't.

Even with no pro applications, some enthusiasts still do RAID0 just for that 'little bit extra' (Or Epenis++ if you want to think of it that way ) and avoid the double probability of failing scenario by only installing the OS and games to it.
For most of these people, reinstalling the OS and software is trivial.

I've gotta ask though mate, if you couldn't ever see a useful application of RAID0, why were you running it?

[charleski]

I've gotta ask though mate, if you couldn't ever see an a useful application of RAID0, why were you running it?

Because I used to think like you, but now I've realised I was wrong
Simple as that, really.

[Agent]

Or maybe out situations are different?

[charleski]

You're just balancing the cost of a RAID5 array against the cost of lost production time due to creating a backup and (potentially) having to restore it. There is also the cost of lost work that happens in between backups to consider, and that can be very expensive.

At best, you're making the case that RAID0 may be effective when your labour costs are negligible and you have no time constraints for delivery. In the majority of situations this is not true for anything beyond the baby's birthday party. Amateurs still have to deliver quality product on-time in many cases.

Sorry if I'm being a bit harsh, but I do think that RAID0 is a bad decision, and that's an opinion fueled by experience.

[mroz]

An interesting discussion which unfortunately demonstrates how easily swayed I am. Now I'm back to being undecided.

Two drives 500GB each; desktop usage & some video work.

Inclined to think I'll RAID 0 the first 20% of each drive & partition that: 150GB scratch & 50GB OS/Apps. Not sure whether to RAID 1 the remaining or keep it as two 400GB drives.

Also need to sort out some kind of external backup via eSata, when I can afford it - any preferred backup software for XP?

Should also look into finally using SMART after all these years of ignoring it, to catch some impending failures.

[aidanjt]

WHERE do you see RAID0 as having a place? This is the obvious question.

On Gentoo I use RAID0 on my server box for my /var/portage directory as well as /tmp & /var/tmp, these directories have no need for fault tolerance at all, but rather performance, and that's where RAID0 shines. There are also other applications where you don't need redundancy, and a simple backup policy will do.

And the whole raid0 = failure_rate * n_disks assumption has already been beaten to death and dismissed on these forums. 1 disk has an equal chance of bricking itself as 1 of a 2 disk setup.

[mroz]

And the whole raid0 = failure_rate * n_disks assumption has already been beaten to death and dismissed on these forums. 1 disk has an equal chance of bricking itself as 1 of a 2 disk setup.

Wtf? Unless you're implying in an n-disk set up the chance of disk failure is highly correlated that must be wrong. Of course there will be some degree of correlation, since some failures will have external causes, potentially resulting in multiple drives being taken out (hence the reason RAID 1 isn't an alternative to external backups).

Still, it's surely approximately correct that the chance of any given disk failing is independent?

Obviously the chance of failure of any one of a group of drives isn't proportional to the number of drives - that's just gibberish. However for independent failures & low failure rate it is a good approximation:

For n independent events each of probability p, the chance of one or more such events is clearly
1-(1-p)^n. For small p this is approximately np.

What part of the beaten to death argument have I missed?

[aidanjt]

You missed the part that RAID arrays are built upon batches of disks, it goes without saying that hardware is manufactured in batches and thus have a fairly equal probability of failing given the same manufacturing process. By making (bad) assumptions on failure rates, you risk your 1 lone disk (and it's data) getting screwed as surely as one disk in a RAID0 array does.

[this_is_gav]

WHERE do you see RAID0 as having a place? This is the obvious question.

If you're an A/V pro who needs fast throughput, then do you think that the savings on not setting up a proper RAID5 array are really worth having to spend hours hoping you can resurrect the drive containing a project that needs to be submitted tomorrow? (And, even if you do resurrect it, you have to hope that there's not going to be a big glitch sitting in the middle of your show-reel).

I'm not an A/V pro (or even mildly close), yet if I didn't have 2 sets of striped Raptors and relied on my single 7200.10 at Christmas, I'd have still been going when the holidays finished. Oh, and the 7200.10 failed anyway, as did my other one, so from my point of view I had much more to lose on the Seagates, as none of my 6 Raptors are showing signs of dying yet (touch wood).

