A RAID zero question/comment

[Gonk42]

I realise that this isn't Mac Pro specific, but a lot of Mac Pro users
also use RAID.

The general comment I've seen on these forums is that with RAID 0 either
disk going down takes the lot (I'm talking two disk for simplicity, of course
it can be extrapolated) so you're drives are twice as likely to fail.

Thinking about it, isn't it worse than that? My thoughts go along these
lines; with every file striped across two disks every read and write operation
will involve both disks (even though it will only be half the bytes per disk).
So not only will you have the factor of two from the either disk failing
issue but each disk will have more stress.

A particular example scenario would be reading data from one file, processing it in some way and then writing it out to another file. If you have the first file on one disk and the second on a different disk then the drive heads will not need to move much. If both disks are combined in RAID 0 then both sets of drive heads will have to jump too and fro between the two files. (I realise that both files will often be on the same disk in a non-RAID setup but I think the point is still valid for a proportion of the time.)

I am no expert, but am interested in comments from those who do know about RAID.

[sammich]

After making sure of my logic by reading the wikipedia article on RAID 0:

A RAID 0 setup will split data in very small chunks, so unless your data is smaller than 512 bytes, it will reside in portions on both drive. That said, if you consider that both drives are 'blind', they both get the command to read a particular file, they will serve up that data as if their half portion is a complete file, only the RAID controller would know what's really happening. So then each drive would be doing as much work as any other single drive setup.

Say you have, 2 disks in RAID 0, and a single file 'A' in 4 parts, A1, A2, A3, A4:
Disk 1: A1 | A3 (A3 immediately follows A1, no gaps)
Disk 2: A2 | A4

A request for file A will mean that the RAID controller will tell both drives to get File A. The controller will get A1 and A2 more or less at the same time, and sends it to the computer. As soon as A1 is completely read, the drive head will sit at the end of A1, ready to read A3, likewise for Disk 2. A3 and A4 will be sent in the same manner to the computer. Now if you look at the point of view of one of the disks, it looks like they are just giving all of A, like any single drive.

[Gonk42]

What you say is correct, but what I was thinking of is that every file will
essentially be two files (one on each drive as you say) so that no matter
which file is accessed both drives will be involved.

I don't know what stresses drives but I would guess it is the seek operation
more than the contiguous read part. If all files were contiguous (or each
half-file in the RAID case) then there would be twice as many seek operations
occurring in the RAID case as in the non-RAID case so the chance of one
drive failing would be doubled (in my rather simplified model) which would
mean that the chance of failure would be quadrupled for any particular drive
(since the death of the other drive will also kill it).

It was just a random thought I had whilst trying to plan my own drive
set up (for a workstation which I've not yet bought).

[whatisthe]

The drives are probably slightly more stressed, hence the more expensive enterprise drives that have higher vibration tolerance and reliability.

Even in NAS/raid units with only a few drives they recommend using enterprise grade drives.

[Tesselator]

Originally Posted by Gonk42

I realise that this isn't Mac Pro specific, but a lot of Mac Pro users
also use RAID.

The general comment I've seen on these forums is that with RAID 0 either
disk going down takes the lot (I'm talking two disk for simplicity, of course
it can be extrapolated) so you're drives are twice as likely to fail.

Nope! Absolutely NOT!

Thinking about it, isn't it worse than that? My thoughts go along these
lines; with every file striped across two disks every read and write operation
will involve both disks (even though it will only be half the bytes per disk).
So not only will you have the factor of two from the either disk failing
issue but each disk will have more stress.

Nope! You double the total capacity so each drive has the SAME as before!

No extra "stress" is put on anything! The drives will last the SAME as if there were 1 or 10!

This is fools logic you're using here. Just think about it with common sense in mind.

A particular example scenario would be reading data from one file, processing it in some way and then writing it out to another file. If you have the first file on one disk and the second on a different disk then the drive heads will not need to move much.

Huh? No, that doesn't even make sense.

If both disks are combined in RAID 0 then both sets of drive heads will have to jump too and fro between the two files. (I realise that both files will often be on the same disk in a non-RAID setup but I think the point is still valid for a proportion of the time.)

Nope!

I am no expert, but am interested in comments from those who do know about RAID.

This reads like troll bait to me. It doesn't make sense, it's recently been covered here about 4 or 5 times over - in GREAT detail, and the assertions are not questions one would ask if they wanted to learn something. They're like, i dunno, unfounded assumptions that are the opposite of everything that's been discussed here in the past month or two.

