HPE Storage Users Group

Hi

We have SSD, FC (10k/15), NL. We have created the following CPG and AO config using all disk types accross each OA config.

T0 = SSD
T1 = FC
T2 = NL

% showcpg
---------------------(MB)----------------------
-Volumes- -Usage- ------- Usr ------- ---- Snp ---- ---- Adm ----
Id Name
16 no-ao.01.t0.r5
17 no-ao.01.t1.r6
18 no-ao.01.t2.r6
19 ao.perf.01.t0.r5
20 ao.perf.01.t1.r6
21 ao.perf.01.t2.r6
22 ao.bala.01.t0.r5
23 ao.bala.01.t1.r6
24 ao.bala.01.t2.r6
25 ao.cost.01.t0.r5
26 ao.cost.01.t1.r6
27 ao.cost.01.t2.r6
--------------------
12 total

% showaocfg
-----------------------CPG------------------------ ---Warn(MB)---- -Limit(MB)-
Id Name T0 T1 T2 T0 T1 T2 T0 T1 T2 Mode
1 ao.perf.01 ao.perf.01.t0.r5 ao.perf.01.t1.r6 ao.perf.01.t2.r6 --- --- --- --- --- --- Performance
2 ao.bala.01 ao.bala.01.t0.r5 ao.bala.01.t1.r6 ao.bala.01.t2.r6 1331200 --- --- --- --- --- Balanced
3 ao.cost.01 ao.cost.01.t0.r5 ao.cost.01.t1.r6 ao.cost.01.t2.r6 204800 --- --- --- --- --- Cost
--------------------------------------------------------------------------------------------------------
3 total

I wonder if there is an argument for keeping things real simple with just 3 CPG's and one AO config.

The CPG's would be

CPG1_AO_T0_SSD (R5 3+1)
CPG2_AO_T1_FC (R5 3+1)
CPG3_AO_T2_NL (R6 4+2)

With a single AO config of set to balanced which contains all 3 CPG's

Reasons being:
-Most customers are unable to define which is their most important data and which should be prioritised. This setup allows AO to figure it out for you
-Even if you are able to define the value and performance requirements of your data now as with any work load this could change in the future. Again let AO control this for you
-By giving the data ability to move freely between all tiers of storage you won't have hot data stuck at a non optimised level where it is hammering the disks at the expense of everything else on the CPG. Vice versa you won't have cold data sitting on expensive disks doing nothing.
-In summary AO is generally better at deciding where data should be than we are
-Plus less CPG's makes reporting and general management easier

I agree on the simplicity front but 2 constraints have led me down the 3 AO policy path:

1. Not enough LFF shelves for cage availability meaning that any AO profile which includes the LFF NL drives would have to be set to mag availability. For top-tier storage (Gold) the business requirement is maximum availability so it has to be cage and therefore has to exclude the NL tier. This means there needs to be at least one other AO profile which does include the NL tier (and has all member CPGs set to Mag).

2. A requirement for top-tier storage to never be on NL. AO is reactive so a top-tier system suddenly accessing previously dormant data that was moved down to NL disk by AO is going to see some bad performance until AO kicks in and moves it up again, which depends on your schedule but in standard configurations is going to be a day. This is not acceptable for top-tier systems that require good performance all of the time. Fortunately the same tier (Gold) created for the previous requirement also fits in here.

These two examples actually mean you could get away with 2 AO profiles and 5 CPGs (2+3) but I thought we might as well create a third Bronze tier for stuff we know is not going to ever require performance so that AO can just move it down to NL and never put it on SSD. We can also schedule this AO to run last as you sort of have an AO "move window" with multiple profiles so it means that AO can concentrate on moving the other higher-tier data within its window and only bother looking at the Bronze tier if it has time.

This is all driven by business requirements though so the simpler approach might be suitable (and indeed preferable) in different environments.