The Soul of Windows Server 2008: Server Core and Hyper-V - 27 Feb 2008

In Windows Server history, each release has been notable for some key technology. Windows 2000 Server was the Active Directory (AD) release. Windows Server 2003 was the security release.

When planning began for Longhorn Server (now Windows Server 2008), Microsoft was preoccupied with Linux. Consequently, the original plans lacked significant innovation: Longhorn Server was an unexciting revision of Windows 2003 with some manageability enhancements. As time passed, the corresponding Longhorn client (now Windows Vista) release continuously slipped, holding back Longhorn Server.

Finally, in 2005, because the original features conceived for Longhorn Server were finished (and to appease Software Assurance customers) Microsoft announced a new cadence of a “minor” release to follow two years after each “major” release such as Windows 2003. The result was Windows Server 2003 R2. R2 was notable for clearing the stage so that the actual Longhorn release could introduce some really interesting technology: Server 2008 debuts a new roles-based management paradigm enabled by componentization of the OS; but the features this release will be notable for are Server Core and native virtualization, Hyper-V (code-named Veridian).

Just as each Server release has been noted for a technology, so has each release’s development been led by a Microsoft engineer. Windows NT was fathered by Dave Cutler. Win2K finally shipped thanks to Brian Valentine. Windows 2003 bears the imprint of Dave Thompson. Responsibility for Server 2008 rests on Bill Laing, general manager of the Windows Server division.

In a recent conversation, Laing discussed Server 2008’s evolution, candidly commenting about the development of key features, lessons learned, what he thinks might be hard for some users, and what surprised him.

The Role of Roles
Forster: What were your goals for Server 2008?
Laing: We always have the basic goals of improving reliability, security, scalability, but the notion of role-based deployment was a big change for Windows. We wanted the server so you could configure it by role, or by workload. The big Aha!moment was that customers actually say “roles.” We didn’t make that word up; it

Forster: Didn’t Windows 2003 start moving toward roles?
Laing: We had Manage Your Server and Configure Your Server, but it wasn’t a natural tool you left up the whole time. Now we literally don’t include the bits for undeployed roles in the directory. They’re on the disk, but if you don’t install the role, the code for that role is noteven there.

Forster: What are the implications of role-based deployment?
Laing: The way I think about it is you’re reducing the surface area, which helps you with management because you’re only exposing the things you need for the role. If you don’t install Media Player, you don’t have to pay any attention to it—whether it’s managing it, or patching it, or whatever. I think of how easy it is with Windows 2003 to turn on File Server. Well, now you have to consciously go through the act of creating a file server role. You’re not accidentally going to create shares, for example.

Forster: Do roles enhance security?
Laing: I’d love to claim it makes Windows Server more secure. It’s a tough thing to claim. But there are fewer moving parts. So the surface area has come down and it should improve security.

Server Core
Forster: The most important innovation is probably Server Core, the stripped-down version of the OS with no GUI. How did Server Core happen?
Laing: Customers told us they wanted it— and I was pleasantly surprised at how much we were able to do in a first release. Actually, the people who had started doing the initial work came from the Embedded Systems group. They’d been thinking about Windows in embedded environments. They’d been doing a lot of analysis and had done maps of different layers of the OS.

Forster: Untangling the dependencies within Windows Server must have been daunting. How did you deal with that?
Laing? When we initially went into componentization, naively, we thought there would be maybe 2,000 components in the OS and we’d just pick and choose the ones we wanted. The problem is you have to test all the ways the components can be combined, so you really have to choose fairly big building blocks. It was clear to me that we could only manage a few layers initially.

Forster: What were the challenges of applying the Embedded Systems team’s work to Windows Server?
Laing: If you build an embedded OS, it’s deployed in the context of, say, a Point of Sales terminal. It’s not some general-purpose thing like an OS that then gets deployed in many scenarios. The people building the terminal can choose their components, integrate the system, and that’s it. So we walked this fine line between how many components do you want and the complexity problem that occurs because components can be assembled in different ways. That’s why we went for Core, plus—as we used to call it—ROS (Rest of the OS), which was the next building block. That was the difference between Server Core, and then Server without the roles, and then each role being separate, and then ideally each feature.

Hyper-V
Forster: Hyper-V was a late addition and actually isn’t a finished part of this release [As I wrote this article, a beta of Hyper-V had shipped in December and another beta was scheduled to ship with Server 2008, with the final Hyper-V set to release within 180 days.] How did Hyper-V come about?
Laing: Around late 2003, we acquired Connectix (Virtual Server and Virtual PC). At that time, people thought of virtualization as an option rather than a core strategy for the company. The initial model was to add Virtual Server 2005 R2 to provide a virtual hosting model. Then came research groups, such as Xen (we actually contributed research into Xen), and the hypervisor model. And the semiconductor industry was developing enhancements to support virtualization. We said, “That’s a core feature of the OS.” That was the change in our thinking—that virtualization was a core featureof the OS.

Continue to page 2

Forster: Will Hyper-V drive demand for Server 2008?
Laing: Oh, yeah, I think it will. That’s probably the main new thing—most other things that we’ve done are somewhat evolutionary. That’s a big-ticket item that people will go for. And the fact that we support Windows 2003 and Red Hat and SUSE Linux on Hyper-V makes it interesting.

Lessons Learned
Forster: What lessons will you take from this release?
Laing: Betas are important, but you don’t get deep insight back from betas. If you do stupid things and you have obvious bugs, you get feedback. But we got most out of deep engagements: TAP [the Technology Adoption Program], the EEC [Enterprise Engineering Center]. In fact, I would increase our investment in those kinds of programs over time because it’s a very rich interaction. [For details about the EEC, see “What You Need to Know About the Microsoft Enterprise Engineering Center,” July 2003, InstantDoc ID 39163.]

Another lesson is that you have to be flexible and have a structure that lets you add or remove things—like it was pretty seamless to add virtualization to the plan. It was technically a lot of hard work, but it impacted the virtualization team, the Server Manager team, and overall project management, but that was about it.

Forster: What will be hard for users to learn in this release?
Laing: Server Core has had a lot of positive feedback, but I wonder how many people are really used to having no GUI—just command-line scripting of everything. Certainly a group of hard-core people will love it, and we’ll get better as we get PowerShell on it.

Forster: What surprised you about this release?
Laing: I was very surprised how popular the RODC [read-only domain controller] is, and that came from people pushing it in directions I didn’t expect. I had a narrow picture of it at the beginning: It was interesting for branches, basically. But people have been pushing it into the front-end Web server so they can push policy out of it. It surprised me how popular that was because it’s a complicated thing to do and a lot of people are deploying that.

Perspective
Windows 2000 was notable for AD. But industry old-timers also remember it as the long-delayed, not-Windows-NT-5.0 release. Thanks to Vista, the Longhorn release cycle will be recalled as suffering from delays and do-overs. But Server 2008 benefited from market developments over the past five years as Microsoft dealt with its Linux paranoia and recognized virtualization’s significance. Nobody will remember Windows 2003 R2 (the original vision for Longhorn), but Server 2008 will be noted as the Server Core and virtualization release. Sometimes delay is a good thing.