Create 2 diskgroups to prevent against EVA metadata loss?

[Fred van Zwieten]

If one puts an EVA into operation, It's metadata is stored on the first 8 disks of the first diskgroup. My SAN architect argues that 2 diskgroups is best for availability _because_ the EVA metadata is then replicated to the second diskgroup, thus preventing a EVA meltdown in the unlikely event of a diskgroup meltdown. Is this true?

A diskgroup meltdown with a single disk group will always cause a EVA meldown, but we are replicating everything with CA to a second EVA, so I argue using a single diskgroup is better for performance (spindle count) and capacity (protection level). Availability is allready covered though CA.

[Donald Kok]

Hi Fred,
The disks in the diskgroup are distributed on the shells. For a diskgroup meltdown, all shells the disk are on would be melt too. You do not have an operational EVA in that case anyway.

The proposal of the architect introduces a third level of redundancy. (CA over diskgroups over RAID). Seems overdone to me.
my 2cents
Donald

[Paul Henderson_2]

I'd agree with Donald on this Fred. Remember that for any diskgroup, a percentage is held back depending on the level of Vraid you are using. The SAN architect is introducing a level of redundancy that is already handled by the EVA, especially if you are using vraid1. And since you are using CA to replicate to a second EVA, you are triply covered.

My SAN architect, from HP, recommended always using a single diskgroup, if possible, to provide the best performance. We haven't had a hiccup in over a year of unbelievable performance (HPTC application), so your architect's suggestion is overkill to the EVA already designed overkill.

[Vincent Fleming]

Using CA, you're mirroring the arrays anyway - why bother with additional redundancy?

It looks like we're all in agreement here.

There are cases where having more than one disk group can increase performance, but we don't do that for redundncy.

Cheers,

Vince

[Fred van Zwieten]

Oke, guys, You convinced me. I'll forward this thread to my Architect.