I don't get it. <shrug>

[ncc1701d]

If you did have 4 drives in RAID 0 (and appropriately backed up) and one of the 4 drives did fail - you can then start up again from the backed up files.

My question is: Before you can replace the data you have to set up drives again, so once you have determined which drives failed, can you replace that drive that failed with another similar drive and add it to the remaining 3 that were working fine, wipe them all and start a new RAID 0?

[jazz1]

I've been running a 2.66 in a RAID0 with 4 Seagate 500GB drives for over a year and I have never had a hiccup.But my first defense against lost is Timemachine on a 2TB RAID1. Wish I had a DROBO.

I used run an old Mac desktop with a 2 drive RAID over SCCI. Now that was always having to be rebuilt.

Currently, I am also using Backblaze in the "cloud" backup service. Oh, yeah I also use another service that just backs up my music online.

Am I paranoid because the RAID0 is likely to fail more? No, because all drives will fail. It is just a matter of when.

[sammich]

Originally Posted by Tesselator

Nope! Absolutely NOT!

Uh...yeah. You're getting mixed up with RAID 1. Which has REDUNDANCY.

Nope! You double the total capacity so each drive has the SAME as before!

What's that got to do with anything you quoted?

No extra "stress" is put on anything! The drives will last the SAME as if there were 1 or 10!

Each drive is an individual. Two drives in RAID 0 will increase the chance of the RAID 0 setup of catastrophic failure by a factor of 2. It's not how long each drive lasts.

This is fools logic you're using here. Just think about it with common sense in mind.

I think you need to re-read what you've said.

[sammich]

Originally Posted by Gonk42

I don't know what stresses drives but I would guess it is the seek operation more than the contiguous read part. If all files were contiguous (or each half-file in the RAID case) then there would be twice as many seek operations occurring in the RAID case as in the non-RAID case so the chance of one drive failing would be doubled (in my rather simplified model) which would mean that the chance of failure would be quadrupled for any particular drive (since the death of the other drive will also kill it).

You're still thinking about it wrong. Sort of. Yes, there are twice as many seek operations per RAID 0 setup. But the number of seeks per drive remains the same, THUS, each drive would only be under as much stress as if it were being used by itself.

[almostinsane]

Originally Posted by Tesselator

Nope! Absolutely NOT!

Dude you have no idea what your talking about.

[Tesselator]

Originally Posted by sammich

Uh...yeah. You're getting mixed up with RAID 1. Which has REDUNDANCY.

Nope. We're talking about RAID level 0 here. I know very well what I'm saying.

Each drive is an individual. Two drives in RAID 0 will increase the chance of the RAID 0 setup of catastrophic failure by a factor of 2. It's not how long each drive lasts.

Nope! It doesn't. This is wrong. It's "popular" to think so but it is wrong.

I think you need to re-read what you've said.

Mmm, no. I know what I said.

[Tesselator]

Originally Posted by almostinsane

Dude you have no idea what your talking about.

Prove it.

You can't so I'm not worried.

[Tesselator]

Originally Posted by sammich

You're still thinking about it wrong. Sort of. Yes, there are twice as many seek operations per RAID 0 setup. But the number of seeks per drive remains the same, THUS, each drive would only be under as much stress as if it were being used by itself.

Yes. This is right for the most part. It's more complex but basically this is right.

[iHateMacs]

Loosing data doesn't concern me too much on my raid 0. It's purely for transient Final Cut work. My working project file is saved to another drive and all the video/audio files have been copied to the raid from somewhere else anyway. If it goes down then it's a slight inconvenience especially as I'm not in business.

I have a raid 0 and a raid 1 on the same pair of drives. First partition on each drive is for my raid 0. Small partition to help keep data to the faster outer edge of the disk.

The remainder of the disks is setup as a raid 1 mirror for longer term data.

[QuantumLo0p]

You could go crazy trying to calculate failure frequencies and I wouldn't bother.

IMO, I suppose if you are consumed by failure rates, about the only factor I would consider is a very basic view that the more raid0 drives you have the higher the chance of a failure. Start with new drives, build your system, use a good backup/recovery strategy then relax and have a home brew. Other factors I would not loose any sleep over.

[Tesselator]

Originally Posted by QuantumLo0p

You could go crazy trying to calculate failure frequencies and I wouldn't bother.

IMO, I suppose if you are consumed by failure rates, about the only factor I would consider is a very basic view that the more raid0 drives you have the higher the chance of a failure. Start with new drives, build your system, use a good backup/recovery strategy then relax and have a home brew. Other factors I would not loose any sleep over.

