topic Re: 'Double' driver failure pretection VS vraid6 in HPE EVA Storage

'Double' driver failure pretection VS vraid6

SuperArmy — Mon, 19 Jan 2015 07:30:55 GMT

As you know, for HP EVA, there are two types of data protection:
1. 'Double' driver failure pretection:
Double failure protection. Reserves space within a disk group for data reconstructions if 2 disks fail. The disk group can tolerate the loss of 2 physical disk and maintain the redundancy (Vraid) levels in the group.
2. vraid6:
With Vraid6, host data is broken into chunks and distributed on all of the physical disks in the disk group in which the virtual disk resides, and a parity chunk is calculated and written twice to the physical disks. If one of the data chunks becomes corrupted, the storage system can automatically reconstruct the data from the remaining data chunks and parity chunks. The data on a Vraid6 virtual disk will not be lost if 2 physical disks in the disk group fail.

Both can tolerate 2 physical disks lost, whats the difference between them? Is it necessary to created a raid6 virtual disk under a 'Double' type disk group? or raid5 is enough?

P.S. this thread has been moevd from General to Storage Area Networks (SAN) (Enterprise). - Hp Forum Moderator

Re: 'Double' driver failure pretection VS vraid6

Torsten. — Thu, 26 Sep 2013 13:22:40 GMT

Don't mix up this!

1. 'Double' driver failure pretection:
Double failure protection. Reserves space within a disk group for data reconstructions if 2 disks fail....

Double or single reserves space equivalent to 4 or 2 disks. Nothing more than that.

A RAID5 has single parity information, RAID6 has double parity information.

No problem for both if a single disk fails.

But if 2 disks fail at the same time in a single RSS group in a RAID5, the LUN is dead.

A LUN in RAID 6 will survive.

This scenario is unlikely, but it happens.

If a single disks fails, the system rebuild and then the next disk fails, again no problem for both, because after the reconstruct after the first disk fail the syste is again fully redundant.

If the setting is "none" and you have enough free space available, the system will rebuild and will be fully redundant again. The reserved space setting comes into play if there is no user space available ONLY.

>> Both can tolerate 2 physical disks lost, whats the difference between them?

Difference: RAID5 can tolerate 1 disk loss at the same time, RAID6 cantolerate 2 disk loss at the same time (within the same RSS group).

Re: 'Double' driver failure pretection VS vraid6

SuperArmy — Fri, 27 Sep 2013 02:36:10 GMT

Hi Torsten., thanks for you reply!

But I am not saying raid5.

'Double' driver failure pretection is on Disk Group level, and vraid5and vraid6 is on virutal disks level.

This morning, my colleage told me that 'Double' driver failure pretection provides two disks lost pretection on the whole disk group range, vraid6 provides two disks lost pretection on RSS range.

Re: 'Double' driver failure pretection VS vraid6

Torsten. — Fri, 27 Sep 2013 05:37:33 GMT

Disagreed.

The protecting level for the diskgroup (none/single/double) just reserves space for reconstructing.

Nothing else than that! Just RESERVE space, to ensure a rebuild can be done even the array is "full".

You protect your data using the vraid level from none (vRAID0) to highest (vRAID6).

Re: 'Double' driver failure pretection VS vraid6

SuperArmy — Sat, 28 Sep 2013 15:24:12 GMT

Thank you Torston, now I clear what the protecting level for the diskgroup is, it's just reservers space.

Suppose there is a disk group,
1. Some day one physical disk in this group becomes corrupted, then the storage will rebuild(leveling) the date which on the failed disk into the reservers space, the result will be
a.) the data on raid0 vdisk will be lost.
b.) the raid 1, raid 5, raid6 vdisk will survive.
2. Again during the leveling period, another physical disk is failed,
If the failed physical disks located in different RSS then the same result with step 1.
If the failed physical disks located in same RSS, then:
a.) The raid0, raid1, raid5 vdisk data will be lost.
b.) the raid6 vdisk will survive.
3. Further more serious, at the same time the 3rd physical disk is failed and these 3 disks are in the same RSS, this time all the vdisks in the disk group will be lost, even raid6 vdisks.

are the above assumptions right?

Re: 'Double' driver failure pretection VS vraid6

Torsten. — Sat, 28 Sep 2013 19:27:58 GMT

The status "leveling" is usually not a concern, but "reconstructing" after a disk failure is. During this process the data is not 100% redundant, but this usually takes a few hours only depending on disk size and system load. Many "disk failures" are predictive failures initiated by the system itself, but it is clever enough to not start this during a reconstruct. So overall, a double or even triple disk failure within a single rss group at the same time is very unlikely.

If you really want to have highest HA, double everything. This means in this case use 2 EVAs and mirror them.