EVA4400 RAID Redundancy question

[Jason Ng Teng Po]

Hi all,

In the EVA technology, I heard of single and double protection for disk redundancy.

In a group of 8 disk with 1 TB teach. If I on single protection, means I can have up to 2 disk fails.

But what about RAID redundancy? If you have 8 disk in a group and I perform a RAID 1 with single protection, what will be my redundancy? Up to 4 disk failure or still 2?

[Uwe Zessin]

The so-called "protection level" does NOT protect any data!! It is the virtualized equivalent of spare disks. On the other hand - the EVA can rebuild redundancy even with protection level = "none" as long as there is enough unused space in a disk group.

The real data protection is on each virtual disk's VRAID-level. But to make it clear: VRAID-0 does NOT offer ANY protection against disk failures no matter what 'protection level' is set.

VRAID-1 and -5 offer guaranteed protection against a single disk failure; VRAID-6 against a double disk failure of any two disks in a group.

The EVA divides larger (12 or more disk drives) disk groups into separate failure domains, so theoretially it could cover more disk drive failures, but one never knows in advance.

[Prokopets]

Uwe, you said "EVA divides larger (12 or more disk drives) disk groups into separate failure domains" am i right and it's true only for x400 EVAs?

[Víctor Cespón]

No, this has been that way on all EVAs. Let's say you have a disk group with 30 disks, when you create a vdisk in RAID5, it does not use 29 data + 1 parity.
The 30 disks are subdivided in smaller groups (typically 8 disks), and the data is stored as 4 data + 1 parity.
If two disks fail at the same time but are not from the same subgroup, no data will be lost.

When a disk fails, the EVA starts rebuilding the RAID with maximum priority, so that we're back to a fully redundant situation as soon as possible. It will use the free space on the disk group, or the reserved space in "disk failure protection", to store the regenerated data (the data that was on the disk that failed).

[Sheldon Smith]

In a group of 8 disks, if using single protection AND you are low on free space in the disk group, you can have up to ONE disk failure.

Single protection costs two disks-worth of storage to allow for all vdisks using vRAID1. Both the blocks from the failed drive and their mirror partner blocks need to be moved elsewhere in the disk group. Double protection costs four disks-worth of storage. And if you have mixed disk sizes, the EVA reserves based on the size of the largest disk in the disk group.

-----
Don't forget to assign points to all the answers! :)

[Steven Clementi]

Using the example above of 30 disks...


...you would have 4 "failure domains".

Theoretically, you could lose a disk from each set (thats 4 disks total) and still be operational without data loss depending upon how much space you have available in the disk group.

BUT, also as stated... you can't possibly know which disk or disks will fail ;o)


Steven

[phillbl1]

anyone point me in the direction of some good documents that explain all this?

Thanks

[Víctor Cespón]

Much of the information about the internal workings of an EVA is not public. In fact the information about RSS status was removed from public view in all recent versions of Command View.

There's a good general overview of EVA systems here:

http://www.scribd.com/doc/2233504/EVA-Overview

Looks like a public presentantion, it does not say HP restricted anywhere, so...