Yes I agree. And by "failure" of course we know you mean a retry failure that usually never happens on drives under 3 to 5 years old and only (usually) affects a single file.

The truth is just like you say. It's the same deal as a single drive for all but the over analytical navel gazers (or where HIGHLY secure systems are required - [like in a bank or something]). Use good, new, healthy drives, keep a backup, and just enjoy the speed.

[sammich]

Originally Posted by Tesselator

Prove it.

You can't so I'm not worried.

If you would be so kind as to explain why, in a pure RAID 0 setup, if one of the disks goes down, it wouldn't cause a total loss of data in that array.

Each drive is an individual. Two drives in RAID 0 will increase the chance of the RAID 0 setup of catastrophic failure by a factor of 2. It's not how long each drive lasts.

And why is this not true?

ALSO: unless you're referring to probability vs. possibility vs. likelihood.

[Tesselator]

Originally Posted by sammich

If you would be so kind as to explain why, in a pure RAID 0 setup, if one of the disks goes down, it wouldn't cause a total loss of data in that array.

The misstatement I was objecting to was: "...so you're drives are twice as likely to fail." which was the predicating part of his proposition that made the entire statement false.

And why is this not true?

Think of a RAID set as a single unit - which is how it behaves. It's not twice as likely to fail because you have two nor three times if you have 3. By that logic my 16-drive RAID 0 in 16x16 RAID 10 configurations (32 drives each) would fail about once a month. They don't. They "catastrophically" fail about once in 5 or 6 years in actuality - just like single drives. MTBF. Mean Time Between [suggested] Replacement is also still approximately 3 years - just like a single drive.

EDIT: While "weakest link" logic does indeed apply I propose to you the question: Does a chain with 200 links break 100 times more often than a chain with 2 links? Or does it last the typical life of "a chain" when cared for and used properly?

ALSO: unless you're referring to probability vs. possibility vs. likelihood.

That's an incomplete sentence so I have no idea what you mean. Yes, those three words mean about the same thing. Huh?

.

[sammich]

Originally Posted by Tesselator

The misstatement I was objecting to was: "...so you're drives are twice as likely to fail." which was the predicating part of his proposition that made the entire statement false.

Think of a RAID set as a single unit - which is how it behaves. It's not twice as likely to fail because you have two nor three times if you have 3. By that logic my 16-drive RAID 0 in 16x16 RAID 10 configurations (32 drives each) would fail about once a month. They don't. They "catastrophically" fail about once in 5 or 6 years in actuality - just like single drives. MTBF. Mean Time Between [suggested] Replacement is also still approximately 3 years - just like a single drive.

Ah, I see your logic now. But for starters, this is a basic RAID 0 setup, so there is 0 redundancy. Your RAID 0+1 setup is mirrored, the only way for that to catastrophically fail would to be if the pair of drives that mirror each other were to fail at roughly the same time.

So you're talking about the 'likelihood' of such a drive failure. Yes, if you are given the typical drive, chances are you will never see it have any problem until 6 years down the track. BUT there is that small chance that something could happen and you lose everything. Adding the second drive will double that small chance. Still a small chance, but still double a single drive nonetheless.

That's an incomplete sentence so I have no idea what you mean. Yes, those three words mean about the same thing. Huh?

Mathematically speaking: probablility is the 'chance' that an event will occur. Possibility if an event is going to happen (ie drive failure/not drive failure is 50%). And likelihood is my way of saying from past experiences and general consensus, the chance that something will happen, ie: realistic.

Probability a drive will fail after 1 month: 1/10^9
2 drive RAID 0 setup: ~2/10^9
** now that means nothing to the average person all they understand is:
Possibility:
Complete data loss after 1 month: 50%. It either happens or it doesn't, but that's non-sensical too
** but in all likelihood, the drives will last at least 4 years before one fails.

[Tesselator]

Originally Posted by sammich

Ah, I see your logic now. But for starters, this is a basic RAID 0 setup, so there is 0 redundancy. Your RAID 0+1 setup is mirrored, the only way for that to catastrophically fail would to be if the pair of drives that mirror each other were to fail at roughly the same time.

Yes, I know. All my comments were pertaining to RAID 0.

[sammich]

EDIT: While "weakest link" logic does indeed apply I propose to you the question: Does a chain with 200 links break 100 times more often than a chain with 2 links? Or does it last the typical life of "a chain" when cared for and used properly?

