As an IT executive at Wachovia, Tony Bishop earned kudos among his peers and the industry for his sophisticated views on virtualization. He oversaw the development of a services-oriented, virtualized next-generation IT infrastructure that allowed the company to reap huge savings while ushering in drastic improvements in application performance, processing times and overall efficiencies.Today, Bishop is CEO of Adaptivity, a start-up IT consultancy. Here he shares his thoughts about what makes a great virtualization strategy, the best tools for use with a virtual infrastructure, and what the future holds in store.

Most enterprises -- 70% by some industry estimates -- either have completed or are engaged in a server virtualization project. While this means people are embracing a logical view of IT -- which is a good thing -- you suggest there's a lot of room for improvement. Explain.

The bad of it is this: People have approached virtualization as a bottom-up, not a top-down, strategy. They're looking at their servers, utilization rates and maybe trends, and they're basically saying, 'I'm going to split these servers up and partition them to get more efficiency.' [This view doesn't take into account that] applications and the information are the consumers of servers. If IT focuses only on the utilization efficiency and doesn't incorporate a top-down assessment of what resources applications and information are consuming, then it's not going to be able to drive the broadest impact for the business through virtualization. That's No. 1. No. 2, [a narrow approach] will cause performance issues that will either negate the value or slow down the adoption of virtualization.

We had seen this at Wachovia, and I've heard it from my peers. Think about it this way. For servers, we used to rely on symmetrical multiprocessing, or SMP, where your Sun and IBM boxes -- the Unix servers -- created logical partitions. They shared memory, processors, I/O and disk. As soon as we did that, we started seeing applications where performance and processes were constrained because of the partitioning. This led people to dedicate SMP boxes to single applications so the applications wouldn't be constrained. And then we moved to clusters, and then dedicated clusters to those applications. Why? We needed to be able to handle peak periods of performance and processing. The bottom-up approach didn't align to the demand side, and that's the same problem we see with virtualization and why a lot of virtualization strategies are failing.