It seems that an ever-increasing number of companies are using Virtual Desktop Infrastructure (VDI). Gartner predicts that by 2012, penetration on the enterprise PC sector will be 60 per cent. The vendors which provide accompanying software and gear are likewise fighting for your investment, as evidenced by the flurry of activity on El Reg storage desk yesterday.
After a Reg piece on Tintri performance covering a recent test run at VMWorld Barcelona, an IT manager at a financial services firm quickly got in touch to say the article should have said more than it did about Tintri's hybrid flash/disk array, VDI-supporting gear.
To recap, Tintri is a start-up selling a hybrid flash and disk drive storage array specifically designed and built to support VMware virtualised servers and use cases such as virtual desktops. There's more about the background to the product and the company here. Suffice it to say it provides flash acceleration and disk drive capacity in a valuer for money, enterprise-class package in a hot market – witness other start-ups targeting this market such as GreenBytes and Whiptail.
The IT manager told us that his firm has bought Tintri for VDI, migrating from EMC CLARiiON and NetApp arrays with FlashCache. He explained the background to the decision:
"Over the years, we have undergone a migration from straight VMFS partitions on Fibre Channel LUNs on EMC CLARiiONs to NetApp hybrid Fibre Channel/NFS and now to Tintri.
"The biggest issue in the traditional array space is that it's very easy for high I/O to either totally saturate the disk or the write cache; write cache saturation on the NetApp has the knock-on effect of crashing I/O for the whole device, which means you have to have plenty of extra disk backing the storage, usually more than you actually need for capacity.
It seems that VDI vendors would do well to make things a little easier for their enterprise customers, as the IT bod commented:
"The device arrived and was 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; the Tintri is more of a point solution, but, so far, it seems extremely effective at what it does."
He added: "This is not just about booting the virtual desktops but about provisioning them from scratch. Booting 1,000 existing desktops ought to be much quicker. … [It's] an important distinction … between booting 1,000 VMs and provisioning 1,000 VMs."
He described the workload performance with reference to linked vs fat clones and cost per I/O in more detail:
"The linked clone vs fat clone distinction is an interesting one, as well. I'm not sure under what circumstances one would build out that many [1,000 say] fat clones, but, assuming that one is using some sort of custom script to spawn a bunch of clones from a single source, I would still expect it to be pretty fast. Arguably, you could do better with a NetApp with Rapid Deployment (assuming you can get it to work), but, in my experience, you get a lot more performance bang for the buck with Tintri than you do with NetApp. I'm not sure what NetApp's rated $/IOPS is, but Tintri comes out to a bit over $2/IOPS."
A contrary view
Max Gill, enterprise sales director of Atlantic Computing, doesn't favour the Tintri approach, or that of other similar start-ups. He tells the Reg:
"Customers are presented with a mind-blowing array of potential options from vendors out to replace the existing shared storage they own with something new. These range from local server SSD (Fusion–IO, OCZ), one of the all SSD array vendors – (Violin, Xtrem-IO, Pure Storage, Whiptail), one of the hybrid storage vendors (Tintri, Nimble storage), or a combined appliance/storage (Nutanix, Pivot 3 etc). In all of these options customers would stop using their existing storage and bring in something dedicated for VDI.
"They promote the same solution for both persistent and stateless desktops. With persistent desktops you are now using an unproven storage provider to store what is now critical data and with stateless desktops you are using expensive storage to store run-time transient data that never actually needs to be stored. In both scenarios user data and profile data is always saved off onto an existing CIFS/NAS share."
So what would he recommend for VDI? He has two primary deployment scenarios, both of which unsurprisingly incorporate his firm's product:
- For persistent desktops (where the data is important and needs to be stored) – use Atlantis ILIO to in-line deduplicate and offload I/O before it hits the existing storage array. Using the existing array, Atlantis ILIO will reduce the amount of storage you need to provision (for both performance and capacity) by up to 90 per cent. This is done through a combination of our I/O offloading, content aware caching, inline dedupe and coalescing engine turning thousands of small random writes into tens of larger sequential writes that shared storage is great at writing. As we are processing all of the writes on the same server as the VM, we are significantly reducing latency and improving performance. All of the extensive enterprise features (that all the new vendors are trying to write) can still be used – nothing is changed through the introduction of Atlantis ILIO.
- For stateless desktops (where the data is purely run time) - use Atlantis ILIO to present local storage (Local disk, SSD or RAM) and remove the need for shared storage completely. Should a server fail, the Broker (Vmware or Citrix) will just start the user a new desktop somewhere else in the estate. If performance is key for customers then Atlantis ILIO Diskless VDI (using RAM for storage) provides the best possible virtual desktop. With regard to how much RAM is required to store the VM – in our latest reference architecture with Citrix and Hyper-V we only needed 129MB per VM."
VDI users: feel the pulsating love from suppliers. It's coming in waves at you and is going to get even stronger. ®