RAID 5 Two Drives in RAID 0?

SolidSnakeUS

[H]ard|Gawd
Joined
May 29, 2008
Messages
1,282
I got a question, someone suggested it to me, since I am getting 3 new WD 640's, how about if I do this:

1 Seagate 320GB 16MB (Perp Record) as the main drive
2 WD 640's in RAID 0 for storage
1 WD 640 in RAID 5 of the RAID 0 drives, so it can back them up

Can that be done and is it a good idea?
 
You need a minimum of 3 drives for R5. Besides, you never, ever, ever, ever use R0 for storage.

Besides, how are you going to backup 1280GB of data onto 640GB of space? :confused:
 
It would probably only be for the first 640GB, something like that. Sorry I've been suggested to do this. I'm still torn on what to do, and I want exactly 3 640's, but I'm still unsure of what to do with them. I mean, hell, if I wanted to, I could just use the drives separately and it would be faster than any of my current drives lol.

This is what my computer looks now:

1 Maxtor 80GB 8MB Cache
1 Maxtor 160GB 8MB Cache
1 Seagate 320GB 16MB Cache
1 External WD 250GB 8MB Cache

This is what is going to be inside the computer:

1 Seagate 320GB
3 WD 640's

The other 80 and 160 will becomes externals

What would you suggest I do?
 
Well what do you want exactly? Fast performance? Data uptime?
 
Run the 3 640GB drives in RAID 5 (this will yield 1280GB of fast RAID storage that can survive a single disc failure) and use the 320GB as a temp drive for various junk. This gives you some performance and data safety benefits of RAID without really risking your data. Also you could use the 320GB as a secondary backup for critical files (however I suspect you will use the externals for this). If this sounds like a bad idea to you, then you might want to reconsider this whole RAID idea or research it A LOT more before screwing up your data.
 
I can say right now, I just want faster drives than what I have now. Other than my Seagate 320GB, the other drives are 3 to 4 years old (that includes the external). Been using the 80GB as my main drive since Jan 2004. Do you think I should just run each drive individually, considering they are pretty fast for 7200RPM anyways? Most of my 320 is taken up, most of my 160 is taken up, most of my 250 is taken up and most of my 80 is taken up (this is kind of why for the past 4 years, I have built up about 1.7TB worth of anime/movie DVDs). I would like speed for things to load faster, but I don't need the fastest. The thing is, this is such a huge jump is storage for me, it's gonna be nuts and confusing how I'm even gonna fill them all lol. Trust me, if I lost any of my data, I would probably beat the crap out of something and be mad as hell lol. Moving files quickly would be nice and loading games would also be a nice thing, but I've heard that RAID 0 can increase the risk of failure to disk drives. But it seems that drives are getting more reliable these days. Ocellaris, that is pretty much what my friend suggested, I guess I worded myself wrong originally (for the 3 640's that is). I would probably use the 80, 160 and 250 that are externals as back ups if need be. They aren't large, but they are large enough to back up the files and folders I want saved.
 
I've heard that RAID 0 can increase the risk of failure to disk drives. But it seems that drives are getting more reliable these days.

As you add drives to a Raid 0, the risk of data loss goes up exponentially, so that is just a bad idea for data. Raid 0 works good for something like and OS drive in an environment where you can quickly rebuild your OS from a saved image if you need it. Also works good if you need as much drive bandwidth as possible for video work.

Never count on a drive to simply not fail :D
 
Ok, then RAID 0 is out of the question. I was also suggested to maybe use JBOD for the 3 640's and make the 320 the main drive? I just want something way faster than what I have, faster transfers, faster loading of games, but I don't need the best. Using the JBOD could use the space on the 640's more efficiently then not having it at all. Would that be a better suggestion?
 
JBOD is usually for mashing multiple small discs into a single (virtual) disc. For example right now I have a 20, 60, and a couple 30 GB drives sitting in a drawer. Well, having a bunch of partitions 30GB or small is pretty much useless and just going to confuse the shit out of me. I could JBOD those into a virtual single 140GB drive, which would be a usable size for me. Also in JBOD, the loss of any one drive means some hefty data loss so that is another thing you want to avoid.
 
So should I just give each drive an individual letter and not in RAID? I would do this so I don't have the problem of losing info spread on different drives, but it would be just the speed of one drive. But to be honest, it would be a lot faster than what I currently have. If I was to go with a drive to each it's own, which one should I make the main drive, a 640 or a 320?
 
Since speed is a concern of yours, I'd say use the 640GB as the main drive. It'll be a boost up from your Seagate 320GB drive.
 
So just use one of the 640's as the main drive and then don't do raid and leave each drive to itself as storage?
 
Just use the 320 for your OS and do the 3x640 in a RAID5 and call it a day is what I say. The 320 is pretty fast for a main drive and honestly what do you plan on putting on your storage drives that need lots of speed?
 