That said, I would still recommend a standard WD or Samsung drive. They're quieter, not slow by any means, and do everything using just that little less power. You'd struggle to notice unless you did dabble in something which required sustained data rates. Certainly loading game levels is no faster on my RAID0 Raptors than a single 250gb WD.

[peterb]

If you are budget limited and the risks of lost data outweigh the gains of what's (almost) double the throughput.

And that really is the heart of the matter. In implementing a RAID 0 array, you need to be aware of the risks against the gains and take appropriate measures to reduce the risk of total data loss to an acceptable level. (The same goes for a single drive solution too - just that the risks are a little less.)

As an aside, while you are getting double throughput, I presume that that is with a dedicated controller card. If so, I would guess the increase wouldn't be as significant with a software solution using the facilities built into a mobo.

[mroz]

You missed the part that RAID arrays are built upon batches of disks, it goes without saying that hardware is manufactured in batches and thus have a fairly equal probability of failing given the same manufacturing process. By making (bad) assumptions on failure rates, you risk your 1 lone disk (and it's data) getting screwed as surely as one disk in a RAID0 array does.

I think I see our point. I take it you're claiming for a subset of a given batch of disks, they're all very likely to fail at roughly the same point in their lifespan? If so, got any data or references to back this up?

Language aside: the way you've phrased the above is a little confusing. All the disks could have the same probability of failure & this wouldn't give you the result you're claiming; rather, that requires the actual failures themselves occur at similar times, so that the expected first failure of a group of n is roughly the same as the expected first failure of a single.

[aidanjt]

I think I see our point. I take it you're claiming for a subset of a given batch of disks, they're all very likely to fail at roughly the same point in their lifespan? If so, got any data or references to back this up?

No, nothing at hand, simple logic suffices, some experience, and knowledge of the principle of mass production, and hey presto.

Language aside: the way you've phrased the above is a little confusing. All the disks could have the same probability of failure & this wouldn't give you the result you're claiming; rather, that requires the actual failures themselves occur at similar times, so that the expected first failure of a group of n is roughly the same as the expected first failure of a single.

What's confusing about 1 disk having an equal chance of failure as 1 of 2 disks?.. Most sensible network administrators replace a set after a single disk failure for the simple reason that I outlined.

[TooNice]

I voted for single disk. If the option was 2x160GB, and space is not an issue, then I would pick that option instead. I see some use in keeping data in two physical drives even though it is possible to just partition the 320GB drive. Perhaps I should've voted for the last option ;P

[mroz]

What's confusing about 1 disk having an equal chance of failure as 1 of 2 disks?.. Most sensible network administrators replace a set after a single disk failure for the simple reason that I outlined.

That isn't what you said & at that point in my post I was writing about how you'd stated your claim, not merely what you were claiming.

You actually stated '...and thus have a fairly equal probability of failing...' from which you inferred a set has the same chance of failure as a single. That reasoning is simply wrong.

I deduced what I thought you meant (which you seem to have confirmed as correct) & stated it in a way that is I hope logically consistent:

'...failures (are likely to) occur at similar times' from which it is possible to infer your original claim, which incidentally seems to be based on experience both first & third hand.

I'm still interested in anyone having data to back this claim up, as anecdotal evidence is just that & can't alone be conclusive regardless of the expertise & depth of experience of the source.

[aidanjt]

That isn't what you said....

Yes it was:

And the whole raid0 = failure_rate * n_disks assumption has already been beaten to death and dismissed on these forums. 1 disk has an equal chance of bricking itself as 1 of a 2 disk setup.

[mroz]

I was commenting on your reasoning in a particular post in order to better understand & clarify what you were trying to say.

The line you've quoted above isn't even from the same post & has nothing to do with the point I was attempting to make.

Never mind. I think I eventually understood you. I've nothing more to say on this in the absense of any evidence of what was originally an interesting assertion.