This issue's cover story is all about just how far you can push the limits of virtualization. One issue that plays heavily into that story is just how many virtual machines a server should support. Is the right number 40, 60, or even 100? What if the number is more like 800? Would that get your attention?

That's the number thrown out by mainframers when they talk about how many virtual Linux instances can be supported on IBM zSeries hardware. It's not news that the zSeries runs Linux, but for most of us living in the distributed world of x86 servers, that fact has been treated like something of a curiosity. After all, the mainframe era pretty much ended in the 1970s, right? Wrong. While mainframes aren't at the center of a modern computing environment, they still do certain things well, and their sales typically grow at least as fast as the national GDP. And let's face it, if there are people out there who know how to run 800 VMs on a single set of hardware, they probably have something to teach us.

Before digging too far for lessons, it's worth noting that from a 50,000-foot view, mainframes aren't all that different from multicore, multichip X86 servers, and the similarity is even greater with the Opteron and Nehalem architectures. The high end of the zSeries sports 64 usable CPUs (with another set dedicated to spares, system work, accounting, and I/O assist), and systems can be configured with more than 1.5 TB of main memory. On the other hand, from a high-level view, a Mercedes and a Chevy look the same, too, and as with the Mercedes, mainframe makers take a no-expense-spared approach to design that results in real hardware improvements. In the mainframe instance, that means ponderous computing and I/O capacity, coupled with highly mature systems software.

Just how little expense is spared? Prices start at six figures and quickly drift into seven figures for larger complete systems. That means there's no free lunch--if you want to run 30 times more virtual machines on a single set of hardware, don't be too surprised if costs go up by a similar factor. The place where we can learn lessons, or at least get a glimpse of a nirvana currently beyond the scope of the x86 ilk, is in the soup-to-nuts performance monitoring and management systems common on mainframes.

For most reading this column, amping up your mainframe resources probably isn't the right way to expand your virtualization efforts. What's more interesting is to watch the x86 slowly catch up with the mainframe in hardware concepts, including chip architectures and technologies like VN-Link (see cover story, "Companies Push The Limits Of Virtualization") for inter-VM communication; software concepts including provisioning and fault management; and application performance concepts, where it now lags the most.

Art Wittmann is director of InformationWeek Analytics. Write to him at awittmann@techweb.com.

To find out more about Art Wittmann, please visit his page.

Register to see all reports at InformationWeekAnalytics.com.