Hosted Desktops Come of Age With Nvidia Grid and Citrix
- 24 May, 2013 13:44
It amazes me that it's taken about 35 years to invent what could be called the "better mainframe"-and even more amazing that it took a software and graphics company partnership to almost do it.
We've been chasing this elusive idea with thin clients, blade PCs and, most recently, cloud services such as OnLive, but each has had its own unique issues with regarding standards, costs, licensing, networking and user experience. The underlying goal was to provide something that approached a mainframe's capability to provide a utility experience to the user, but with a no-compromise, low-cost PC-type level of performance.
This may still prove elusive, but this week's Nvidia-Citrix Systems partnership built on Nvidia's revolutionary Grid platform suggests we're getting very close. Let's take a moment, then, to review the historical solutions that attempted to provide a PC experience that was close to a telephony experience and discuss why each failed to displace the PC.
Novell Hosted Desktops: Cheap But Slow
My first time using a hosted PC experience was working at Dataquest, which used Novell to host its desktops. This was cost effective. I recall AT&T (the old AT&T) coming by and boasting it could save us $1,000 per year per desktop by switching us to Windows 95; we laugh the reps out of the office because our total desktop support spending was around $450 a year.
Unfortunately, the Novell solution was also dependent on the network, and networks back then weren't very reliable. There were actually days when we were sent home because the network failed and we couldn't do any work until the failure was corrected. Dataquest eventually abandoned this solution; in addition to being unreliable, it was slow and couldn't keep up with the performance increases that defined the late 1990s.
Thin Clients: Great Potential, Poor Execution
Sun Microsystems' Scott McNealy and Oracle's Larry Ellison had a better idea: Build a PC appliance and provide an experience much closer to what you'd get on a telephone. Problem is, they had no idea how telephones worked and didn't appear to have the brains needed to spell "PC," let alone replace one.
In addition, they didn't seem to agree on critical things like pricing. Sun seemed to think folks would pay a huge premium for an appliance-like experience. However the efforts were marred by a lack of standards, huge pricing differences and that same inability to keep up with PC performance increases. Most pronounced was the tendency to use servers to host the workload. Servers simply weren't designed to handle the very different loads of lots of individual PC users.
The Sun Ray 1 was impressive, though, and the smart card authentication-which moved work, while retaining state, from station to station-demonstrated very well. That said, Sun employees tied to the system complained bitterly about low performance, and both companies exited the segment they founded. Currently Dell and Hewlett-Packard own the two largest companies selling into this segment, but thin clients realize but a fraction, in terms of market penetration, of their potential.
Blade PCs: Good Idea, But Too Slow, Costly, Immobile to Catch On
This idea, started by a company called ClearCube Technology, actually had the most promise in the late 1990s. You basically repackaged a PC onto a Blade and centrally located it. It was cost competitive with PC purchases of the time, providing similar performance while offering scale, security and reliability advantages-if one system failed, for example, the service would reboot onto another working system.
HP would later move on this opportunity, but the solution was known for underperformance. It was slow to move to a normal network model, as initial users needed dedicated wired links back to the PC blades. There were few cross vendor standards for blade or rack design, either. Finally, massive reductions in PC prices during the 2000s left PC blades too costly to compete. Mobile had become more powerful over time, too, and both the blade PC and the thin client approach lacked a strong mobile solution. Both Clear Cube and HP continue in this segment, but it's all but invisible.
PC as a Service: Cheap But Slow 2.0
Cloud gaming provider OnLive pioneered this most recent attempt to change the desktop experience. This service cost a fraction of the services that came before it and showed the potential to massively reduce the costs associated with providing a PC desktop.
However, network bandwidth and latency remain a problem. Plus, Microsoft licensing wasn't fully sorted out. Finally, OnLive had to develop its own cloud hardware because nothing in the market was designed to provide high-end desktop performance for users sharing resources. Internal development costs and Microsoft licensing issues were so harmful that OnLive had to restructure to keep from failing.
Citrix: An Expensive Hosted Desktop
Citrix was behind much of this, providing what was thought to be the most advanced software package for delivering a hosted desktop solution. However, this was mostly used to host difficult-to-port applications and was rarely used to supply the full desktop experience-even though that was initially the solution's primary focus. This was largely due to hardware limitations and Citrix pricing-which generally has been above, for full desktop delivery, the perceived IT value for the service. This has been a long-term problem for Citrix, holding back the company's growth and market performance.
NVIDIA Grid: Finally, Performance Isn't a Problem
NVIDIA Grid is the first attempt to provide cloud hardware that can provide full PC performance that scales for both the individual users and many users. Providing workstation levels of performance at peak times, providing shared performance for users with more modest current needs and dynamically switching between modes is the back end architecture that appears to address the long-standing requirement for server hardware that's designed from the ground up to handle the kind of loading that users uniquely require-which is vastly different than the loads driven by typical server applications.
This marks the first time a central technology provider of NVIDIA's class stepped into this market to provide a solution that could eventually be vendor independent.
NVIDIA, Citrix a Hosted Desktop 'Perfect Storm'
In the end we get the best software for a hosted desktop coupled with the best hardware. The price problem with Citrix continues, but that's as much a factor of perception as it is of cost choices. Given that Citrix is a software company, this can be addressed through both marketing (to increase perceived value) and price reductions (to better match current perceptions).
Neither path is mutually exclusive. In the end, this latest combination comes closest to the perfect storm of an enterprise hosted PC solution in the market. On paper, it represents the biggest step toward a hosted desktop.
There's little doubt in my mind that the PC desktop of the future will be hosted, as this will let users move much more seamlessly and safely across an ever-increasing number of screens and shift much of the support load to central resources that can be more effectively and cheaply managed.
We're still waiting for cloud services to scale to provide this broadly for small business users, but this latest move from Citrix and NVIDA, while targeted initially at high-performance workstation users, showcases that an enterprise, if it wants to go early, now has a workable solution. I expect prices to be sorted reasonably quickly, too-either by Citrix or a competitor.
Rob Enderle is president and principal analyst of the Enderle Group. Previously, he was the Senior Research Fellow for Forrester Research and the Giga Information Group. Prior to that he worked for IBM and held positions in Internal Audit, Competitive Analysis, Marketing, Finance and Security. Currently, Enderle writes on emerging technology, security and Linux for a variety of publications and appears on national news TV shows that include CNBC, FOX, Bloomberg and NPR.
Read more about cpu/processors in CIO's CPU/Processors Drilldown.