Vraid Level

[simon_164]

For a test I had an 8 disk Vraid 1 with single protection. I have created a Vdisk and put some information on that and presented to a Windows system. For a test, I have removed the first disk--> no problem everything is good. I removed second disk --> no problem also everything is good. I removed the third disk and now my Vdisk is red and my diskgroup is not available anymore. on Windows the disk is gone. Shouldn't I be able to loose half of the disks and stay healthy? what is the deal here? I reput the disks in different order, now i got the information back but i should restart windows to do that.

[Uwe Zessin · Accepted solution]

> Shouldn't I be able to loose half of the disks and stay healthy?

Unfortunately, it is not that easy. VRAID-1 tries to guarantee that a data block is stored on two different physical disk drives in two different disk drive enclosures, but the internal data structures can become confused after disk group/ungroup operations. Did you check the "RSS Disk state"? It is in the upper right corner of the 'Disk Group Properties page. If you managed to pull the 'right' pair of disks before the EVA could restore redundancy, you have caused data loss.

But there is more behind the scene. For example: some disks also hold meta-data. If you happen too pull to many of these disks, the disk group or the whole array can become inoperable, too.

> I reput the disks in different order, now i got the information back

The EVA can deal with 'drive roaming', but I have never seen it being documented / supported.

> should restart windows to do that.

Yes, I think that is a good idea to get a freshly rebooted Windows server.

[simon_164]

Thank you Uwe,
so if i wait enough time, I can remove half of the disks and still have my disk presented and operational.
Do you have any document describing the whole information? now what if we r talking about Raid 5

[Steven Clementi]

Simon:

How many disk shelves do you have in your EVA?

As you may have figured out already by Uwe's comments, if you pulled out the "wrong" drives, then yes... you would lose access to the vdisk(s).

I believe, for Vraid5, You would need to have the right combination of drives in a 6D configuration in order to be able to lose multiple disks, or in a 8D configuration (and better), you should be able to lose the whole shelf.

Again, this all depends on your RSS groupings and where they are currently located.


Steven

[simon_164]

I have 2 disk enclosure in EVA 4000, i have 2 disk groups:
1- has 12 disks and is configured to hold raid 5 ( i mean i will put on it only disks with raid 5, advice)
2- has 8 disks and is configured to hold raid 1 ( again i will put on it only disks with raid 1, advice)

i guess the first disk group has 2 RSS each with 6 ?
the second has only one with 8 disks.

all has single protection level,
from the configuration above i guess i will be able to loose 2 disks at the same time and still be operation on the first disk group and loose 2 disks at the same time and still be operational on the second disk group.
please advice.

[Uwe Zessin]

A disk group with VRAID-5 virtual disks is even less forgiving (IMO):

the EVA's RAID-5 implementation uses a 4D+1P algorithm. 4 data blocks are combined to a parity block. All 5 blocks are stored on 5 different physical disk drives.

A disk group with more than 11 disk drives is divided into multiply Redundant Storage Sets (RSS) which have between 6 and 11 disk drives. These form separate 'failure domains', so that a disk group can withstand multiple disk drive failures as long as they happen in different RSSes.

But how is the 4D+1P stored in an RSS with 6,7,8,9,11 disk drives? According to a HP support person, different stripes are equally distributed over the RSS, e.g.

(view with a fixed font)
D-1,D-2,D-3,D-4,D-5,D-6
-----------------------
1-1,1-2,1-3,1-4,1-P,2-1
2-2,2-3,2-4,2-P,3-1,3-2
3-3,3-4,3-P
and so on.

D-1,D-2... are disk drives within the RSS. 1-x or 2-x are the stripes; -P is the parity in that stripe.

It is clear that the RSS can withstand the failure of a single disk drive, but what happens if, for example, D-5 *and* D-6 fail?
Stripes 1- and 2- will loose a single disk and can continue, but 3- will loose disks 3-1 and 3-2 which cannot be recovered by a single parity :-(
That will make it very likely that *any* 2-disk failure within a RSS will cause loss of data.

[Uwe Zessin]

> all has single protection level,
> from the configuration above i guess i will
> be able to loose 2 disks at the same time
> and still be operation on the first disk group
> and loose 2 disks at the same time and still be
> operational on the second disk group.

Nope, sorry. The disk group "protection level" is simply bad terminology, because it *does NOT* actively 'protect' anything. It does NOT create some additional redundancy.

The 'protection level' is only a space reservation. Think of it as a set of spare disk drives which have joined the disk group and their capacity 'rotated' by 90 degrees so that it is equally distributed over all disk drives in the group. Or think of it as some highwater mark that prevents you from accidently filling the disk group with virtuals disks and not leaving space for restoring VRAID redundancy.


You can set the 'protection level' to 2/double and it will reserve space that is 4 times the size of the largest disk in the group. Now create a virtual disk in VRAID-0.
Pull a disk drive and that virtual disk becomes inaccessible, because VRAID-0 does not imply any redundancy.

On the other hand, you can set the 'protection level' to 0/none. As long as you leave enough free space in the disk group, the EVA _can_ recover virtual disks in VRAID-1 or VRAID-5.
In fact, whenever a disk drive fails, the EVA will first draw from the space that is available for virtual disks and try to keep the 'protection space' as long as possible.


So be fair, when I started working with the EVA some years ago, I was also very confused.