VDI has been a hot topic recently, and it’s certainly something we see with growing demand out in the field.
We recently conducted a series of tests with VMware at their labs in Palo Alto. And, we were really pleased to see Chris Mellor at The Register pick up on a VDI story that was introduced to him by a customer of ours that wanted to get something straight: Tintri is not just about booting the virtual desktops, but about provisioning them from scratch.
With all of that in mind, we wanted to share some of our own thoughts on VDI, and how it relates to storage.
Storage is not simply about performance. While it’s certainly a major factor in today’s storage discussion – notions like “slap more disk on it” or “throw some flash storage at it” try to address the pure performance problems – there’s an equally important factor that is too often overlooked by storage solutions: manageability.
How can you make your life simpler as a storage admin, sysadmin or a virtualization admin when it comes to storage in the virtual environment? Have the storage do more of the tedious work for you.
If you are working with traditional storage abstractions like LUNs and volumes, you continually have to put all the elements in the right place and carve them out just so. With Tintri VMstore, you have just a single datastore to worry about; there are no LUNs or other storage abstractions with different performance characteristics — just VMs and virtual disks. This effectively eliminates storage configuration and management issues—or, as our customer put it:
Tintri arrives racked, stacked, and configured for connectivity in under an hour by a sysadmin who'd never even heard of it. There are definitely higher-performance solutions out there, but most of them come with significant administrative overhead or have other limitations.
What’s the secret sauce to making our system work for virtualization and VDI?
A file system that is purpose-built for virtualization!
Here are some of the features that make Tintri truly VM-aware:
Cloning and Snapshots: Array side cloning and snapshotting of individual VM base images (that is, cloning with per-VM rather than per-volume granularity) helps with overall VDI performance and capacity. Tintri also simplifies the choice between linked clones, full clones, or alternative image management techniques such as those from Unidesk. With Tintri, administrators can be confident that any of these techniques will perform well and will be simple to administer. With traditional storage, full clones take longer to create and end up taking more space, but by offloading the cloning to a Tintri array (using Tintri’s VAAI plugin) these concerns disappear.
Predictable VM performance: The Tintrí VMstore file system delivers optimal flash levels of performance for each VM without manual configuration or VM placement. Traditional storage must place base-image replicas and virtual servers into specially configured volumes, and mandates that replicas and servers must never be placed into the same datastore as virtual desktops, because their quality-of-service needs differ so wildly. Only Tintri allows replicas, desktops, and even virtual servers to be mixed within the same datastore without any special tuning or configuration. Only VM-aware storage dynamically self-tunes and allocates exactly the resources required by each independent virtual disk, regardless of type (replica, linked-clone, desktop, or server).
Inline de-duplication, data compression and working-set analysis: Tintri provides data inline de-duplication and compression as it arrives. This greatly reduces the cost of flash performance. All applications (including virtual desktop applications) have a significant amount of “cold” or infrequently accessed blocks. Tintri’s working-set analysis allows infrequently accessed 8 KB blocks to be “paged back” to hard disk drives rather than wasting space in flash storage.
Instant performance bottleneck visualization: Real-time VM and vDisk-level insight on IO, throughput, end-to-end latency and other key metrics enables rapid VDI performance diagnosis.
We don’t actually know which customer went to bat for us, as they wish to remain anonymous. We thank them all the same for their support, and for making it clear that although performance is an important functionality, manageability is a larger factor to consider when optimizing storage performance for VDI.
Unique control with VM-level actions for infrastructure functions including snapshots, replication and QoS make protection and performance certain in production, and accelerate test and development cycles.