Disable AO for a CPG

[brent78]

I want to have some VLUNs that are pure SSD, how do I tell the 3PAR not to AO a CPG?

[Darking]

I would suggest you create a new SSD CPG and move the volumes to that.

Regards

Christian

[Cleanur]

Assuming you want to continue using AO for some volumes, If you;'re going to pin volumes to an SSD tier. As well as creating a new CPG outside the existing AO config, make sure those pinned volumes are fully provisioned.

AO's job is to fill as much of the SSD as possible, if your pinned SSD volumes are thin and AO is still running against the SSD tier, it will very likely eat up all the space causing your thin volumes to run out of space in that tier.

[hdtvguy]

AO's job is to fill as much of the SSD as possible, if your pinned SSD volumes are thin and AO is still running against the SSD tier, it will very likely eat up all the space causing your thin volumes to run out of space in that tier.

Actually that is not what HP has told me over and over again, and was confirmed at fairly high level of the product team. AO's code is designed such that it wants to always push data down and that the settings in AO (Performance, Balanced, Cost) only throttle how aggressively AO pushes data down. It will move data up when it meets the criteria, but the way it is coded is designed for the legacy service provider model. We have been urging them to change AO to better utilize higher tiers. Our top 2 tiers are more then 50% empty, yet are NL drives get crushed with IO because AO does not take advantage of empty higher tiers unless the data meets the criteria even if it causes the lower tier to suffer in performance.

[Cleanur]

Thanks, but the point wasn't to get into a debate around AO's movement algorithms or suitability for your particular environment. If your data meets the criteria set by the algorithm, then AO will quite happily fill the SSD tier to 100%. I was simply trying to ensure the questioner understood what can potentially happen if they have both pinned thin volumes and AO operating on the same disks. In this case I think ensuring the solution is implemented correctly takes precedence over your particular gripe.