Wednesday, May 5, 2010

To centralize, or not to centralize

You wouldn't do this with eggs. Why would you do it with your company's desktops?













Desktop virtualization holds great promise to dramatically reduce IT support costs, while allowing end users unprecedented access and flexibility. There are now dozens of offerings to choose from. But beware – your approach matters. A lot.

Picture the last time your PC crashed. Now imagine it happening to everyone in your company. All at the same time.

It makes complete sense to centralize certain aspects of desktop management. Access and policy control, reporting and image updating, definitely. And certain compute-intensive applications. But it makes little sense to centralize execution.

Desktop execution should (almost) always be decentralized.

A decentralized system is inherently more resilient than a centralized one. No single incident can bring down the productivity of the whole. This has been a truism in computer science, and it’s proven itself in many other systems, not least of which is the Internet itself. On the Internet, no single router or gateway failure can bring down the connectivity of all the endpoints. Instead, in the case of a failure, packets are rerouted around the downed node and transmissions successfully proceed.

Similarly in a decentralized desktop environment, a single PC failure, or even the failure of the management server itself, does not stop the productivity of the whole. Sure, a single user may be inconvenienced (and we all know that certain users are more important than others :) ), but there is no chance that the entire system can come down. When we say “no chance," we don’t actually mean “low probability,” or “five 9s,” etc. With decentralization, there is *no* chance of systemic failure. Nada.

Moreover, a decentralized desktop system is usually lower cost because consumer CPU and storage is much cheaper in aggregate than the equivalent resources in the datacenter. And decentralized execution provides the best user experience, since the user can be online or offline, does not have to worry about bandwidth, and local CPU provides better performance than a centralized remote one.

Now, certain narrow use cases do warrant centralization. But the vast majority of desktops should remain decentralized.

Something to chew on

Even with 14 years of experience and a bazillion dollars, Google’s search services went down last May. What’s the likelihood that your VDI will fare better?





Caveat VDI

Your existing physical desktop environment is already an inherently resilient system. Your company (and your career) can easily survive the occasional user hard drive crash or network issue. But now you’re thinking about scrapping that beautiful architecture, and replacing it at enormous upfront and ongoing costs with an inherently more fragile and risky one. There may be a legitimate cost-risk-benefit reason for you to do this – just be sure you’ve done the analysis.

A better way

Distributed Virtual Desktops (DVDs), a term coined by IDC, represent the low cost, highly flexible world of the New Desktop (capitalization intended). In this model, like your physical desktop solution, DVDs are controlled centrally, so access control, policies, reporting and image management happen efficiently by a single team. But desktops are virtualized so the same golden image can run on any platform, and any hardware configuration. This makes it better than your current physical desktop solution.

And, in stark contrast to VDI, DVDs execute in a decentralized fashion. This means that issues (and we all know issues can happen) are isolated to a single user. As with VDI, there are multiple offerings in the DVD space—MokaFive is one. At MokaFive, the tenet of decentralized execution has been imbued from the very beginning and throughout every aspect of our product design. We believe it’s the only way to go.

Burt Toma, Director of Products

No comments:

Post a Comment