Virtualization is a hot topic in the data center, maybe too hot. While the economic gains from virtualization can be substantial -- a reduction of infrastructure costs, along with associated energy and space savings -- the risks can’t be ignored.

Successful virtualization projects are more complicated than vendors are willing to admit. With more resources concentrated on fewer physical servers, these physical servers run the risk of being maxed out to the point where performance suffers.

Worse, when end users start calling IT to complain, IT finds that it can’t pinpoint the problem. Is it a network anomaly? Peak of usage? A physical resource capacity? A momentary misallocation of resources? Something on the end-user side?

Too often, IT simply doesn’t know. They say they’re looking into the problem, and they cross their fingers and hope that it goes away.

The reason for this is that server performance management tools aren’t keeping up with the demands created by virtualization. As we layer application after application onto single servers, we are pushing the hardware to the edge of its performance capabilities.

One of the misconceptions about virtualization is that a reduction in hardware leads to simplified management. The opposite is often true. A server with a single application is predictable. A server with several applications, multiple operating systems and dynamic resources reallocations, is very unpredictable.

When virtual environments are initially designed, compatibility and capacity are based on broad averages and daily profiles. Traditional server monitoring relies on few-minutes sampling, which is fine when a single application is attached to a single physical server.

In a virtual environment, however, a one-minute sample can miss most of the meaningful details including CPU combined with waiting threads, memory constraints, I/O constraints, and more.

Virtual server management needs to dig deep into data quality, measuring at five- or even one-second intervals, correlating performance metrics at the server level. Virtual server saturation characterization requires real-time correlation between technical metrics like CPU utilization, pending threads -- disk IOs -- and network IOs in order to reflect the real picture about the level of service that is provided to users. It needs to be able to detect and measure heterogeneous Virtual Machines with multiple operating systems -- all from a unified point of view.

With blunt metrics, incompatible applications may reside side by side on the same server. These applications may have synchronized traffic peaks that are overlooked, and, as a result, micro-saturation occurs. Micro-saturation is not detected by standard monitoring tools, and it can completely undercut hardware gains -- the very gains that were the point of virtualization in the first place.

If the virtual environment is hosting email, you might not even notice. A two-minute waiting period for incoming email won’t cause anyone too much grief. But if the server is handling financial transactions or stock trades, however, any performance glitch can have dire consequences.

What is needed is a performance and capacity management solution designed for virtual, rather than traditional, server environments. Virtual performance management solutions need to gather fine-grained metrics, while at the same time looking at the big picture, and weighing decisions from the end-user’s point of view.

If performance management is not good, the consequences are serious. When a traditional single-application server underperforms, the problems associated with it are manageable. After all, it involves only a single application, which is often confined to a small part of the enterprise.

When a virtual server underperforms, on the other hand, the entire enterprise could come to a grinding halt.

Yves Charles is CEO of Sysload Software Corp., NA, a provider of server performance, event monitoring and capacity planning for data centers and virtualized environments with 400 customers and 70,000  servers  under management. You can contact him at This e-mail address is being protected from spambots. You need JavaScript enabled to view it .