Games, large file transfers and decompressing files, the usual stuff. Also, faster readings since I use Vista and the stupid auto-read crap.
 
^ Aren't games and such going on the main drive? Plus RAID5 doesn't really slow down reads, but the writes are a bit slower but nothing mind blowingly slower. I bet the RAID5 on even a software raid would be faster than on some of your older drives.
 
What I might think of doing for now is that I will use one of the 640's for a main drive, use 640 and the 320 for now. Leave the other 2 drives in the computer alone, but if in the future, I might want to give them some kind of RAID. Seriously, the 640 and the 320 combined is larger than my computer had previously for storage, by a lot lol.

Maybe a RAID 1 on the 2 drives later on.
 
^ Aren't games and such going on the main drive? Plus RAID5 doesn't really slow down reads, but the writes are a bit slower but nothing mind blowingly slower. I bet the RAID5 on even a software raid would be faster than on some of your older drives.

R5 for an OS is bad. It's pretty slow compared to a single drive.

What I might think of doing for now is that I will use one of the 640's for a main drive, use 640 and the 320 for now. Leave the other 2 drives in the computer alone, but if in the future, I might want to give them some kind of RAID. Seriously, the 640 and the 320 combined is larger than my computer had previously for storage, by a lot lol.

Maybe a RAID 1 on the 2 drives later on.

I'm have a system that has 20 drives in it. I have a R5 for 8x320GB drives and the rest are single disks (no RAID, no JBOD). If you decide on doing RAID, you'll want to back up all your data and create an array. You don't want to risk losing all that data.
 
I talked to another friend of mine and explained everything to me. I am going to have my Seagate 320GB as my main drive and the RAID 5 the 3 640's, because that sounds real nice on having speed boost, redundancy and fault tolerance (such as if a drive dies, put a new one in and it will rebuild what was on the failed drive). So it's gonna be this:

1 Seagate 320GB (Main Disk)
3 WD 640's in RAID 5 (Storage)

Externals:
1 Maxtor 80GB
1 Maxtor 160GB
1 WD 250GB

This should end up to be about 1.9 TB worth of space when formatted (maybe a tad less).

Luckily my MSI P7N Platinum mobo supports RAID 5, so that is nice.
 
I would go with a separate card. They are easier to find than motherboards after a few years. Dan_D, unhappy_mage, and Dew should be able to explain why a whole lot better than I on why a separate card is better than using the motherboard.
 
If I were to use a controller, can you suggest one that is like less than $50?

Wow, I just noticed, there are not a lot of controller cards with RAID 5 that are cheap...
 
Less than $50, no. Highpoint 2300 would be my recommendation for a cheap R5 card. There is nothing wrong with keeping disks separate either. I should say this, you probably won't notice a huge difference with RAID on your motherboard and a separate card, but I'm looking at it in the way of migrating your array or possibly recovering it if the controller dies.
 
I got a question, even though my motherboard can do RAID 5 and the card is easier to use, can't I just not spend money and figure out how to work it on the mobo?
 
I got a question, even though my motherboard can do RAID 5 and the card is easier to use, can't I just not spend money and figure out how to work it on the mobo?

You can. But ease of use isn't the reason why we recommend getting separate card. The onboard RAID is pretty easy to setup these days. The main reason we recommend a separate card is for ease of migration and recovering if the motherboard dies.

The problem with using a budget RAID5 card or the motherboard RAID is the CPU usage. Budget RAID cards and motherboard RAID chips are hardware assisted software RAID. Meaning that they setup and maintain the RAID but all of parity calculations is done by the CPU. This means that you'll see steady CPU usage even when the PC is idle. Depending on the CPU and size of the array, you're looking at as low as 10% CPU usage or up to 50% CPU usage at all times.
 
Well right now I have this as my system:

Intel Xeon E3110 (Identical to the E8400)
Wintec 4GB DDR2-800
MSI P7N Platinum mobo
eVGA 8800GT 512MB

Think it would run it fine if I did onboard?
 
Should run fine. Though I'd test out the RAID5 for a week or so before putting any valuable data on it. See if running the RAID5 impacts your game performance.
 
I will probably do that. I am going to update the drivers for my mobo chipset so I get the latest drivers for the RAID drivers. Remember, it's the weekend, so I can't get the drives today or tomorrow lol.
 
Perhaps I should rephrase that. Why would you want to use R0 for storage? :confused:
I suppose we'll just end up arguing what "storage" means. What else would you possibly use a hard disk drive for?

RAID0 is great at sustained transfer rates, not so hot at random I/O, and has a risk of data loss. Given these attributes, it seems like it's well-suited for any application where I/O jobs play to its strengths, and its weaknesses are mitigated by a backup strategy or data that's readily re-created.
 
I'd love to see your math showing the exponential increase

It's about MTBF.

If you use 2 drives in RAID 0, the MTBF of the RAID set is 1/4th of the MTBF of a single drive.

