Until recently, the two biggest obstacles to VDI adoption have been graphics and storage. At long last, we have viable, relatively inexpensive solutions to both of these problems that will allow organizations to deploy virtual desktops for a much larger user base than ever before.
It took the efforts of NVIDIA and Atlantis to get us there (on the backs of many other people and companies through competition, I'm sure), but I think the argument can be made that, for the first time in its six-year history, VDI is now ready for prime time. Here's why:
Virtual GPU tech wows
At Citrix Synergy 2013, the demo showing off NVIDIA's new GPUs for complex graphics delivery had a lot of people excited. The virtual GPUs have been designed to work with XenApp and XenDesktop on XenServer (vSphere support is coming, but it was a Citrix show). In the keynote's demo, the presenters wowed us by manipulating a 3-D rendering of a car in RTT (an app that was previously all but untouchable through a remote desktop), a full-featured Photoshop experience and editing of a 4k video, complete with real-time effects in Adobe Premiere.
More from Citrix Synergy 2013
Citrix kills App Streaming, turns customers to App-V
Thin client providers focus on management
Citrix announces XenDesktop 7 with combined architecture
To be fair, the VDI graphics demo was with dedicated GPUs, possibly even on dedicated workstations running behind the stage, but still being remoted to laptops and iPads on stage using HDX. It was impressive, to say the least, and it highlighted NVIDIAs release of their K1 and K2 GPUs that can be dedicated or shared between virtual desktop sessions.
This addresses one of the two main challenges that still affect VDI. All data center-hosted desktops are plagued by a certain aspect that we call the resource triangle. With remote desktops, you can have good experience, low CPU and low cost -- pick two. But NVIDIA's technology effectively solves that by allowing you to deliver a good experience with low CPU at a low cost. Before you start talking about how expensive this will be and how the model still holds up, let's look at the cost breakdown.
The K1 cards retail for $2,287 (per Dell's configurator for an R720 server) and come with four GPUs on a card. The K2 cards have just two GPUs, but they are higher end and are listed at $4,178 on the same site. Each GPU can be shared by eight sessions, which means that a Dell R720 with two slots can conceivably support 64 VDI sessions (eight sessions with two cards and four GPUs per card).
That means that a solution using K1 cards can cost as little as $71 per user. Higher-end users that require a shared K2 GPU would cost as little as $261 (the cost of a decent discrete GPU anyway). Does it cost more than the status quo? Sure. But VDI isn't about saving money, remember?
Getting past IOPS
I also checked out Atlantis' demo of in-memory VDI, which was eye opening in person; I can see why Brian Madden developed an IT crush on the technology when he first saw it.
Atlantis' ILIO Persistent VDI uses the RAM that's built into a server as the primary storage for desktops. VDI hosts can have large numbers of desktops, each consuming tens of gigabytes of storage space, so Atlantis uses what's called in-line deduplication to deduplicate the redundant data on the fly. That's different from regular deduplication, which runs in the background on big iron storage solutions.
This results in an average storage footprint of less than 1.5 GB per VM. When placed in what is essentially a RAMDisk, VDI sessions absolutely scream.
In many environments, the challenge with sizing VDI comes from IOPS. We hear statements like "Windows 7 only needs 10 IOPS per user and 25 for a power user," which we believe despite knowing that generalizations are usually bad. Perhaps when you average out the desktop storage I/O over a long enough period of time you can arrive at these numbers, but active desktops require the IOPS they need, when they need them, and they won't be happy with 25.
Solutions for this so far have relied on big iron or solid-state storage, both of which are expensive and limited in performance by other factors. Some other more complex tools trump up hybrid storage, using SSD for the frequently used bits while putting the more static data on spinning disks. This technology turns that on its head and almost eliminates the need to even care how many IOPS each desktop needs.
With these two exciting additions to the VDI landscape, perhaps Citrix didn’t need to announce some terrific future product after all.