Why yes, if each link in the chain had a small but discreet probability of failure.

A billion drives in RAID 0 is almost guaranteed to fail in, say, a month.

[Tesselator]

Originally Posted by sammich

Why yes, if each link in the chain had a small but discreet probability of failure.

Nope!

A billion drives in RAID 0 is almost guaranteed to fail in, say, a month.

Prove it.

[drumhum]

Tessalator,
Are you really saying that a RAID 0 is not twice as likely to fail than a single disk?

You have pointed out the apparent misunderstand of raid 0 reliability but not explained much.

You have really gotten me thinking about this whole probability of failure for raid0 thing. I've registered here because of what you have written.

Lets think in terms of probability and chance...

I have thought of the situation compared to using dice. Consider throwing a 4 as a drive failure. I have 1 in 6 chance of a four for every throw. A one in 6 chance of drive failure.

If I have two dice I now have 36 combinations (1,1; 1,2; ... 6,5; 6,6).
Out of those combinations, twelve contain a 4 (1,4; 2,4; ... 6,4 and 4,1; 4,2; ... 4,6)
So, I know have a chance of throwing a 4 of 12 in 36. That is, 1 in 3.

My probability has halved. I am twice as likely to throw a 4. I am twice as likely to experience a storage failure.



Now I know "probability" is a confusing subject and I've certainly been a victim of misunderstanding. Please can you explain why I am not twice as likely to have a 2 disk raid 0 fail on me?

You have certainly made me think about all this.

[Tesselator]

Originally Posted by drumhum

Tessalator,
Are you really saying that a RAID 0 is not twice as likely to fail than a single disk?

Yes! This is as obvious as 2+2=4

The warrantee on a drive is 3 or 5 years or whatever. And they do not fail (typically) before then. It might be said that a single drive is likely to fail in 3 to 5 years. What the initial poster said means to me that he's trying to say that in a 2-drive raid 0 it's likely to fail in 1.5 to 2.5 years, and a 4-drive raid is likely to fail in 9 months to 15 months.

This is, of course, absurd and ridiculous!

The only argument in this thread which even seems sensical actually is incorrect as well. This is the Sammich proposition that "there is that small chance that something could happen and you lose everything. Adding the second drive will double that small chance." It might seem correct to say that:

Month 1 = 0.1% chance to fail in single drive. (1 in 1,000)
Month 1 = 0.2% chance to fail in 2-drive raid. (2 in 1,000)
Month 1 = 0.3% chance to fail in 3-drive raid. (3 in 1,000)

and after that:

Year 1 = 0.001% chance to fail in single drive. (1 in 100,000)
Year 1 = 0.002% chance to fail in 2-drive raid. (2 in 100,000)
Year 1 = 0.003% chance to fail in 3-drive raid. (3 in 100,000)

Year 2 = 0.01% chance to fail in single drive. (1 in 10,000)
Year 2 = 0.02% chance to fail in 2-drive raid. (2 in 10,000)
Year 2 = 0.03% chance to fail in 3-drive raid. (3 in 10,000)

and etc. But this is actually mathematically incorrect when the myriad of other variables are considered as one would, who was actually concerned with such detail. Like the fact that being in a raid set actually reduces and distributes workload. Such social variables as owner concern and upkeep that comes into play differently than for the majority of single drive users. Environmental variables like backup, recovery, faulty upstream components, and differing internal error handling. Natural variables like lightning strikes, floods, and line surges which are not changed by any calculable degree from having more or less drives in the system.

But who the heck cares about all that? It's just not going to happen so why even discuss it? If it does happen it happened within the same relative odds as a single drive or any sized raid set one of us here is likely to put together (2~8). The topic itself is retarded unless something VERY VERY important is being stored on there - like tens of millions of dollars worth of transactions - and then one only needs to consider such things so they can design proper system security to avoid the disaster from happening at all - ever.


It's navel gazing and it's not even fun navel gazing as there's no pay-off from doing so - implied or actual. QuantumLo0p had it exactly right when he said:

"You could go crazy trying to calculate failure frequencies and I wouldn't bother.

IMO, I suppose if you are consumed by failure rates, about the only factor I would consider is a very basic view that the more raid0 drives you have the higher the chance of a [file error type] failure. Start with new drives, build your system, use a good backup/recovery strategy then relax and have a home brew. Other factors I would not loose any sleep over."

Yup! A new chain lasts X number of years in a given environment for certain uses. It doesn't matter if there are 2 or 100 links.


.