EVA 4400 Disk Protection Level

[UselessUser]

Hi,

I understand the fundamental workings of EVA, and I also understand that this so called "Protection Level" is not really a protection level, the VRAID level is this.

However I have double protection set at the moment on 8 1TB FATA disks. I really need some more space and am having a hard time justifying the purchase of new disks, when I can adjust the protection level to single and free up 2TB in a single hit...

Can someone explain what the technical difference is...

[IBaltay · Accepted solution]

Hi,

Single/Double disk protection

Protection level influences availability
The protection level defines the number of disk failureâ auto reconstruction cycles that the array can accomplish without replacement of a failed disk. Following a disk failure, the controller re-creates the missing data from the parity information. The data is still available after the disk failure, but it is not protected from another disk failure until the reconstruction operation completes.
For example, â singleâ protection level provides continued operation in the event of 2 disk failures, assuming the reconstruction of the first failed disk completes before the second and so on till the number of the failed disk of 4 in the double disk protection.

None disk protection

For Vraid1 and Vraid5, protection level â noneâ provides resilience to a single disk failure; whereas Vraid6 provides resilience to a dual disk failure; however, this is not a best practice configuration. Vraid0 offers no protection from a disk failure.

[gstonian]

For only 8 disks, adjust the Protection Level to 1...

[Uwe Zessin]

Right, the so-called "protection level" has nothing to do with RAID redundancy - that is provided by the VRAID-level: 1, 5, 6.

The EVA's virtualization DOES NOT add redundancy (some people I've talked to really believed this!). VRAID-0 does not provide protection against a disk drive failure.

Think of the "protection level" as some kind of "virtualized set of spare disks". The capacity is 'striped' across all disk drives in a disk group.

Level 0/none does not reserve any capacity.

Level 1/single reserves 2 times the size of the largest disk drive in the disk group.

Level 2/double reserves 4 times. I'm sure by now you understand that it just reserves more space for recovering redundancy, but it does not make your data "more safe". That level is really intended for systems which are not closely monitored and/or replacement parts do not come by quick (in my opinion).

2/4 times is due to the way the EVA stores VRAID-1 redundant information. It does always apply, even if you don't use VRAID-1 - but the EVA must care, because you could change your mind one day and completely fill a disk group with VRAID-1 vdisks.

[Adrian Parker]

Just a query on this subject as I was directed to this thread, I'm curoius to know if it possible to survive a second drive failure during the reconstruction process?

Say we have a 50 disk group with a mix of vraid5 and vraid1 vdisks, we lose one disk and reconstruction starts. We then lose a second disk, will we lose all vraid5 and vraid1 luns?

Rightly or wrongly, my understanding is that the group is split into smaller storage sets and you may be able to survive concurrent failures in different storage sets, or different vraid1 couples.

Thanks
Adrian

[Uwe Zessin]

A disk group is divided into one or more separate 'failure domains' (called RSS - Redundant Storage Set - and no, there is no redundancy for VRAID-0 - if one disk fails you have lost data!). A reconstruct is done completely within that RSS first. Only after it has finished the data is 'leveled' across all disks in the group. If the second failure is outside of the affected disk, yes, no data is lost.

The problem is:
You (and the EVA) don't know in advance which disk drive fails during the reconstruction.

[McCready]

To put it another way, single/double protection keeps you from having to figure out how much space to leave in your disk group to survive a disk failure - Single leaves enough space to survive a single failure, double allows you to survive two failures at the same time, assuming, of course, your VRAID1/5 setting has basically copied your data to other disks to allow the EVA to redeploy the data once a failure happens.

My personal thought on a 8 FATA drive disk group would be to set single protection but deploy your data in VRAID6 - just make sure you can replace a failed disk quickly (24 hours, in my opinion) so you can get your full redundant space cushion back.

My question to those with more knowledge than I would be what happens using VRAID6 if you do get the second disk failing before the first has completed reconstruction - I'm assuming the data would be safe, but would you still be operational as you might not have enough space to do a proper rebuild without the replacment of the failed disks?

As an aside, if this is your only disk group, you might want to take a more conservative approach, as FATA drives are more prone to failure under a high workload than Fiberchannel drives. If the disk group is mainly for archival purposes with low I/O, then I would not be as worried. I can't say exactly what low and high would be.