Tuesday, November 01, 2005

HP, EVA's and databases

Getting the next generation disk array from HP and I'm trying to figure out the best way to configure it. The EVA has a new concept called "disk groups" that I just can't get my hands around. The lowest level of storage that the EVA uses is a Redundancy Storage Set (RSS). The EVA basically groups your disks into these RSS's based on the number of disks in your disk group.

The "disk group" basically sits on top of the RSS and is a logical collection of disks. The disk group is stiped across all disks in the disk group. Each "virtual disk" (basically LUN) is then created from a single disk group and assigned a "vRaid" level. You can create as many "vDisks" in your disk group as you need. A single vDisk can not span multiple disk groups unless you use a LVM like vxvm.

The documents I'm reading (here, and here) are conflicting at best. One says that a database server should have one disk group effectively implementing SAME (stripe and mirror everything) for you. The other document says to configure the number of disk groups for the type of I/O you will encounter. Knowing that the redo logs will be sequential access and the datafiles will be a mix of random and sequential access, I think there should be at least two disk groups.

Performance wise, every fiber of my being tells me one filesystem just isn't going to cut it no matter how many physical disks are under it. Do I really want all 48 disks spinning for a 8k write?

One of my fears is that if I lose a disk group, I lose everything. HP, of course, says you will never lose a disk group. Experience on other HP products tells me the first time we have a failure, we'll get some bad advice and wipe something out. Am I just being paranoid about losing a disk group?

Does anybody have practical experience on HP EVA's that can offer some insight?


Quadro said...

We do separate LUNs for redo/data files even on HP XP 1024. We also have EVA 6000 with separate VG.

This configuration based on experiences from another our sites.

Anonymous said...

We went through this not too long ago.

A group of sys admins at another location already had an EVA in their data center and had set it up with the one disk group approach.

We were working on a project in another data center and the project had budgeted to purchase an EVA for this project and future uses.

We argued for months against the one disk group approach. An HP rep. and the sys admins from that other business had convinced the project leaders and sponsors that more than 1 disk group was not needed and would increase the budget too much to buy the extra disks needed or if we split the disks we had purchased into more than 1 group performance would suffer to the point of making it not usable.

Thankfully that EVA to this day has not had any issues and is in the process of being relocated.

The original one I mentioned above however did lose the one and only disk group it had for multiple production databases.

I never did get the full explanation of what happened but one disk group is no longer an option and on some of our more critical systems one EVA is not an option either.

quadro said...

"The original one I mentioned above however did lose the one and only disk group it had for multiple production databases."

Two in a row (horrible thing). About two months ago our EVA 6000 also lost the entire VG - in response to some steps told us by HP engeeners.

Thankfully enought we already migrated our 24x7 DB to XP 1024 and was able to restore VG from Data Protector without massive lose of service. One day of a downtime for our reporting system.

Jeff Hunter said...

Thanks for all the responses. That confirms what I originally thought.