Yes, that's One-Quarter of the life expectancy of a single drive.
 
I suppose we'll just end up arguing what "storage" means. What else would you possibly use a hard disk drive for?

RAID0 is great at sustained transfer rates, not so hot at random I/O, and has a risk of data loss. Given these attributes, it seems like it's well-suited for any application where I/O jobs play to its strengths, and its weaknesses are mitigated by a backup strategy or data that's readily re-created.

I'm not talking about speed at all. I am merely speaking about an increase in data loss. 1TB is a lot of data to recover and takes a long time with DVDs (I know that one well).

It's about MTBF.

If you use 2 drives in RAID 0, the MTBF of the RAID set is 1/4th of the MTBF of a single drive.

Yes, that's One-Quarter of the life expectancy of a single drive.

How does RAID decrease life expectancy? :confused: I think you mean to say the probability of data loss increases with R0. Right?
 
I'd love to see your math showing the exponential increase

Assuming that I use the binomial distribution to model this case, let's take X~Bi(1,p) for a 1-drive scenario vs. Y~Bi(2,p) for a 2-drive scenario.

X and Y are the number of drives that fail given the limits of their discrete distribution (0 to 1 for X, and 0 to 2 for Y).

P(X=0) = (1-p), whereas P(Y=0) = (1-p)^2, where p is the probability of drive failure within a given period of time (although for measuring mean time before failure, it would be better to use a Poisson distribution). In proper English, what this means is that the probability that no drives will fail in the given period of time for the 1-drive scenario is equal to 1-p, whereas the probability that no drives will fail for the 2-drive scenario is equal to (1-p)^2. Note that (1-p)^2 is always smaller than 1-p, unless p = 0, which realistically, is impossible. Let's say that 10% of drives fail within this given time period, making p=0.1. Also, we note that in order to halve the chance of success going from the 1-drive scenario to the 2-drive scenario (basically, get the case where P(X=0) = 2P(Y=0)), we require that p >= 0.5, since for p < 0.5, (1-p) / (1-p)^2 < 2. How likely is it that a drive has a 50% (or greater) chance of failure within a reasonable time range (say, 3 years)? I'd say, subjectively, and from my own experience, that it is very unlikely, and therefore, it would be difficult to make the conclusion that as you increase the number of drives, the chance of success decreases exponentially (although, you could make the argument that the chance of success decreases logarithmically).

Comments on my math are welcome. If there's anything that you feel I've done incorrectly or applied incorrectly, I'd like to know. I've actually surprised myself with this calculation, since I also thought that drive failure decreased the chance of success exponentially...
 
How does RAID decrease life expectancy? :confused: I think you mean to say the probability of data loss increases with R0. Right?

You lose one drive in a RAID 0 set, you lose all of your data.

The probability of disk failure increases when you use R0.

The simple analogy was that if you have a single "brand x" drive with MTBF of 190,000 hours runtime, then the MTBF of a striped pair of "brand x" drives would be 1/2 that of a single drive because you're relying on 2 drives.

IBM, home of some fantastic scientists and engineers, produced a good article about MTBF as applicable to hard drives and RAIDs at IBM's site.

They advised against using RAID 0 because (cue funky math problem...)

If a single drive has an MTBF(single) of "T", then
(as a first order approximation)
MTBF(RAID 0, n drives) = MTBF(single) / n^2

In english, the MTBF of a RAID0 of n drives is equal to the single drive MTBF divided by the square of the number of drives.

If you stripe 2 drives, then the MTBF of the stripe set is a quarter of the MTBF of a single drive.

MTBF (RAID0, 2 drives) = MTBF(single)/2^2

for grins, let's plug in the 190,000 hour estimate
MTBF (RAID0, 2 drives) = MTBF(190,000)/2^2
MTBF (RAID0, 2 drives) = MTBF(190,000)/4
MTBF (RAID0, 2 drives) = MTBF(47,500)

MTBF of a RAID0, 2 drive set = 47,500 hrs using the single drive rate of 190,000 hours

The MTBF decreases exponentially - or logarithmically? IDK, it's been a while since I took a math course - the more drives you add to the RAID0 set.
 
Increase of data failure =/ increase of drive failure. If a R0 array dies for whatever reason, you can still use the disks (unless of course they are the reason for the death). There are many reasons why a R0 can die: drive dies, controller dies, data corruption, etc. Just because a R0 array goes, it does not mean the drives cannot be used. ;)
 
I'm not talking about speed at all. I am merely speaking about an increase in data loss. 1TB is a lot of data to recover and takes a long time with DVDs (I know that one well).
What you said is "Besides, you never, ever, ever, ever use R0 for storage." So, yes, I know you're not considering speed. The problem I have is with "never ever ever ever". There are cases where RAID0 is an acceptable idea, even a chance of data loss.
 
Back
Top