StoreServe 7200 volume sizing best practice

[Abstraction]

I've been trying to find some documetation to help me figure out how best to carve up a new 7200 into LUNs for VMware datastores, but not finding much help.

I have a full shelf of 24x900 GB drives in a raid 5 CPG (default FC_r5 CPG).  It shows up as 19.6 TB total space.  I made 3x5 TB virutal volumes to present to VMware hosts and thick provisioend them.  Now it states that I have 18.4 TB in volumes and CPGs, 250GB in system internal, and 819 GB in system spare.  I guess the 900 GB disks actually show up as 819 GB.  I get a critical warning that FC raw usage is above 95% with only 155 GB free.

 Since this is thick provisioend it used up all the space, so is this really a bad thing?  Those volume shouldn't grow any more. I wont be doing any snapshots.  Should I remove those volumes and recreate them as 4 TB ones in order to clear that alarm?

[Cali]

Most People doing TP on 3PAR, but if using FP I configure the Volumes to use 94%.

It is sometimes a bit tricky to reach it.

[Abstraction]

I'm only doing FP since that was what people at this organization before me had done.  I think mostly because they were afraid of running out of space if they over provisioned.  We don't have dedicated storage admins here, so the VMware guys (me) have to double hat.  Since we aren't experts and have a lot of other things to worry about, having storage be lower on the things to worry about list was the goal  I think.

[Sheldon Smith]

---

@Abstraction wrote:

I'm only doing FP since that was what people at this organization before me had done.  I think mostly because they were afraid of running out of space if they over provisioned. 

---

That's what HPE StoreFront Remote is for and it's free!
First, go to https://www.storefrontremote.com/ and log in with your HPE Passport. Then take 12 minutes and review the first three help videos (at least): Introduction, Overview and Registering Your System.
Follow the instructions to register your system and allow the 24 to 48 hours mentioned for the registration token to filter back to StoreFront Remote. Now go back and watch the remaining Help Videos at StoreFront Remote.

You can leave the datastores as fully-provisioned volumes and have the ESX hosts manage the thin provisioning, or you can offload the thin provisioning management by making the datastores be thinly provisioned and using Thick Eager Zeroed VMDKs (an HPE Best Practice). You could use thinly provisioned volumes and thin VMDKs, but that's just adding administrative overhead on your end that would have very little benefit.

Since you have used up most of the storage, you would need to free up what you can. Note that when used physical disk chunklets are freed up, they are automatically sent to the reinitialization queue where they are cleaned for reuse. This can take a while. I would suggest deleting any empty datastores, and look for any VMs that could be vMotioned from the least-filled datastore to the other current datastores. Then create a 5TB thin volume and start moving what you can from a fully provisioned datastore to a thin datastore, having VMware change the VMDKs to Thick Eager on the way.

Repeat as the chunklets are initialized until all your datastores are thinly provisioned.

[Abstraction]

A delayed thank you for this insight.  I had oftened wondered what the best practice for thin volumes/datastores vs VMDK thin provisioning was.  I had figured doing thin on both was a waste, but wasn't sure which one was best to do it on.  Of course if HPE recommends thin on their volumes they are a bit biased.  VMware might say the opposite.  I think I'll go with your recommendation though.  Luckily this is a new array that I haven't put any data on yet.  I can easily blow away the thick volumes and reprovision them as thin.  I can convert the VMs to think eager zero as I storage vmotion them onto the new storage from the old lefthand iSCSI array they live on now.

Edit:  Thank's for the storefront remote link; however, all of my systems are on non internet connected networks.  I have to do the old fashion morning checks.

[Sheldon Smith]

Too bad your systems' Service Processors are not enabled to report back. HPE provides advanced performance and monitoring analytics with improvement being made constantly.

You can do the thin provisioning on either side, 3PAR or ESX. They do the same thing; thin is thin. Main thing to remember is ESX has to use Host CPU cycles to perform thin provisioning. The 3PAR does it with dedicated ASIC chips, offloading the provisioning overhead from the ESX Hosts.

As I wrote above, as you delete a fully provisioned volume, the chunklets will be flagged Uninitialized and need to be cleaned before the system will use them again. The CLI command "showpd -c" will show how many are flagged Uninitialized. Wait 10 minutes, repeat. Then do the math for a rough idea of how long it will take for all of them to be initialized.