Last year, according to a Yankee Group survey, 84% of businesses were either considering or had already begun to adopt a service oriented architecture (SOA). However, in the vast majority of cases, these organ-isations are still limiting their adoption of SOA to discrete business purposes, such as customer-facing web portals or support services. A key reason for this lack of ambition is the relatively under developed extent of most organisations’ underlying infrastructure.

Indeed, as a growing number of organisations are coming to realise, SOA is not the panacea for an infinitely flexible information technology infrastructure that it is often made out to be. To be sure, SOA principles harnessed to web services standards and technologies can dramatically enhance the interoperability of an organisation’s application software assets.

But, creating a “loosely coupled” application layer is only half the battle: to be truly agile and responsive organisations must build an equally flexible and dynamic platform to support it – a service oriented infrastructure (SOI).

Essentially, an SOI is a set of physical and system software level IT resources which are geared to meeting the demands of an SOA-defined application environment, and in many ways mirrors the attributes that are commonly associated with SOA environments. For instance, in the same way that an SOA-based application environment assumes a set of logical assets made interoperable by web services interfaces, connected over a common communications infrastructure such as an enterprise service bus (ESB), so to does an SOI rely on standards such as the light weight directory protocol (LDAP) to create common links between physical assets such as processors, storage and network devices.

Both worlds share many other ideas and techniques in common, but apply them to different levels of the IT “stack”. Thus in an SOI, physical assets are not associated with specific tasks, but are treated as interchangeable resources that must be “orchestrated” in real time. Similarly, both regimes are designed to deal with the demands of asset discovery and management, security, and problem management and diagnosis.

The key difference between the two realms, of course, is in the manner that they treat the dynamic provisioning of physical resources. In SOA, applications are orchestrated to respond to changing business demands in a manner that assumes that the requisite structural resources are available – the SOI is designed to ensure that they are.

To do this, the SOI must embrace technologies that can automate processes such as bare metal provisioning, performance management and tuning, capacity planning and load balancing. In this respect an SOI is actually providing many of the automated services that have formerly been defined with proprietary architectures for autonomic or on-demand computing platforms.

However, unlike these architectures – that have long been promised, but only ever partially delivered by the likes of IBM, Hewlett-Packard and Sun – SOI is expected to provide all of the same capabilities in a non-proprietary way.

Ordinarily, this would be bad news, because of the typically lengthy time it can take for advanced technologies that have started life in proprietary development efforts to be fully supported in heterogeneous environments.

In this case though, it may already be possible for organisations to build a more complete SOI platform, than by investing in an inevitably expensive set of single-vendor products. The cause for this optimism is the rapid strides that have already been taken by businesses that have adopted Intel-based virtualisation products.

Virtualisation technology is really the cornerstone of the next generation SOI platforms. By enabling multiple instances of one operating system to run on one physical machine or, alternatively, by making it appear as if multiple physical resources, such as disks or memory, are one virtual device, virtualisation accomplishes the most essential element of any SOI – final division of any hard-wired link between applications function and the underlying infrastructure.

Such virtualisation capabilities have been available to users of high-end proprietary mainframe and RISC servers for a number of years, and have allowed vendors to offer a degree of proprietary autonomic capability.

More recently however, the development of virtual machine software (VMS) for Intel-based machines has made the technology more widely available and, in the process, provoked the development of a raft of supporting products that can populate Intel-based environments with different types of SOI services, such as automated device orchestration and distributed systems provisioning.

Indeed, the recent explosive growth of VMS sales (from zero to $580 million in five years and, according to IDC, on course to hit $1.8 billion in 2010) has been one of the most dramatic features of the IT sector in the past several years.

So far, this growth has largely been fuelled by sales of VMware’s virtual infrastructure products to customers set on using virtualisation to consolidate and shrink their Intel server estates.

Now though, there is evidence that a new open source-friendly, and standards-conscious software industry is growing up around VMware’s products. As it does so, it is encouraging more users to embark on refreshing their existing physical infrastructure, and replacing it with something far more flexible. Some of them, perhaps eventually all of them, are calling this platform an SOI.