HPE Storage Users Group

AO seems designed to adapt to the previous day's workload on the assumption that the following day will be more of the same but with better performance due to what AO has done with the blocks.

What about if you have fairly predictable trends in workload requirements such that Monday sees some systems used heavily, Tuesday has a different bunch, the weekend is a whole lot of other stuff and then it repeats? Seems that lots of business processes fit a weekly or monthly recurrence rather than daily.

Is anybody setting up AO profiles which do the analysis some days in advance of the actual move and if so, how's that working out?

Depending on how many volumes you have doing weekly AO would be difficult. the system only tracts so many data points, and actually in 3.12 at some point it could not even keep 12 hours of data for our system. IT has enough issues dealing with a day or data and having enough itime to move the data between tiers.

well that's a shame. I'm yet to implement AO, currently planing for what the initial config/schedules are going to look like and I thought that AO might be able to work in this way given the options which are available in the IMC to start and stop the analysis several days in advance of the actual move. The implication is that AO can store the analysis information, calculate the moves but delay it to the specified schedule. I guess I'll see the reality of AO's effectiveness when I start actually applying it.

AO has limits, like most technologies and depending on the size of your array and volume count you may hit those limits.

If you have a small subset of LUN's that get busy on a schedule you could script a DO move of the affected volume into a higher performance CPG for the busy period and then move it back after the run.

Not workable with a large number of apps obviously.