HPE Storage Users Group

So im very new to 3par, and AO.

And i am trying to wrap my head around the AO function of the product.

I have an installation with 32 450GB mLC SSDs, 40 900GB FC 10K, and 12 4TB NL.

I would like to utilize as much of the flash as possible, and have therefore created 10 thin vluns at the moment, using the default FC_r5 CPG. i have setup AO on the CPG, and told the AO configuration it should also use the SSD_r5 as tier 0, and NL_r6 as tier 3.

From what i can see of datamovement at the moment, most of the data flows down to NL, which i suppose is expected. But i would like to have more utilization of my SSDs really.

When looking at my current CPG allocation it looks something like this: 60GB on SSD, 700GB on FC and 1TB on NL.

Since i have so much Flash, i would have assumed that capacity wouldnt be "wasted" by not being utilized..

I know that AO needs to look at the data for some time to make its moves, i had just hoped more would be moved the the higher tier.

You can do a few things. First have you AO schedule look only at the busiest time of the day. We run our AO at 7pm and tell it to only go back 12 hours. This way it only make data moves based on our business day. Next set the AO configuration Performance mode this way ti will want to push more data up. The last step is tricky, you use the CPG growth warning settings to help guide AO to where you want to land more data. You can set a growth warning (note warning only, never set the growth limit) which will tell AO to not move data into a CPG if that CPG is at size that the growth warning is set to. We do this on out NL tier to help AO move more that it normally would to higher tiers.

Some things to keep in mind. If a (non NL) CPG is full in that the underlying tier has no more space the array will not deny writes if it has available NL drives, it will write to NL to prevent a denied write. This is not ideal, but better than crashing an attached host due to a denied write. AO 3.1.2 has a bug where the AO task may not be able to go back far enough depending on how much meta data you have (volumes, tiers, etc). We do find that in general even with Performance mode set the array moves a lot of data downward. Could be validation that many blocks do not benefit from higher speed tiers. The downside that we have discovered is that sure these blocks are lightly used and woudl be fine on NL in general, but when you look at the size of most NL tiers in proportion to the upper tiers the issue you wind up having is contention for the NL drives because while the overall number of "infrequently accessed" blocks is high the overall times NL gets hit is high due to the size in comparison to the smaller faster tiers. This lead to over-subscribed NL drives in many cases and thus has a potential impact on overall performance.
I believe 3par is aware of this concern will eventually come up with a solution, but be careful how busy you NL tier gets.

Be careful with growth warnings, I used them yesterday to free up some space for a new AO policy and the existing AO policies ended up completely emptying the tier! I had to run a second run of AO to get data back into the SSD tier but it definitely did not do as good of a job as it had previously as latency has been higher today than in the recent past. I'm hoping tonight's AO run will smooth things out a bit.

Hmm, well AO move a lot of blocks around but the one AO policy that had taken all the space last night failed to give any up so my other policy that had hot blocks to move up to Tier0 was unable to due to lack of space, any idea how I can get AO to limit the amount of space without completely emptying an entire CPG?

According to 3par setting a growth warning does NOT cause AO to move data out of a tier unless AO would have moved it out anyway. If your tier emptied it was because AO determined the data in that tier did. It need to be there any more and the lack of new data was due to the growth warning not allowing any new data that AO would have wanted to put there.

I guess i just think it would make more sense to atlest be able to do what i mentioned.

My array is ment to last me for a certain timeframe, and we normally buy the capacity we need in advance, and dont do intermediate expansions

Therefor i have a great amount of flash that could easily fit all my production data, but i want to be able.. over time to tier the less used data down to slower tiers.

I had hoped there was some way to do that, without having "day-to-day" administration of storage. My resource is not really dedicated to handling our storage, and therefor i expect my solution to work as optimal as possible at all time (hey who doesnt want that).

I assume i can achive what i want buy going away from AO and just use Volume tuning, to move all data to an R5 SSD tier. and then reevalute as our data grows.