EVA Disk Groups

[Timothy Walsh]

Have 3 EVAs installed in our data center, these EVAs have multiple host machines attached. The host machines perform different functions, some are SQL servers, some Notes Mail Servers, and others are file servers. During our initial training and setup of the EVAs we were told a single Disk Group was the best practice and that's how we setup the first 2 EVAs.

When the 3rd EVA was recently installed the HP Field Service Engineers recommended multiple disk groups. That we'd see better performance from the EVA if for example the SQL servers had their own disk group, the mail servers had their own and the file servers had their own disk groups.

Looking at the white paper found at this link: http://h71028.www7.hp.com/ERC/downloads/4AA1-4202ENW.pdf?jumpid=reg_R1002_USEN

it appears that multiple disk groups have a benefit from a data protection and redundancy perspective, but might actually impact performance.

In the forum it appears that a single disk group is still favored, but some of the information appears to be dated.

So, anyone have any current thoughts on what the best practice would be for multiple disk groups for an EVA hosting SQL, Mail and File storage clients?

Thanks for your thoughts and opinions and any supporting links.

Tim

[IBaltay · Accepted solution]

Hi,

one single DG is still the best solution from the performance perspective, because thus you have all the logical disks distributed across all the spindles.

If you atomize the EVA disk back-end then you are suffering from the spindles.

But from the availability point of view when having multiple disk groups if any of them fails, it affects only its vdisks...

[Víctor Cespón]

There's no performance gain from separating the disks in several disk groups. The firmware is not multi-threading yet so all I/Os go to a single queue and if a disk group is slow, it will affect the traffic to the other disk groups too.

We have seen this many times on cases where customer calls complaining about performance.

There's a request to modify the firmware so each disk group is really independent of the others (rebuilding, leveling, vdisk management, snapshots, I/O operations, etc are done on each disk group regardless of what happens on the others). Until this is done, there's no real independence between disk groups.

[Mark...]

Hi,
The reason that some people get confused about disk groups etc is that there are 3 best practice type scenarios:
Performance
Availability
Cost.
What is best for cost is seldom best for performace and availability !!
Foe example best cost = Vraid0 but this is worst for availability
best perf = 1 disk group
Using data bases ? two disk groups 1 for data and 1 for log files (you can mix data and log files in one group but obviously not the same db/log files in one!) the list goes on.
I would suggest that you need to look more closely at how you use your EVA's and then check out a few whitepapers and/or advice in itrc to decide on how many disk groups are best for your situtaion/use.
Mark...