The value proposition of server virtualization in the data center is compelling. Businesses are consolidating their sprawl of underutilized Windows, Linux and Solaris systems, reaping the benefits of saving precious space and energy.

Data center infrastructure capital costs are almost equal to the annualized costs of purchasing and operating IT hardware¹ and virtualization is bringing relief on both these fronts.

However what works for the Fortune 500 may not be of any value to the "Fortunate 100,000." Virtualization by necessity adds complexity. Consolidation also introduces a new intensity. With these attributes come substantial downsides. The compelling value proposition doesn't apply to all situations, and for smaller data centers and computer rooms it may not apply at all. Here's why:

1. Ensuring virtualization delivers an acceptable level of service calls for new management infrastructure.
With multiple virtual server workloads bundled into each physical VMware server it's important to keep physical server platforms up and running. Sys admins need tools to monitor and manage all their new environments, both at the rack-side and remotely. Unfortunately, the act of virtualizing also disconnects many of the simple old tools like serial consoles, KVM switches and LCD drawers that administrators used to manage their sprawl of single operating system platforms. These tools were fine for controlling servers with physical keyboard mouse ports and real operating system environments, but are of zero value in accessing the service processors in headless blade servers, the hypervisors themselves or the virtual Linux/Windows server running on the hypervisor on the blade. Managers will need to upgrade their infrastructure management facilities with next-generation offerings like Opengear's KCS6000 (www.opengear.com) or Mincom's KVM.netII (www.minicom).

Compounding this situation is the supply of appropriate management tools that lag the virtualization uptake. VMware, the current virtualization leader, is slowly rolling out management tools that manage its platform and the hosted servers. But VMware's Virtual Center runs on a limited number of platforms and isn't designed to manage cross-platform (i.e., RedHat's Xen/Qumranet or Microsoft's Hyper-V), leaving sys admins to use an array of such tools.

Increased server uptime demand also call for a new take on power management at the rack. Currently, only a small percentage of data centers monitor power consumption and temperature profiles at each rack. PDUs and UPSs are now critical pieces of infrastructure that have to be controlled at each rack making it imperative for managers to look to new vendor-agnostic general management tools like Opengear KCS before committing to virtualization.

2. The "Fortunate 100,000" may not have the breadth or depth of IT staff to deal with increased complexity.
Consolidation gives an appearance of decreased workload (fewer physical servers to manage) but in fact the server count has increased. There is still the same number of servers running the same applications (but they now are all virtual and more complex to manage) plus the IT staff now has a layer of hypervisor software to manage (using VMotion to relocate virtual servers pre-service, etc).

Compounding this dilemma is the fact that virtualization "appears" to make it simpler and cheaper for the business to add more applications. No need to go through the corporate complexity of purchasing new hardware, the applications can be up and running much easier without waiting for delivery/installation. So the level of IT services invariably expands and demands on staff increase.

3. The data center thermo-mechanical layout and power and cooling facilities may not be sophisticated enough to manage consolidation
Many data centers don't have the planned rack layouts to manage the "hot spots" that invariably result from consolidation. This shortfall makes smart folks of the shareholders of Schneidner Electric in Paris, the owners of APC, and IT managers need to bring in expert advice before committing to virtualization so they too can continue to smile. Virtualization always brings increased processor utilization and is often accompanied by moves to more energy-intense blade servers and extra hardware for high availability, which all contribute to increased power-burn per rack (often growing tenfold to burns of kilowatts/sq foot and more in spots)

When you look closely at the TCO of data centers over time you see that rising power, cooling and infrastructure costs are offset by falling system acquisition costs and in reality it is the management and administration costs that are ramping. The cost of managing the additional complexity introduced with virtualization may be much greater than the power and space savings.

It is essential businesses critically assess what ROI they will really derive from virtualizing ... and don't get sold on the "cloud." 

1. The March 2008 paper "A Simple Model for Determining True Total Cost of Ownership for Data Centers" by Dr. Jonathan Koomey jgkoomey@stanford.edu has good data here.