thin on thin provisioning, do you do it?

[Thuleman]

I currently do not thin provision at the array level, mainly because I don't have an implemented alert solution and I didn't see the need when I first started out. However, while I am still only at 40% storage utilization on the array, it is certainly filling up a bit faster than expected and I am considering to switch to thin provisioning on the array.

I read through Thin on Thin? Where should you do Thin Provisioning – vSphere 4.0 or Array-Level? @ Virtual Geek 04/09. Generally I find VG to be a very reliable resource, in this particular case though I wouldn't mind some feedback from our storage resident(s).

Thin on thin? Yes/No

 

[lopoetve]

... What's the array, and are you on vSphere?

 

[Thuleman]

The array is an EQL PS5000E, and yes, it's on vSphere.

 

[geiger]

Really depends on your storage mentality, and duty structure.

I am the virtualization and storage admin. I have dedicated disks set to virtualization and manage it as such. I do thin provisioning at the virtualization level. I have my vmdk's balanced over datastore types that reflect workload.

I am also a DBA, so thin provisioning at the storage level throws up all sorts of red flags for me.

Here is a simple way to look at the 2 methods:
Thick: I/O intensive Applications(DB, Email, etc)
Thin: Variable capacity(File Server, Non Critical OS, etc.)

anecdotal, but relevant....
On our first san(cx-400) we were capacity and budget bound. We ran oracle and exchange on the same disk pools due to storage and budget limitations. Performance was horrid because we tried to lump together storage requirements based on capacity requirements.

 

[lopoetve]

I need to check to make sure equallogic thin provisioning is supported in vSphere.

Thin works. It's an awesome tool. It's also something you can shoot yourself in the foot with. You HAVE to be careful. In addition, if it's something where large amounts of data are going to be going in and out regularly after the initial install, you're not going to gain anything - vmdks grow, they do not shrink. Plus, you've now introduced the ability to have fragmentation in VMFS - if it's something where you do the initial install and app install, and then only a little data is added afterwards and it just runs, it's not an issue. If it's a DB that you install and app install, and then 3 weeks later it goes live and takes a huge chunk of data and keeps swapping, not only have you saved no space but you've got your drive broken into two parts. This may not matter, as the OS and data portions will be contiguous-ish, but it is fragmentation. Especially as you get close to full on the datastore, this can get you in trouble. And don't forget that you still want 10% free minimum at all times.