Server Den: AMD Emphasizes Energy Efficient Opterons

Stage three is, how do I do that across the whole enterprise. And stage four is where you want to integrate resources that come from outside as if they were internal.

Stage one got us to about 15% of the server installed base doing virtualization. What's interesting from a marketing standpoint is that 15% is just before the tipping point. This is where you move from early adopter into early majority.

I think we're about to get to the maturity in virtualization, where we're going to start to ask ourselves not would we run this as a standalone machine. Instead, we're going to ask, why isn't our production system running under the virtualization engine?

InformationWeek: What are people implementing today?

Dessau: It's real easy to consolidate servers, to take a thousand file and print servers and put them on two large machines. That's not business-critical work; if it goes down for 20 minutes, nobody cares. It's a bigger deal to take my production Oracle or SAP system into that environment.

Many corporations haven't spent that much on hardware in the past year or so. Assuming those customers start buying again, you should see them virtualizing more heterogeneous, stage two workloads, because they're going to be buying servers powerful enough for virtualization, with better management suites, etc. I think you'll find a big jump into that over the next 12 to 18 months.

InformationWeek: Let's turn to blades. Where are they headed?

Dessau: I think the focus is going to be on scale-out servers. Blades are not as innovative as some of the designs we're starting to see of small, almost disposable servers that massively scale out.

InformationWeek: You mean like the servers Google builds with no enclosures?

Dessau: I'm not going to talk about Google. But you can envisage building a server now for a few hundred dollars. It doesn't have to have multiple USB ports. What do you need? A processor, memory, and maybe a 1-GB or 10-GB Ethernet port. That makes a pretty good server engine. It's a few hundred dollars, depending on the processor.

You're not going to build your cloud out of the world's fastest processor, because it's just going to sit idle most of the time. You're much more interested in the balance. Make sure you've got enough memory, which is now by the way the single biggest cost of the server, along with access to good networking.

Let's say that costs you $500 to build. How much is it going to cost you to maintain? Because if it costs you more than $500, throw the thing away when it stops working. Well, take the memory DIMMs out and put them in something else.

In five years' time, someone's going to say, why do I even build servers like that? You're going to see the emergence of much more interesting fabrics. [These are massively scalable servers where] you want to be able to plug as many processors, or as much memory, or as much I/O channel as possible into a big pinboard. I think that's where we're going.

Recommended Reading: Server Den Q&A: Dell CTO Elucidates Efficient Enterprise Server Den: Inside HP's Converged Infrastructure Server Den Asks Infoblox: What's Infrastructure 2.0?" CES Den: Cisco Video Thrust Telegraphs Bandwidth-Bandit Strategy Server Den: Architectural Differentiation To Dominate In 2010 HP Revs Data Center Strategy, Stabbing At Cisco AMD, Intel Remake Servers From Processor Up Q&A: HP ProCurve Chief Technology Officer Paul Congdon Intel CTO Envisions On-Chip Data Centers Wolfe's Den Interview: Pacific Labs CIO Talks Cloud Computing Security Wolfe's Den Podcast: Trend Micro Takes Security To The Cloud Wolfe's Den: Less Client, More Cloud For Microsoft After Windows 7

Network Computing has published an in-depth report on the state of enterprise storage. Download it here (registration required).

What's your take? Let me know, by leaving a comment below or e-mailing me directly at [email protected].

Follow me on Twitter: (@awolfe58)

Alex Wolfe is editor-in-chief of InformationWeek.com.