Virtual unrealities exposed

…entirely different hardware. This entails a big performance hit, however. VMware and Microsoft's Virtual Server, by contrast, only virtualise enough of the hardware to allow operating systems to run in isolation, but the operating system has to be designed for the hardware it's running on. Such an approach still entails a significant processing overhead, however.

Some approaches, such as Solaris Containers, BSD jails and SWsoft's Virtuozzo, virtualise at the operating system level, reducing the load significantly. Only one type of operating system can run on a single physical machine, and all the operating system instances use the same kernel. For putting up with this limitation, the benefits are low overhead, improved performance and massive scalability — potentially hundreds of server instances per machine.

The approach taken by Xen, called paravirtualisation, doesn't simulate hardware at all, instead offering an API that gives each operating system direct hardware access. This means very little overhead and the associated performance improvements, but requires modifications to the operating system. That's not a problem for an open source OS, but poses more of a problem for the likes of Windows.

The current introduction of virtualisation support in hardware — Intel's Virtualization Technology (VT) and AMD's Secure Virtual Machine (SVM) — allows Windows to run under a paravirtualising "hypervisor" such as Xen without modification.

So what's the appeal, exactly? Back in the late 1990s, VMware's customers initially found virtualisation a useful way to build particular virtual environments for testing applications, or for testing software patches before deploying them on production systems.

Around 2001 — amid budget cutbacks, and in the wake of the late-1990s server glut — users began getting seriously interested in consolidation for their production servers. (Sun estimates that most production servers are running at about 15 percent utilisation). The idea was that "server sprawl" could be brought under control, and processing capacity used more efficiently, by loading a number of independent servers onto a single system. This was part of a broader trend towards "utility computing", which encompasses the idea of linking large numbers of heterogeneous servers together into a single pool of resources that virtualisation can divide up at will.

"Fundamentally, we've been approaching a crisis point in the amount of complexity, ever since the birth of client-server," says analyst Gary Barnett of Ovum. "Since the turn of the millennium, we've been reaping the results of our over-exuberance, in buying all these little bits of heterogeneous technology. People have been saying, 'Let's get this under control,' and virtualisation is one of the tools key to doing that, alongside things like clustering."

Advanced uses
According to VMware's view of things, companies have now moved on from simple clustering services to what it calls infrastructure virtualisation. This includes a range of advanced management services, each of which VMware says have seen significant customer takeup.

One is disaster recovery, in essence the capability of automatically shifting a running server from one machine to another, with little or no interruption, in the case of hardware failure. Most server virtualisation offerings…