In contrast to relatively wan first-generation virtual machines, virtual servers can be workhorses, suitable for intense data center operations in all but the most extreme environments.

VMware CTO Stephen Herrod said virtual machines can safely support much larger workloads than previously assumed, and many thousands of virtual machines will one day be managed through a single management console.

He used those examples in a keynote at VMworld Europe to explain what the virtual data center operating system is all about. In the early days, x86 instruction set virtualization was used to run one workstation application at a time, fooling the resident operating system into believing it was one of its own applications.

But such modest beginnings led to an assumption that virtual machines tend to be slender, thinly provisioned shadows of a real server and couldn't be used to run complex transactions or applications requiring high data throughput. Those notions should be shelved, Herrod said in his speech.

First-generation virtual machines on servers tended to be configured to use one or two virtual CPUs (an assigned share of the server's real CPUs), 4 Gb of memory, and 300 Kbps of network bandwidth.

Those resources were sufficient for many early virtual machine tasks, Herrod said. But as companies rely more heavily on virtualization, they're equipping virtual machines with four virtual CPUs, 64 Gb of memory, and 9 Gbps of bandwidth.

As server cores increase to six and eight per CPU, it will soon be routine to build virtual machines for large workloads that rely on eight virtual CPUs, 256 Gb, of memory and 40 Gbps of bandwidth. Such virtual machines will be able to produce 200,000 I/O operations per second and keep up with database workloads in all but the most extreme environments, Herrod predicted.

An Oracle 11g database running on a virtual machine under Red Hat Enterprise Linux on a "next-generation," eight-way Xeon server will be able to do 24,000 transactions per second, based on its use of eight virtual CPUs, Herrod said in his talk. The specific chip architecture wasn't named during the speech.