Hi,

 

lets say I have an unraid System with 8 x 2 TB. Then the first of the discs is the parity disc, right?

 

But what happens when the parity-disc fails, do I then loose all the data on the parity disc AND on all the other 7 discs?

 

Or can I replace the parity disc with a new disc without any data loss?

 

It would be nice if someone could answer my question.

 

Cheers,

 

Chris

Yes, the first disk (or 'a' disk) would be assigned as the parity disk.

 

If the parity disk fails you lose no data except the parity info on the party disk.

 

In that situation, yes, you replace the parity disk with a new one and rebuild your parity onto it.

 

If a non parity disk fails then the parity disk will step in and you will have full access to data that was on the failed disk until you replace the failed drive and the data will be rebuilt onto the disk.

... this sounds pretty safe.

 

This is basically the same safety level that a RAID5 provides, isn't it?

 

Cheers,

 

Chris

I would like to point out emphatically that if you lose the parity drive NO data is lost at all. You only lose your protection and as soon as the disk is replaced and parity is re-created you get your protection back. If another disk is lost before the parity drive is replaced and re-created than the data from the lost disk and the lost disk only is lost. If instead of the parity disk you had lost a data disk, than similar to the other case, no data is lost, just the protection. though if you lose a second disk before replacement than the data from both disks are lost.

 

I like to think it about it visually:

Say we have 5 disks (4 data 1 Parity) the data disks have a random set of 1s & 0s, Parity is than created using XOR which can be thought of simply as a odd even gate (0 if even 1 if odd)

 

11001101010 - Data 1

10010110101 - Data 2

10010010100 - Data 3

01011010100 - Data 4

 

10010011111 - Parity

 

See the party drive just hold whether that column is odd or even. Now lets say we lose a drive:

 

11001101010 - Data 1

10010110101 - Data 2

10010010100 - Data 3

01011010100 - Data 4

 

10010011111 - Parity

 

Now we have to replace the disk and recreate the data. It simply does this by looking at all the data disks and saying, for example with the fist column, ok the numbers are even but parity says odd so the missing bit is a 1, next column, the numbers are even and the parity says even it must have been a 0, and so on until the data is recreated on the new disk. As you can also see from this example is that if the parity dive is lost, no data is lost for the data is not touched. my example is a bit dumbed down, and I am sure has some large factual errors, but the base idea is correct.

 

So I am just restating what boof said

 

If the parity disk fails you lose no data except the parity info on the party disk.

If a non parity disk fails then the parity disk will step in and you will have full access to data that was on the failed disk until you replace the failed drive and the data will be rebuilt onto the disk.

... this sounds pretty safe.

 

This is basically the same safety level that a RAID5 provides, isn't it?

 

Cheers,

 

Chris

 

In simple terms yes, both will tolerate the failure of 'a' disk with no issue. The failure of 2, or more, disks at the same time is where you will have problems.

 

In that instance I would suggest unraid is better as you will then only lost data on the disks that actually failed. Data on the rest of the disks will be untouched.

 

In the same situation under a striped raid configuration you would lose the entire stripe and therefore lose all the data on the array.

 

One of the big benefits of unraid.

... this sounds pretty safe.

 

This is basically the same safety level that a RAID5 provides, isn't it?

 

Cheers,

 

Chris

 

In simple terms yes, both will tolerate the failure of 'a' disk with no issue. The failure of 2, or more, disks at the same time is where you will have problems.

 

In that instance I would suggest unraid is better as you will then only lost data on the disks that actually failed. Data on the rest of the disks will be untouched.

 

In the same situation under a striped raid configuration you would lose the entire stripe and therefore lose all the data on the array.

 

One of the big benefits of unraid.

 

But I believe that striping has an advantage in data reliability.  unRAID maintains parity, but does not use it in any way unless there is a failure or error of some type.  It could be completed FUBARed and no one would know or care until such time as a failure occurs and you go to do a recovery and wind up with garbage.  It is not even self-correcting.  Performing monthly parity checks helps guard against this occurring, but unRAID has no real-time data verification going on.

 

RAID5 (as I understand it and please correct me if this is wrong) is reading the striped data from all of the disks simultaneously, and is validating that the integrity of the stripe is correct.  If a disk goes sideways and starts spewing garbage, RAID5 will instantly realize it and have the ability to isolate and disable the bad disk.  Therefore RAID5 is quicker to identify a bad disk and does not allow subtle errors to creep into an array as unRAID can at certain times.  But you do get the other adavantages with unRAID - in terms of being able to recover individual disks in the event of a disaster.  Lacking fire or theft, the risk of losing the entire array with unRAID is extremely low, as compared to a real possibility with RAID5.

A good point, though I'm not sure any striped system (excepting possibly zfs? though I'm not sure if it does..) verifies data on read. All it returns is it can read from the disks, I don't believe it performs a read, checks the parity of the stripe, then gives you the data if it's correct.

 

The equivalent of the parity check on unraid would be scrubbing the raid array which you should either be doing continuously, or at least regularly as you do in unraid.

 

But you're likely right a failed (or failing) disk would be identified much more quickly if only because it will be in use more.

To my knowledge, only ZFS does what bjp999 described.  unRAID does it during every parity check, whereas ZFS does it in real time during every file read.

 

The way I see it, RAID5 in general (without ZFS) is less secure than unRAID.  Here's why:

 

- If 2 or more disks fail, you lose the entire array.

- It is very difficult and can be expensive to recover data from a striped array (I don't know of any free tools that allow you to do this).

- RAID5 requires identical drives.  Considering that most people would buy multiple of the same drive for convenience's sake, this greatly increases your chances of a multiple drive failure (since they all could come from the same bad or defective batch).

- RAID5 wears the drives evenly.  Again, this increases your chances of a multiple drive failure since identical drives worn evenly are more likely to all fail at once.

- Hardware RAID controllers can be expensive to replace, and hard to find if they are old.  If your hardware RAID controller dies (even if your disks are all healthy), you will need to replace it with the exact same one in order to rescue your data.

 

In my view, RAID5 is designed for speed, whereas unRAID is designed for preventing data loss.  Granted RAID5 is much better than just keeping your important data on a single disk.  Also, unRAID is fairly speedy, just not as fast as RAID5.

 

The above all applies to RAID6 as well, except that the number of disks that must fail before you see data loss increases to 3.