How virtualization changes server economics

In a marketplace of “better, faster, cheaper,” server virtualization certainly has exceeded the “better, faster” threshold, but frequently overlooked is the how it serves as an enabler to permanently reduce enterprise-class software expenses by 20-50%. How it can enable such large savings is not particularly obvious, and requires a deeper exploration of the new relationship between the server and the software stack in a virtualized server infrastructure.

This new relationship starts with the technical capability to instantiate and move a virtual machine independent of the hardware compute resource. Technologists value this for the nearly unconstrained provisioning of compute resources to virtual machines, within a relatively fixed hardware layer, with no disruption to online systems. An implied result is improved hardware utilization and financial savings. Companies should value this attribute because it allows servers to be easily replaced, at any time, with no disruption to an online system, thereby enabling just-in-time (JIT) server replacement, lower software license requirements, and real financial savings.

How this comes about is the result of many factors, but the most important to consider are:

1. The mathematical relationship between a server and software stack changes with virtualization.
2. Software cost three to ten times that of a server.
3. The high annual rates of processor performance will continue.

Virtualized servers, and compute clusters, break the traditional one-to-one relationship between the server and the application stack. Now there may exist one-to-many (i.e., one processor supporting multiple application stacks) and many-to-one (i.e., many processors supporting a single application stack) relationships. The former is more import in economic terms.

This change effectively makes virtualized servers a commodity pool of compute resources, available to many virtual machines. What it also implies is the set sizes (i.e., the number of processors and cores) of the compute resource pool are important because they represent the minimum replacement size of how the pool can be updated and replaced.

Next >>