Raid LvLs differences

[Tiberium Wolf]

I searched in google and in wiki but I still don't understand the other raid modes except the 0, 1, 0+1 or 1+0.

The other modes are striped too but they have parity data. What does this parity data do exactly? If you lose a HDD you lose all the info but you still be able to work with the rest of HDD and remaining data, right?

[Sephi]

I think wiki sort of has explains it well.

What different Raid numbers are.
More explained about Raid

And what parity is.

[Ledgem]

Originally Posted by Tiberium Wolf

The other modes are striped too but they have parity data. What does this parity data do exactly? If you lose a HDD you lose all the info but you still be able to work with the rest of HDD and remaining data, right?

Parity data isn't an exact replica of other files, but it contains data that can help to recover or restore files. The only example I can give is with PAR files, which are often sent along with a batch of files on Usenet. The PAR data contains CRC32 or MD5 checksums to ensure that all of the files match the original. If a file is missing or corrupt, the parity data is utilized to try and reconstruct the file. If you don't have enough parity data, the file may not be recoverable. If you ever see it work successfully, you'd probably think it was the coolest thing ever (I sure did).

In the case of RAID 3 (one drive is a parity drive) or RAID 5 (you can have multiple parity drives), if one HD in the RAID is removed then the parity data is used in an attempt to reconstruct that drive's data. The parity drives can't be used as data drives themselves, though - the RAID will be unusable until the drive is rebuilt. If there isn't enough data to recover the drive, then I presume that the RAID is lost, or there's massive file corruption. The chances of successful recovery with parity data increase as you increase the amount of parity data. Thus, a RAID 5 is better than a RAID 3. If you find yourself using the same number of parity drives as you have data drives, you might as well revert to RAID 1, of course - which is the most solid backup solution

[jpwong]

Technically RAID3 and 5 shouldn't see much of a difference, I believe that RAID5 is somewhat faster because of the distrobuted parity.

RAID5 is NOT multiple partity drives, rather it's the same principle as RAID3 except that the partity is spread across all disks in the RAID rather than being stored on it's own seperate disk.

In a RAID3 or 5 array, you can loose a MAX of 1 drive and be able to recover your data.

I believe RAID6 allows 2 drive failures because it has the equivelent of 2 partity disks distrobuted across the array, but most boards only seem to support 0, 1, 3/5 and one of 0+1 or 1+0.

Parity is created by using the XOR command on your data. If a drive failes, your computer can reverse calculate the XOR by using that parity block and the remainning drives and get the original data back.

As mentioned, a RAID with partity that looses a disk isn't useable until the missing drive is rebuilt. The exception is RAID3 where if you loose your partity disk, you should be able to operate as normal, except you'd basically be operating in RAID0 mode then.

[Tiberium Wolf]

So basically it's not 100% sure that you can reconstruct the failed drive.

[jpwong]

Nothing's ever 100%, so you have to look at what will work best for you.

RAID0 has no protection. If you loose a disk, all you data is lost.

RAID1 is probably the most protection, but is costs a shitload since you need to have 2x the drives.
RAID3/5 offers parity and offers some of the speed benefits you'd get from a RAID0 array. You can loose a single drive out of the array and still have your data restorable.

RAID1 is space consuming, more costly but generally is the most reliable. That's to say, the chance of the same disk in both arrays crapping out at the same time is possible, but it's not likely.

RAID3/5 offer a medium. The chances of two disks in an array failling simultaneously increases as you add more drives to the array, but for small arrays of 3-5 or so drives, it might be more suitable than trying something else.

As for drive recover on a 3 or 5, basically, as long as only 1 drive has failed, and none of the other drives have been corrupted or otherwise damaged, you should be able to replace the failed drive and reconstruct all the data.

Yes, it's possible that another drive has suffered some other damage and you will only be able to recover part/most of the drive, but I mean, it's also possible that all your drives could fail tommorrow.

IMO it would be better to have disks in a fault tolerant RAID than not. At least that way, you have the opertunity to potentially recover your data should disaster hit and knock out a drive.

[Jinto]

@Tiberium Wolf,

Imagine you split data among 4 drives and have a parity drive that contains checksums and parity data for the data. This allows you to use only one pseudo-mirror drive per raid, which is more economic than, e.g. mirror every piece of data 1:1.

How does the backup work? Well imagine you split the word 1101 0001 1010 1111 on 4 drives. So drive 1 contains 1101, drive 2 0001, drive 3 1010 and drive 4 1111. Imagine a fith's drive contains (lets assume something very easy...) a checksum generated out of the sum of the decimal values of each half-byte:

1101 = 13
0001 = 1
1010 = 10
1111 = 15

sum: 13 + 1 + 10 + 15 = 39

Now an additional step is done... if the sum is greater than 31, then subtract 32.

So in this case... 39-32 = 7 (00111)

So the fith drive contains a checksum of 7 (00111) for the given word.

If one of the drives fails. A chunk of data of the word got lost. That means one cannot use the word anymore, its corrupt. But one got a checksum for it. Lets assume drive 3 failed (1010), the following data remains:

1101
0001
????
1111

What is the checksum for this?

13 + 1 + x + 15 = 29+x

The actual checksum was 7:

Since 7 is smaller than 29+x, 32 are added.
Now the equation 39 = 29 + x, will provide the checksum for the single half-byte that got lost: 10
since 10 is 1010, the half-byte can be recovered. (with the information of the remaining data and the special checksum given). This encoding scheme is thus capable of 4 bit recoverage for words of 16 bit length.

This was a very simple example, the calculation of parity data is actually based on a more complicated algorithm, that utilizes the (trans)information of data better. Because as one can see the checksum is one bit longer, than the half-byte data chunks on each drive, which means the parity drive would be filled with 25% more data than the other disks. But thats because of the simplicity/inefficiency of the example given.
Though the principle is basically the same. If in this example, two drives get lost (50% of the data), data cannot reliably be recovered. The information in the checksum provides not enough information to do this.

[Ledgem]

There is one factor that hasn't been mentioned yet: the RAID controller. I've heard that Linux can handle most of these RAID types, but I believe Mac OS X can create RAID 0 and RAID 1, as well as what they call RAID 10 (0+1). I'd imagine that Windows should have a software RAID manager but I've never seen it in use and I'm not sure how you'd use it.

When you get into the more complicated RAID types (0+1, 3, 5) you may have to use a hardware RAID controller. It's basically a PCI expansion card that you connect your drives to, and it handles the RAID functionality. This is where the problem arises: if that expansion card is ever lost or damaged, you lose access to your RAID unless you can reconnect it to the same or a very similar card. I've heard a lot of horror stories about this, and the common advice seems to be that if you'll be relying heavily on the RAID, buy two controller cards of the same type.

A software RAID may not offer as many RAID types and by comparison it may take up a bit more system resources (in the case of RAID 3 and 5), but the benefit is that you can move it from one computer to another of the same operating system and have it recognized.

[jpwong]

XP can only do RAID0. A Windows Server platform such as Windows2003 will allow you to use RAID0, 1, and 5.

XP can be tricked into thinking it's running server platform so that you can access the RAID 1 and 5 options (they are in the OS, MS just hides them from you if you have an XP machine)

If you have a windows software RAID 1/5 and put it into a XP machine, it can read the drive as long as you go into disk management and tell it to remount every single time you boot up the machine (and then it spends like 5 hours verifying the data, but you can still access it). This won't happen (it automatically detects and mounts) if you put it into a machine that has the proper files that lets windows know that it can use those RAID modes (IE Windows Server2003 or like).

[Tiberium Wolf]

Oh. Thx for the explanation ppl.