Physically consolidating computer resources in the data centre can simplify the management of servers; it cannot address the problem of optimising utilisation. This requires a further step: virtualisation.
Systems virtualisation is not a new idea. Virtual machine operating systems have long allowed one large server to be treated as several individual machines, hosting multiple applications that share resources invisibly to one another. It has long been regarded as a way of optimising the capacity of high-end machines such as mainframes.
But today’s approach to virtualisation takes a different tack. Modern virtualisation software can make a host of different servers or storage devices appear to be a single logical entity. The spare capacity that is typically allocated on a ‘just-in-case’; basis to one application, running on a single server connected to a dedicated storage device, can now be allocated to other applications which are running out of storage capacity or which need more computing power.
The effect on resource utilisation can be dramatic. Studies show that a typical application server is rarely run beyond 50% its capacity; and Meta Group reports that typical usage is actually below 25%. The situation is even worse in storage, where the fear of running out of disk to support online transaction systems is so entrenched among IT managers, that even in years when server sales have declined, demand for storage continues apace.
Not only is this an expensive procurement option, the management costs associated with running multiple machines are higher. By adopting virtualisation technology, server and storage utilisation rates can potentially be safely pushed to over 70%, says Meta, allowing procurement and management budgets to be cut, even though capacity has effectively been increased.
However, there is still a significant gap between virtualisation theory and practice. For example, in storage, the adoption of storage area network (SAN) and Network Attached Storage (NAS) systems, which effectively create ‘clouds’ of storage that can be shared by all computers connected to a network, is already increasingly popular. But industry standards that will provide interoperability between heterogeneous storage environments are still maturing.
Progress is even slower in the server domain although this is starting to accelerate thanks to the influence of blade servers. Blades, as well as making it easier to physically consolidate servers, also bring with them dedicated management units which control the physical provisioning and de-provisioning of blades, and their allocation individually or en masse to different allocations.
These are not virtualisation controllers per se, but they are likely to become important platforms for a generation of ‘autonomic’ systems software that will radically enhance business’ ability to closely match server resources to business requirements, based on a predefined policy.