HPE Storage Users Group

Just got a new 7400 for an environment that will be almost exclusively for a Hyper-V 2012 R2 platform, a handful of other physical servers but not many. SQL DBs will be on this too as the SQL servers will be VMs. No plan to use any passthrough disks (RDM's for the vmware folk)

There are three storage tiers. Mainly FC, less NL, even less SSD, pretty standard stuff. We want to create tiers of storage performance for the VMs such that we have Tier 1, 2 & 3 for priority VMs, everyday servers and low-priority servers.

I am considering creating these CPG's and AO policies and wondered about some feedback.

TIER1_AO.PERF_FC.R5
tier1_ao.perf_ssd.r5

TIER2_AO.BAL_FC.R5
tier2_ao.bal_ssd.r5
tier2_ao.bal_nl.r6

TIER3_AO.COST_FC.R5
tier3_ao.cost_nl.r6

TEMPDB_ssd.r5
SNAPSHOTS_nl.r6

I was then going to create Virtual Volumes for Hyper-V in the associated CPG along the lines of:

TIER1_SYSTEMDRIVES_CSV1,2,3 etc.
TIER1_DATADRIVES_CSV1
TIER1_DATABASES_CSV1
TIER1_LOGS_CSV1
TIER1_TEMPDB_THICK_CSV1 <--- fully provisioned volume
...etc.

TIER2_SYSTEMDRIVES_CSV1
TIER2_DATADRIVES_CSV1
TIER2_DATABASES_CSV1
TIER2_LOGS_CSV1
...etc.

TIER3_SYSTEMDRIVES_CSV1
TIER3_DATADRIVES_CSV1
TIER3_DATABASES_CSV1
TIER3_LOGS_CSV1
...etc.

I'm setting this up for handover to SAN admins so I need to make it as clear as possible. I was hoping that the CAPS on the CPG names would make it simpler to see which CPG to create a VV in as I don't want VVs created in the other tiers of an AO policy. I'm aware that the TIER names here refer to how I want the storage to be tiered to the VMs, not the tiers that apply in an AO policy (0,1,2) so there could be some confusion there but I'm thinking that once these AO policies are setup there would be no need to worry about that, all that matters is that VVs go in the right CPG for their tier.

I was planning on giving the TIER1 AO policy access to more SSD than the the TIER2 AO policy so that TIER1 could become higher performance than TIER2.

Any comments on this approach? Caveats, better ideas etc would be appreciated.

Thanks

Here is what I would do:

CPG-1 SSD_R5_S4 (R5 3+1)
CPG-2 FC_R5_S4 (R5 3+1)
CPG-3 NL_R5_S4 (R5 3+1)

If you have main FC drives have the default CPG for all system be the FC_R5_S4

3par Raid 5 at 3+1 performs at like 96% the performance level of R1 unless you have real heavy IO system that thrash the 7400 cache I bet you will never see it. IF you are nervous about your databases then use the above tier for your VMs and create one more set of CPGs for your DBs

CPG-4 SSD_R5_S4 (R5 3+1)
CPG-5 FC_R5_S1 (R1)
CPG-6 NL_R5_S4 (R5 3+1)

Set this AO config to Performance and set the top one to Balanced.

One thing to keep in mind is that if you run your capacity tight you can run into issues as the AO configs know nothing of each other and you can fill up a storage tier. You use the CPG Growth Warning as a trigger for AO not to exceed.

thanks for the reply
if you see the AO config I have made tthe FC the target GPG for all the AO profiles. Also I've used RAID5 for SSD and FC as I agree about the minimal overhead compared to RAID1.

I don't like the idea of having any NL in an AO profile designed for performance optimisation. Anything dormant that suddenly becomes in demand is going to suck for a while and the idea of top tier machines is that performance never sucks. Also if you have system drive blocks that are rarely used moved down to NL and you need to restart the machine it's going to take a while to come back up. I think I might even remove the different CSV tiers for system drives and just make sure all C: drives are on the performance AO profile.

Usually we do this:
AO Performance
CPG_AO1_T0_SSD (R5 3+1)
CPG_AO1_T1_FC (R5 3+1)

AO Balanced
CPG_AO2_T0_SSD (R5 3+1)
CPG_AO2_T1_FC (R5 3+1)
CPG_AO2_T2_NL (R6 4++2)

AO Cost
CPG_AO3_T1_FC (R5 3+1)
CPG_AO3_T2_NL (R6 4++2)

No special CPG for Copy Space
No special CPG for TempDB
I use same CPG for Copy Space and User Space : tier 1 or FC.
There is no need of R1 for SSD anymore: customers don't use their performance at all... More space, more perf.

slink,

NL is not a bad thing as long as you do not crush them. The 3par caching is very effective. We recently did a DR test and our DR array has most of its data on NL due to inactive blocks. The performance ewe very good, of course it was not a full suite of apps running, but the occasional system restart will not be bad if the dormant blocks are on NL. NL gets bad when you crush it all the time with IO. That said with newer 10K drives now in 1.2TB size I would probably just fill an array with some SSD and 10K drives like we did with our 7400, it has 20TB in SSD and about 250TB of 900GB 10K drives and the array screams!

so there appears to be a difference in opinion about whether to have a single AO profile with 3 tiers or multiple profiles. The argument I think is that multiple AO profiles can conflict with eachother and a tier of disk can run out of space.

RitonLaBevue - Do you have an issues with your setup? Do you use the growth warnings to limit the amount of disk an AO profile can use?
Also, would it not be better to use higher set sizes in your Cost profile to favour space for a slight hit in performance?

Well, that's not my setup but the one we usually set with our customers... When they got 3 tiers ^^
We set warnings on tier 0 and tier 2 after OOTB, and 2 months later when the production is running we have a look at AO graphs to resize tiers.
No AO configuration is fixed for ever as production is moving all the time.

The key is not to have VVs out of AO CPGs on tier 2. Else you could run into some bottlenecks on NL drives...

ssz on tier 2 vary on how many cages your drives are spread.

Slink, I like your plan, I think you can optimize it a little more yet.

You have 3 different R5 SSD CPGs which you should try to consolidate down to 1. The same with your NL, try to consolidate it down to one as well. Your FC tier where everything starts will be where you distinguish Tier 1,2,etc. Avoid pinning volumes to SSD or NL and let AO manage these. I suspect that pinning your tempDB to SSD might be a waste of your dynamic performance potential, I suggest testing it in the tier1 cpg to see if the AO region mover will serve your application well enough.

Here is what I came up with:

CPG1_TIER1_FC-R5
CPG2_TIER2_FC-R5
CPG3_TIER3_FC-R5
CPG4_TIER0_SSD-R5
CPG5_TIER4_SATA-R6

AO_Policy1_TIER1_FC-SSD_PERF
AO_Policy2_TIER2_FC-SSD-NL_BAL
AO_Policy3_TIER3_FC-NL_COST

Thanks for you suggestions Richard but I think I'm missing something here. How can I be using SSD in two AO profiles (Perf/Balance) if I only have 1 SSD CPG? You can only use a CPG once in an AO profile. Same goes for the NL.

That would be me my mistake, not yours. Apparently I am confused as to how multiple AO CPGs would share the SSD disks and need to test more with multiple policies instead of just 1.