2. Desktop virtualization
Desktop virtualization has been with us in one form or another seemingly forever. You could probably even say that it's been emerging since the mid-1990s. But there's more to desktop virtualization today than most of us could have imagined even two or three years ago. Yet another milestone is just around the corner: truly emergent technology in the guise of the desktop hypervisor.
Long the leader in this space, Citrix System's XenApp and XenDesktop are examples of how desktop virtualization just might put a desktop server farm in every datacenter and a thin client on every desktop. XenApp weaves together all the prevalent desktop and application virtualization technologies into a single package: traditional application and desktop sessions, application streaming, and VDI (Virtual Desktop Infrastructure). No matter which way you turn, the detriments of each is generally backed up by the benefits of another.
The client hypervisor takes desktop virtualization the last mile. Picture each desktop running its own bare-metal virtualization layer that abstracts the baseline hardware to whatever VM you wish to push to the desktop, where it can be centrally managed, synced with a mirror on a server, and easily replaced (or even reset by the user) when things go wrong. Citrix isn't alone with this concept -- VMware is developing a similar solution, and both promise to hit the market in 2010.
Regardless of what solutions are available today and what solutions may be on the horizon, enterprise desktop management remains one of the biggest points of pain in any organization. While the model for datacenter architecture has changed systemically in the past 20 years, the model for deploying desktops hasn't. In most places, it's still one fat box per user, with some mishmash of management tools layered across the top to protect the users from themselves and protect the network from the users.
Whether any of the desktop virtualization technologies are applicable to your enterprise is wholly dependent on the nature of the business. Call center and health care treatment room terminals are a relative no-brainer, but you can quickly run into problems with noncompliant applications in other implementations. As the blend of desktop virtualization technologies reaches a critical mass, the wide variety of ways to ship a Start menu to a user offers a better chance that at least one will apply in every instance. Certainly, if the world turns its back on fat clients at every desk, IT will be a happier place. As for the users, the client hypervisor may give both IT and the most ardent fat client holdouts what they need.
Why on earth would InfoWorld pick a programming framework for distributed data processing as the most important emerging technology of 2009? Because MapReduce enables enterprises to plunge into analyzing undreamed of quantities of data at commodity prices, a capability that promises to change business forever.
IDC has predicted a tenfold growth in digital information between 2006 and 2011, from just under 180 exabytes to 1800 exabytes (that's 1 trillion and 800 billion gigabytes!). This explosion represents a challenge, of course (how to store, retrieve, and archive all that data), but also a huge opportunity for enterprises. After all, everything in that sea of data is potentially information -- information that could be used to guide business decisions.
Until recently, enterprises that might want to process petabytes of independent data to find business-relevant relationships would need an extremely good reason to invest in such a venture; the costs and time required were prohibitive. But this is quickly changing as enterprises begin to adopt highly distributed processing techniques, most notably MapReduce, a programming framework that has enabled Google, Yahoo, Facebook, MySpace, and others to process their vast data sets.
In its simplest form, MapReduce divides processing into many small blocks of work, distributes them throughout a cluster of computing nodes (typically commodity servers), and collects the results. Supporting highly scalable parallel processing, MapReduce is fast, cheap, and safe. If one node goes down, the lost work is confined to that individual node.
Google introduced the MapReduce framework in 2004, but there are many implementations today, including Apache Hadoop, Qizmt, Disco, Skynet, and Greenplum. Apache Hadoop is the leading open source implementation. Amazon taps Hadoop to offer MapReduce as an Amazon Web Service. Cloudera, which bills itself as offering "Apache Hadoop for the Enterprise," is making significant inroads.
Support for MapReduce programming is also delivered in several enterprise software products such as GigaSpaces eXtreme Application Platform, GridGain Cloud Development Platform, IBM WebSphere eXtreme Scale, and Oracle Coherence, to name a few.
The inexorable growth of data is a fact of life. As vendors drive the MapReduce framework into product offerings, we have a new window into what all those petabytes mean. It's difficult to imagine how, just 30 years ago, businesses could function without the benefit of business intelligence software or even spreadsheets. When MapReduce becomes part of the culture, business strategists in the not-too-distant future may look back on our era in the same way.
-- Savio Rodrigues