I’m stepping out of my normal NetWorker zone here to briefly discuss what I think is a fundamental flaw with the current state of thin provisioning.
The notion of thin provisioning has effectively been around for ages, since it’s effectively from the mainframe age, but we started to see it come back into focus a while ago with the notion of “expanding disks” for virtualisation products. Ironically these started initially in the workstation products (VMware Workstation, Parallels Desktop, etc.) before starting to gain popularity at the enterprise virtualisation layer.
Yet thin provisioning doesn’t stop there – it’s also available at the array level, particularly in NAS devices as well. So what happens when you mix guest thin provisioning in a hypervisor with thin provisioning at the array/NAS level providing storage to the hypervisor?
Chaos.
Multiple layers of thin provisioning is potentially a major management headache in systems storage allocation. Why? It makes determining what storage you have available and allocated, when looking at any one layer, practically impossible. vSphere for instance may see that you’ve got 2TB of free space in storage that’s currently unallocated, and your NAS may be telling it there’s 2TB of free space, but it may actually only have 500GB free. Compounding the issue, the individual operating systems leveraging that storage as guests will also each have their own ideas about how much storage is available for use. One system suffering unexpected data growth (e.g., a patch provided by a vendor without warning that it’ll generate thousands of log messages a minute) might cause the entire thin provisioning sand castle to collapse around you.
This leads me to my concern about what’s missing in thin provisioning: a consolidated dashboard. A cross platform, cross vendor dashboard where every product that advertises “thin provisioning” can share information in the storage realm so that you, the storage administrator, can instantly see an exact display of allocated vs available real capacity.
This isn’t something that’s going to appear tomorrow, but I’d suggest that if all the vendors currently running around shouting about “thin provisioning” are really serious about it, they’d come up with a common, published API that can be used by any product to query through the entire storage-access vertical. I regret to say the C-word, but it’s clear there needs to be an inter-vendor Committee to discuss this requirement. That’s right, NetApp and EMC, HDS and HP, VMware and Microsoft (just to name a few) all need to sit at the same table and agree on a common framework that can be leveraged.
Without this, we’ll just keep going down the current rather chaotic and hazardous thin provisioning pathway. It’s like an uncleared minefield – you may manage to stagger through it without being blown up, but the odds are against you.
Surely even the vendors can see the logical imperative to reduce those odds.
Disclaimer: I’m prepared to admit that I’m completely wrong, and that vendors have already tackled this and I missed the announcement. Someone, please prove me wrong.