At Tintri I talk with a lot of customers and prospects about their virtualization environments and how it relates to their storage configurations. Virtual machine provisioning discussions come up quite a bit, so I thought I would write about some new features that Tintri just introduced.
The method in which we deploy virtual machines over the past many years has certainly changed on the storage side of the house. Thin Provisioned, Eager Zero Thick, Lazy Zero Thick; there has always been a long menu of choices when deciding how to deploy your virtual machine’s that support your applications. This has also created some confusion for people around “which choice is right for me when I deploy my virtual machine?” I have also noticed recently that many customers thought they had deployed thin provisioned vmdk’s but in fact they were running thick due to default values being selected.
First let me start off by saying Tintri is “pro virtual machine thin provisioning”. You might be saying, wait a second, you’re NFS on vSphere, you are thin provisioned by default! This is true, but with our VAAI implementation we can observe any of the other types of provisioning methods from VMware as well. Let’s say you do a storage vMotion and move an inefficient thick provisioned virtual machine from an existing block storage environment over to a Tintri VMstore. If VAAI is installed, we will observe the specifications of the existing format and retain this .vmdk format and punch zero’s. (unless you decide to change the option when migrating).
Let me make note, there is no need to use older “Thick” provisioning methods when deploying workloads on Tintri. Our VMstore operating system is designed to understand the workloads of every virtual machine down to an 8KB block. Tintri has QoS built into our datastore to adapt as your VM’s change from a performance perspective.
It’s all about Efficiency
With our new T800 platform, we have upped the bar on giving you more value from your Tintri VMstore investment. We have enabled compression at rest on all of the new models to help drive your storage costs down. This allows your organization to run as efficiently as possible from a capacity perspective. With our current shipping version of Tintri OS (220.127.116.11) we now add in some great capacity management features which I will highlight below.
I deployed a few VM’s for illustration in the lab, they are empty, no operating system, you can see some are eager zero thick provisioned, one is lazy, and one is thin in the screenshot shown below:
Here is the overall capacity of the VMstore prior to making changes to the virtual machine formatting:
In the example above you can see our compression ratio numbers are a little low, so let’s examine why. If a virtual machine is thick provisioned per VAAI, according to the specifications, you must “hard back” the zero’s, or reserve the space inside the virtual machine. If you were to thin provision the .vmdk file, then compression would allow us to reclaim the white space. This process typically involves doing storage VMotion so you can run the conversion process. Not any more!
Convert to Thin!
Tintri has built in some great ways to help examine and fix how you can optimize your virtual infrastructure. In the example below you can see the “Provisioned Type” field on the far right that I have exposed in our user interface to identify which VM’s are thick provisioned.
Let’s go ahead and right click and convert these VM’s within the Tintri user interface to thin disks!
This conversion process is instantaneous, and you can now see in the Tintri user interface we have converted our inefficient thick provisioned vm’s to thin without having to perform a storage VMotion.
You can see below the vSphere Web client now reflects an accurate savings on our capacity on each virtual machine:
Below you can now see the Tintri VMstore overall compression ratio is gone from 1.7x to 2.7x since we have migrated the virtual machines to thin provisioned vdisks!
Set it and forget it
Tintri has taken this one step further to help our customers (and thank you customers for your continuous feedback, this is a result!). We now have a global option within the datastore settings to keep all virtual machines that get migrated to Tintri as a thin provisioned regardless! No more going back to reclaim on accidental vm’s that were migrated over.
I hope you found this write up useful, let me know if you have any questions!