HPE Storage Users Group

what is Best Practices to set up AO ?

There isn't really one, we went balanced running once a day with AO only considering the activity from 6 to 6 as we have minimal batch processing requirements and we didn't want backup jobs skewing the ao decisions. Other shops might want to optimize for performance or cost or might need better backup or batch processing performance so they might use completely opposite settings. You should be able to capture usage reports using system reporter to figure out when the array is busy and match that with your business requirements to figure out how to setup ao.

I second that, also be careful not to place “noisy” data on AO VV. For example, VMs that has been assigned insufficient memory. Although swapping VM’s runs great when AO has moved the data to SSD Tier, it will not be the most cost effective approach to say the least

afidel wrote:... and we didn't want backup jobs skewing the ao decisions

I have been thinking about this recently… I too have read and been told that one should not measure during the backup window but will this really affect AO decisions?
I mean, AO measures how often a region is being accessed over a given time. Sure, backup operations can generate a lot of IO but one specific chuck of data will only be touched once.

The array is supposed to not include backups in it AO calculation, my guess is it ignores those bulk sequential reads. That said we only have our AO look at our business day (7am-7pm) because we do not want regular system maintenance (DB re-indexing and such) which occur during the to impact AO calculations. We then let AO run until the next business day starts. I will say I have not been happy with AO since 3.1.2 when they moved it back into the controllers. They have made several AO "tweaks" since then and I do not feel AO works as well as it did prior to 3.1.2, we are constantly trying to force AO to do what we think it should do. What I will say is that I think the current AO is not very well optimized for multiple AO configurations unless you are going to use the CPG growth warning setting to skew AO the way you want.

I'm using AO on two systems, one is Inserv 2.3.1 and the other is 3.1.1

Your CPG strategy is an important consideration to your AO config.

In our environment, we have our CPGS layer out as:

PRD_TIER2 (FC, Cage Safe, Raid 5, optimal set size)
DEV_TIER2 (FC, Mag Safe, Raid 5 maxxed set size)
PRD_TIER3 (NL, Cage Safe, raid 5 optimal set size)
DEV_TIER3 (NL, Cage Safe, maxxed raid 5 set size)

Then we added SSD, and created 2 new CPGS:
PRD_TIER0 (SSD, cage safe, raid1)
PRD_TIER1 (FC, Cage Safe, Raid 5, optimal set size)

We then used Dynamic Optimizer to move selected LUNs we desire AO to consider for SSD optimization from PRD_TIER2 into PRD_TIER1. We never create luns directly in PRD_TIER0.

We then set an AO policy to be "performance optimized" and set the SSD CPG (PRD_TIER0) as the AO Tier 0 with a capacity = the amount of usable SSD, and the Tier 1 CPG as the AO Tier 1 CPG with the amount usable of 999999. The goal here is to have AO use 100% of the SSD for optimization. We do not have a Tier 2 set in the AO policy.