EVA 4400 Resilience to Enclosure Failure

[Tristan Bunch]

Hi,

Quick question re: the EVA 4400. I have an EVA 4400 with four enclosures running single disk protection. All my VDisks are VRAID5 and I leave 2 disks worth of free space for 1 x PDM event. My disk groups are 16 disks, two per tray.
Is may data safe in the event of complete loss of an enclosure, i.e. loss of four disks from the group in one hit and is this affected by the number of disks in a group per enclosure.
I.e. would I need to go to 8 enclosures with 2 disks per enclosure to ensure resilience in the event of an enclosure failure or do the EVA's not work like that?

Many Thanks,
Tristan

[TTr]

Vraid5 uses ALL disks in the group but a different set of 5 disks for each set of 5 data blocks that make up the vraid5 volume.

So even if you used 2 disks per enclosure you still would not be protected from an enclosure failure. Maybe some of the smaller LUNs would be safe if they did not use any vraid5 data blocks in the failed disks.

[Tristan Bunch]

Thanks for your reply which of course raises further questions: -

Is the only way to achieve tray resilience to have one disk per enclosure i.e. maximum of eight disks in a disk group?

Is tray resilience something to be worried about with an EVA 4400 or do people typically run risk on it, as HP recommend as many disks as possible in the disk group?

Many Thanks

[Uwe Zessin]

> the smaller LUNs

The virtualization is so efficient that even a 1GByte vdisk can be striped across all 96 disk drives.

Redundancy & virtualization on the EVA is a _very complex_ matter and cannot be cut down to a hand full of rules. Even if you are protected one day the loss of a single disk drive over night can remove that protection.

If you want improved protection - replicate the data to a second EVA.

[Tristan Bunch]

Thanks Uwe, but unfortunately another EVA is out of the budget for the moment. I guess RAID 0+1 on two trays would be OK for enclosure resilience, taking into consideration the extra disk required. On the other hand, performance would be good! (or as good as 12 spindles can be)

[TTr]

> Is the only way to achieve...

Yes. With non virtual arrays we setup the LUNs vertically on the trays using one disk (or 2 for raid6)

> Is tray resilience something...

Trays are quite resilient and they do have dual power and dual disk interfaces so they very rarely fail. And their redundant components can be replaced live if they fail.

> out of the budget for the moment
If your data uptime is extremely critical then budget should not come to play. If it does, then you have to accept the posibility of some downtime even if that possibility is extremely small.

[Tom O'Toole · Accepted solution]

With four enclosures you would always lose access to your data in the event of a complete enclosure failure. What really determines it is the number of disks in the same RSS in the same enclosure. With only four enclosures, it has no choice but to use more than one drive from the same shelf. As previously mentioned, this is not that big a deal because the shelf redundancy makes this an extremely unlikely event - and upon repair, you will probably regain access to all your data.

Losing two drives in the same RSS (thankfully also very unlikely) might even be be a more likely event and this is certainly worse, because you will actually lose all the data in the disk group. You would have to go to VRAID1, and/or host based shadow to multiple arrays to further reduce your chances of data loss, subject to your budget of course.

[Tom O'Toole]

Forgot to add, the nice thing about EVA (one of the many nice things!) is you can easily choose vraid level on a per vdisk basis according to your business needs. If are able to identify more and less critical storage, it might be within your current infrastructure to do this.

[Tristan Bunch]

Thanks for your help guys, I have a better understanding of the EVA now.