VRAID and diskgroups

[Jacques Tessier]

I am trying to get a better understanding of vraid and disk groups.
I want to setup mirrored disks.

First I must create a disk group and from what I have seen on the web, I need a minimum of 8 disks in this group.
Then I take this group and then apply VRAID 1.
The data somehow gets spread evenly across all disks.
This leads me to believe that I do not have one data drive and one mirrored drive.
Can someone explain to me how the data is laided out for Vraid 1.

One other question I have is regarding VRAID 5. Can I create a 7 data + 1 parity or is 4+1 the only available VRAID 5 I can implement. Again I am not sure how the data gets spread across the disks in the disk group.

I am going to implement a EVA 8100 as my disk array platform

[Phillip Thayer · Accepted solution]

The data in an EVA of any type is virtualized. What this means is that the EVA will create disk sets called Redundant Storage Sets (RSS) of 8 disks each. This is where the minimum of 8 disks comes into play.

So if you have an EVA with 56 disks on it you will have 7 RSS's. If the number of disks is not evenly divided by 8 then it will create the last RSS with between 6 and 11 disks. In the event that the number of disks in your EVA is not evenly divided by 8 then it will adjust the last RSS to accommodate the additional disks. (Sometime the last two RSS's are adjusted if needed.)

In any case, after the RSS's are created then all the disk blocks are mapped to a META table in the controllers. This meta table is how the controller will ensure that data blocks in a VRAID structure are correctly placed based on what VRAID type is chosen when creating the LUN. The idea of 7 data + 1 parity or 4+1 is not really applicable as much here. The data and parity are spread across disks and if possible across RSS's as well.

Phil

[Ivan Ferreira]

>> First I must create a disk group and from what I have seen on the web, I need a minimum of 8 disks in this group.

Correct

>>> Then I take this group and then apply VRAID 1.

Not exactly like that. The disk group does not have any RAID type associated. On the same disk group, you can create VRAID 0, VRAID 1, VRAID 6.

>>> The data somehow gets spread evenly across all disks.

Yes.

>> This leads me to believe that I do not have one data drive and one mirrored drive.

The data is mirrored in a special form, but you DO have mirrored data. The data and it's mirror will be spread across all disks, ensuring that you can loose any disk without loosing the data. Think like if you have a VRAID1 of 500 GB and your disk group has a capacity of 1 TB:

data, mirror, data, mirror, data, mirror
data, mirror, data, mirror, data, mirror
free, free, free, free, free, free
free, free, free, free, free, free
....

>>> One other question I have is regarding VRAID 5. Can I create a 7 data + 1 parity or is 4+1 the only available VRAID 5 I can implement. Again I am not sure how the data gets spread across the disks in the disk group.

EVA does not manage these concepts, but basically is a 7+1. So, if you want to tolerate a disk enclosure failure, you need 8 enclosures with one disk per enclosure.

>>> I am going to implement a EVA 8100 as my disk array platform

I'm not an EVA fan....

[Tom O'Toole]

There is a data disk and mirror disk in some sense for vraid1. It's just that the mirroring is applied to chunks of data in the vdisk, not the whole vdisk.

The data on any vdisk is broken up into chunks and striped across drives in a disk group. For vraid1, a copy of this data is maintained on (to simplify quite a bit) every other drive. As Philip said, for each particular chunk of data, it and its copy always live within the same RSS. Pairs of drives within each rss are identified as 'married', and contain copies of the same chunks of vraid1 data.

Note that unlike other arrays, with the eva, a diskgroup can contain different raid levels. Therefore these married disks might also have different data from other non-vraid1 vdisks in the group, so it's not completely analogous to a traditional mirrored drive.

[Tom O'Toole]

Hate to keep harping on this, but the word is "lose", not "loose", Ivan.

lose - verb meaning to not have something any more

loose - verb meaning to release something.

[Uwe Zessin]

> EVA does not manage these concepts, but basically is a 7+1.

It is 4D+1P and the data is distributed across all disks within an RSS.

