Thin provisioned luns on netapp, thick vmdk disks.

[ecleppe]

Dear all,

We run our vmware environment on Netapp filers with deduplication enabled.  Our LUNs are thin provisioned on the storage side.  As I've read documents in the past that thin on thin isn't that good performance wise, the decision has been made to use thick provisioned vmdk's for our virtual machine.

Logical problem now is that when having 8 VM's on a datastore of 500 GB, the datastore is in alert state because there is only a fraction of space available, but on the Netapp system manager it only shows 20% in use (example).

What do you do in this case? Do you ignore the alerts in VC? Disable them or something?  What should I refer to?

Thanks for any advice,

Erik

[AndreTheGiant]

You cannot ignore the problem.

Because on datastore (at VMware side) you may  not have enough free space for snap or swap file.

Increase the datastore  to have more logical space.

Andre

[ecleppe]

But my luns are provisioned with 510gb space, my vmfs datastores are 500 gb. 
For Netapp only 150 gb space is in use, but for vmware it's 450 gb.

If I grow my datastores, I have to grow my luns on netapp aswell eating up all the space on my aggregates for nothing...

[AndreTheGiant]

If you use thick disk, on VMware side, you will see pre-allocated space, that isn't the real used space.

Or you use thin disk also on VMware side, or you have to pre-allocate less space (or have a bigger datastore).

Andre

[vmroyale]

Hello.

As I've read documents in the past that thin on thin isn't that good performance wise, the decision has been made to use thick provisioned vmdk's for our virtual machine.

Just out of curiosity, which documents have you read that suggested this, and what is the workload that you are using the thick VMDKs for?

Good Luck!

[ecleppe]

Hi,

Basically we have 90 % virtualization here.  AD / Exchange / SQL / etc.... just some components from our Lync environment are not cause they are passing audio.

At this moment we have every vmdk thick (drives bigger than 100 / 150 gb are RAW device mappings anyway).

And our datastores are thin (storage).

I'll look up the document and will post a link when i find it again.  It's not something that I've read recently tho, so it might be outdated.

Rgds,

Erik

[BobL201110141]

I ran into the same issue and did some researching. I am running NetApp version 7.3.5.1 and ESXi 4.1  You can over allocate volumes/luns that are advertised out to the VM environment be setting your space setting to none and fractional reserve to zero. In my scenario I also enable the 'auto size' options of "allow snapshots to grow automatically" and snapshot auto delete" but this will depend on your snapshot retention policy.

So I have a filer that has 3.2TBs of physical space. I have 5TBs worth of Volumes/luns that are advertised as space as datastores to VM. All of my luns/datastores say they have more data than is actually on the disk due to either thin/thick provisioning and utilizing the dedup on netapp. At the aggregate level I still have 2.5 TBs of space. This means I can continue to grow my storage to to make VMWare happy and manage my space at the aggregate level and add more storage to the filer if needed.

Hope this helps

Bob