Second thoughts on virtual servers

Bill O'Brien takes a looks at the buzz around server virtualization and warns that if something's virtual it only seems to be what it really is.

If you haven't already perused the Tech Update special report on server virtualization you might want to mosey on over to see what it's all about. In case you're on the last 10 minutes of your five-minute coffee break, the short story for virtualization is that it's very much like partitioning but with a whole lot more bells and whistles.

As Tech Update editor and virtualization maven Todd Volz posited in the introduction to the report, "Imagine that you need to add 20 new servers to your data center posthaste. Your first thought is to purchase 20 more boxes and chain them to your existing server farm, but your budget's too tight. Why not split your existing machines in half?" That's what virtualization will do for you. But sometimes, there's a little more to the story.

Imagine you have one really fast server to handle your transactional operations and a few attached servers of a lesser god into which you route all the back-office stuff. You're not a giant corporation. The kid who works in the closet down the hall comes into your office and convinces you that you should switch to virtualization. He says virtualization would let you get rid of all those satellite servers, recover the square footage they're using, stop paying for the electrical power they're consuming, and have a few less boxes to manage. It sounds like a TCO win-win situation. But what happens to your really fast server? You're going to force it to do mindless back-office tasks, that's what. The speed you paid for is, in part, being wasted. That thudding sound you just heard was your ROI hitting the floor.

Of course, if you ask any of the virtual big three--VMware, Connectix, and SWsoft--about the situation, they'll tell you that it's not the case at all. Their mantra is that all servers are under-used and have spare capacity. All you're doing by virtualizing is putting that overlooked use capacity to work. Further, they argue, as your capacity increases, you can reduce the demands of the virtual servers to accommodate it. All of that is true. What you can't do, however, is kill them entirely. Imagine trying to run your server without mail. But that's just what you may need to do if the demands on your server peaked--either that or add another server, which is what the whole virtualization scenario was supposed to prevent. And with budget cuts forcing you to lower your server expectations, peak traffic may occur much more often than if you'd spent the extra dollars to buy all that "unneeded" capacity in the first place.

Is that a viable assumption? Press the virtual big three a bit and they'll reveal that maybe we shouldn't be talking about one server--or even two servers. Virtualization, apparently, starts to hit its stride at a somewhat arbitrary three servers or so. According to popular belief, when you've reached that level (or higher) you've built in enough spare capacity (i.e., you've misjudged your needs and over-purchased) to make virtualization a practical reality. But if you're in one of those monster crunch situations where, "You need to add 20 new servers to your data center posthaste," ask yourself, "If I have enough spare capacity to add those 20 virtual servers on my existing equipment, why do I need to add them?"

There's no easy answer. If you've allowed your server farm to sprawl and unravel during the halcyon days of black ink budgets then virtualization will certainly help you get things tighter and under more control. On the other hand, if you've been frugal and mapped your servers out to accommodate the new era of tight purse strings, you might want to think the issue through a little more.

Have you tried virtualization to tap your company's "unused" server capacity? TalkBack below, or tell us about your virtualization experiences by e-mail.