[Del_3]

Uwe, Are you saying that a Vdisk is only spread across an RSS and not the entire Disk Group?

[Sheldon Smith]

No, the 4D+1P is repeated across ALL the RSS groups in the Disk Group.

Once the physical Disks are added to a Disk Group, from that point on, everything is done with "chunks"1 of disk space. Chunks are striped across all the disks in the disk group. Chunks are paired and mirrored across all the disks in the disk group. Chunks are grouped in nD+1P groups across all the disks in the disk group.

-----
1 "Chunks" are technically referred to as PSEGs (Physical SEGments.)

[Uwe Zessin]

A chunk (D/P) is not the same thing as a PSEG. A PSEG holds 16 chunks.

PSEGs are used to devide the storage of a physical disk drive into smaller, manageable entities.

[Del_3]

There's some vrarid vodoo going on here.

[Del_3]

There's some vraid vodoo going on here.

[Jacques Tessier]

what I understand from this is the EVA
does its own strip mirroring in a disk group.

Why do we need to specify a VRAID if the group is already doing some sort of RAID 1/0?

[Ivan Ferreira]

>>> what I understand from this is the EVA
does its own strip mirroring in a disk group.

Yes.

>>> Why do we need to specify a VRAID if the group is already doing some sort of RAID 1/0?

Based on that information, more parity information will be created. For example, a disk group with only vraid 5 will have:

data data data parity data data data parity
data data data parity data data data parity

A disk group with only vraid 1 will have:

data parity data parity data parity
data parity data parity data parity

Of course you can mix them, and the same logic apply. This will give you the ability to loose one or more disks.

[Uwe Zessin]

> Why do we need to specify a VRAID if the group is already doing some sort of RAID 1/0?

A disk group DOES NOT do any (V-)RAID at all!!

A disk group is a collection of disk drives...

- which provides storage space for virtual disks

- usually the so-called 'protection space' to have capacity to be able to rebuild VRAID-1 or VRAID-5 virtual disks in case a physical disk drive fails

- reserves some space on some disk drives to store the EVA meta data

A virtual disk's VRAID is implemented 'on top' of the disk group.

Think about it:
if a disk group would provide RAID-1 functionality, what is the purpose of a VRAID-0 virtual disk and why will you lose that virtual disk if a physical disk drive fails in that group.

Maybe you are confused, that the virtual RAID level is called VRAID-1 while is DOES stripe mirror chunks.

[Ivan Ferreira]

Sorry I have to correct myself:

>>> what I understand from this is the EVA
does its own strip mirroring in a disk group.

Disk group does provide stripping, but does not provide any fault tolerance.

[Phillip Thayer]

But disk groups are created with RSS's which provide some protection within each RSS. This allows you to lose (hope I got that right there Tom O'T) a disk and remain functioning.

A disk group is not simply a bunch of disks grouped together.

Phil

[Jacques Tessier]

So if I understand it correctly
the disks are striped and since there is a spare disk there is fault tolerance.


I guess I have more research to do.

[Ivan Ferreira]

>> the disks are striped
Yes.

>>> and since there is a spare disk there is fault tolerance.

No. Spare disk (space in case of eva) does not provides any fault tolerance. They are dumb. The fault tolerance is provided by the RAID level, and if a disk fail, the spare disk is used for data "reconstruction".

Suppose this, if you have RAID 0 and spare disk, if one disk fail, your data is not available anymore, because RAID 0 does not provides data redundancy, so, how would you rebuild data into spare disks if you don't have data redundancy?

[Uwe Zessin]

> But disk groups are created with RSS's which provide some protection within each RSS.

RSSes are an organizational means to create multiple failure domains within a disk group for virtual disks with VRAID-1 or VRAID-5 protection, but an RSS does not provide any protection itself. You can think of them as some kind of "sub-disk group".

VRAID-0 virtual disks do not have ANY protection against physical disk drive failures, no matter how many RSSes you have or which 'protection level' you choose. The "0" is there for a reason ;-)