Protection level and vRaid

[mr_red]

Hi all!

I did lots of research about this subject and I'm now confused because maybe I asked the wrong guys ^^

So my questions are :

- If I use a vRaid0 in a diskgroup of protection level set to "one" or "double", and then I get one or two disk failure(s), are my data lost ?

- Is it possible to make a vRaid5 in a diskgroup of protection level set to "none" ? If yes, then when I get a disk failure can I continue working with my data ? Are they lost ? and are they reconstructed somewhere in the free disk space or not?

- If I use a vRaid5 in a diskgroup of protection level set to "double", I can support only 2 disks simultaneous failures without data loss or 3 disk failures ? or more (hum)

Thank you in advance :)

[Henry Fauni]

---

@mr_red wrote:

Hi all!

I did lots of research about this subject and I'm now confused because maybe I asked the wrong guys ^^

So my questions are :

- If I use a vRaid0 in a diskgroup of protection level set to "one" or "double", and then I get one or two disk failure(s), are my data lost ?

--> Selecting vRaid0 to create a virtual disk provides NO redundancy. If you lose one or more disks, you lose data.

- Is it possible to make a vRaid5 in a diskgroup of protection level set to "none" ? If yes, then when I get a disk failure can I continue working with my data ? Are they lost ? and are they reconstructed somewhere in the free disk space or not?

 --> Yes, you can select vRaid5 to create a virtual disk within a Diskgroup with protection level set to none. If you lose 1 disk, data is still available. After you replace the bad disk, the parity data will be reconstructed using the remaining good disks. But let's say you lose a second disk before you get the first bad disk replaced, then you may lose data. You can think of the "Protection level" for a disk group as a "spare disk" but not exactly.  If you setup a "Protection Level" to at least 1, if you lose a disk, it will automatically start to rebuild the data on this "spare disk" to maintain the redundancy level. Note that the Protection Level will only kick in for created virtual disks with built-in redundancy (Vraid1, Vraid5, Vraid6). This will not help you with Vraid0.

- If I use a vRaid5 in a diskgroup of protection level set to "double", I can support only 2 disks simultaneous failures without data loss or 3 disk failures ? or more (hum)

--> Setting the protection level to "double" means you are reserving a space in the DiskGroup if 2 disks fail. For a Vraid5 virtual disks within a DiskGoup with "double" protection level - Since Vraid5 has built in redundancy, it can survive without data loss up to 3 disk failures as long as data reconstruction finishes before the 2nd or the third disk fails. There is still potential for data loss if these 3 disks fails simultaneously.  

---

[mr_red]

Thanks very much, I can see much clear now.

One last thing to see perfectly clear.

Let's consider this configuration : vRaid1 on diskgroup of protection level set to "single". (Normally you can support two simultaneous disk failures with no data loss if I understood what you said)

I've read that setting this level reserve a space equal to the largest disk you have for data rebuilding or a "spare disk" like you said.

If you loose a disk, data will be reconstructed in that area, no trouble until now.

But if you loose a second disk you may loose data, and that depends on which disk you have lost. Because there must be a disk in which are written both blocks and their mirrors (using vRaid1).

If this hypothesis is wrong, then I think that the only solution is that EVA reserve more than " the largest disk ". It actually reserves the largest disk size + the space needed to spread those mirrors on the other disks than the disk where their source blocks are written (that remembers me the "but not exactly" in your reply. And that way you can loose the second disk without any data loss.

It's a bit hard to explain it without a schema, I hope you see what I want to say ^^'.

Is the second hypothesis is the right one ? Correct me if I'm wrong.

[Torsten.]

>> (Normally you can support two simultaneous disk failures with no data loss if I understood what you said)


This really depends.

Don't get "protection level" wrong.

This only means how much space is reserved for rebuilds. If you have enough free space, you could set the level to none and you still have "double" or more ...


The protection level only reserves additional space in that disk group (2x or 4x the larges disk), so you cannot use this space for your data.

Let's say you have a vraid1.

A single block will be written to a single disk and to the "partner" disk for mirroring. If now both disks fail at the same time, the block (your data) is lost.

On the other hand you could loose about 50% of all of your disks at the same time, as long as each block is still present (assume you loose all of your mirrors).

But this is only theory, not relevant for real life.



Related to real life - one disk fails, so you need space to rebuild. If now a second disk fails without replacing the first, you need more space for the rebuild. So after rebuilding the 2 failed disks your array is still fully redundant - this is the important point here!

But why needs the protection level "single" the space of 2 disks?

Let's assume the same vRAID1. A single block is written to disk1 and mirrored to the partner disk disk2. If disk1 fails now, the system is trying to rebuild the block to another pair of disks, let's say disk3 and disk4. Since disk2 has no partner anymore, it cannot be used for such mirrored data any longer, until there is a new disk1.

I try to keep this simple, the data layout inside an EVA is a bit more complicated and doesn't fit in the "quick reply" window

;-)

[mr_red]

I meant by "simultaneous disk failures" : two disks failures, one after the other without disk replacement and with enough time between them to complete the first data rebuild .

(I know I had to explain it before asking)

[Torsten.]

It always depends what is on the second disk.

Loosing a second disk in the same RSS is always critical and may lead to data loss.

Loosing a disk in another RSS has nothing to do with the first disk.


In an ideal configuration of let's say 8 disk enclosures you can loose a whole enclosure with all disks inside without loosing your data.

[Torsten.]

Maybe the better answer is - as long as there are enough available disks and disk space the array always want to make your data redundant again after a disk failure, no matter if you replace the failed disk or not. The goal is to have the same status as before the disk has failed.

Keep in mind that a decreased number of disks (after a disk failure) may force a reconstruct of RSS - this changes the internal layout to make it "perfect" again.


As said, in * theory* (!) you can loose 50% of all the disks without loosing the data, but there is no redundancy anymore. But this assumes you loose the mirrors only, hence I said "in theory".

[mr_red]

Let's have a small case to understand a little how this EVA works.

Suppose I have vRaid1 on Single disk protection (which means 2x largest disk space are reserved).

I lost one disk (here am sure that no data are lost), and after the data rebuilding in the "reserved area" I lost a second disk.

Am I sure here that "No" data are lost or it still depends on which disk is lost ??

[Torsten. · Accepted solution]

After the rebuild of the first failed disk has finished, the status is the same like before the first fails; everything is fully redundant - this means you can loose the next disk now, no matter where this disk is.

But now it depends if there is still enough space to make the array redundant again by rebuilding the data.

If not enough space available, your data may stay not redundant until you replace the failed disk.