News

Embracing Windows Server 2003: A Cautious, Long-term Migration

• By Linda Briggs
• December 18, 2003

In mid-October of last year, McBride was in the early portion of a carefully orchestrated move from Windows 2000 to 2003. If things proceed according to plan, by April the migration will be substantially complete, with all major servers and their Active Directory infrastructure converted to Windows 2003.

But first, many challenges await. So far, things are moving relatively smoothly, but as McBride says, "We're going to test the devil out of this thing before we put it into production."

The Devil's in the Details

Ernex is a relatively small organization, but its size is deceptive given its need for 24/7 uptime and intense processing power. The company's software system uses the Internet and X.25 connections to securely collect and deliver customer information in real time for retail outlets. That lets clients like Nike and Eddie Bauer immediately access customer purchase data and offer reward programs at the point-of-sale.

The 50 employees at Ernex each have a desktop computer, but another 50 machines, mostly Win2K servers now, are used for the processor-intensive work of collecting and analyzing customer data collected from point-of-sale terminals throughout North America.

Ernex runs Win2K on its 20 core production and AD servers, with a few machines still on NT for backward compatibility.

McBride is orchestrating the cautious rollout in three parts: Early testing among staff and on member servers, which he's just completing; several months of use on eight to 10 servers in the test area; then finally the production rollout. McBride's technical team of David Bailie, Dragan Babovic and Peter Wilson, along with McBride, has installed the new OS a number of times and each is using it on a personal desktop system.

The company will mostly roll out the Standard edition, using Windows 2003 Enterprise Edition only where clustering is used or might be needed later.

The production rollout itself will be carefully divided into four tiers: Beginning at the bottom, McBride will roll out the development servers first, then the Tier 2 production servers that provide non-mission critical services or secondary applications, moving up gradually to Tier 1 servers such as 24/7 SQL servers, then to the infrastructure servers running applications like Exchange, SMS, and SUS. Only when all of those are running successfully will he upgrade the all-important domain controllers.

For now, since the new OS is the operating platform for Bailie and Babovic, they've been able to work out some of the most obvious kinks--and discover some features they really like.

"I have to say that the server roles are a great feature," Bailie says. "Right now, you install the server and it does nothing. You add a role [as needed]. IIS out of the box isn't even installed any more. The only role I've added is IIS. That's just great."

McBride intercedes cautiously, "[Adding roles] is, however, one of the areas we'll be paying close attention to."

"I was concerned about the interaction with SQL 2000," Babovic says, "so I tried different approaches. I compared execution plans and different scenarios. With SQL 2000 against Windows 2003, it's brilliant," Bailie continues.

Challenges Aplenty

During the course of testing SQL Server, however, one bump appeared in the upgrade road. "It seems one of our main SQL servers in our testing environment is unable to boot after the upgrade," McBride says. "This is almost certainly a hardware problem. However we get a very, very quick flash of the blue screen with the explanation and then the machine reboots. I have seen this before with EIDE, SCSI, and other drive controllers, and we already know there was a problem with one of our machines. However because of the fast-flashing blue screen we cannot use the Windows information to determine the cause of the problem. We can't even boot into safe mode, so it will not be easy for us to troubleshoot this machine."

McBride says their troubleshooting plan is to reinstall Windows 2003 and see what happens. "It is possible there is an EIDE or BIOS problem (or some other problem) that did not get exposed in the original install. My hope is that a rebuild will NOT destroy all of our SQL setup. If it does, no problem; this is a test environment designated for just this thing."

The bulk of other problems so far relate to SharePoint Server. According to Bailie, "We currently run SharePoint 2001 as our corporate portal, but it [doesn't run] on Windows 2003. We'll have to upgrade to SharePoint 2003." As McBride points out, that leads to another issue, since they're running a version of Office 2000 that interacts only with SharePoint 2001. "That may cause us to delay implementation on our core production server," McBride says, "which actually is a host for SharePoint." A corresponding bump: Office 2003 isn't integrating with SharePoint 2001, so they've stopped rolling out Office 2003 for now.

McBride estimates that his team of three has spent 20 to 30 hours on installations so far. In the test phase, he plans four to eight hours per machine to fully upgrade and stabilize--an hour or two on the upgrade itself, then the remainder checking infrastructure components. Ernex also has many custom services, each of which will have to be tested.

The production rollout should go faster; McBride estimates one to two hours per machine. "I hope that by that time I'll have determined some automated process, and it will be login-and-upgrade."

Software compatibility is one more thing they'll be watching carefully. "We're concerned," McBride says. "We have a number of proprietary applications and we're concerned that some of them won't work."

Still, McBride is cautiously optimistic as the rollout gets underway. "AD is probably the riskiest thing to play with," he says. His main worry: That changes to the AD with Windows 2003 may damage some AD elements (he says Microsoft already acknowledges that some attributes "will get mangled") or affect their group policies. Ernex's AD is "fully formed" and used for network and SQL authentication, "and we have a number of GPOs to assist us with network and systems management." Because they expect lots of account cleanup, he probably won't use the AD migration tool.

Hoping for Five Nines

Rather than performance or security, McBride says his biggest single upgrade driver is server reboots. "We have a 24/7 [Win2K] server that I had to reboot twice today for security patches... I'm hoping that Windows 2003 will address that. Early indications are that it has fewer reboots."

Babovic adds, "We're hoping for fewer patches as well," or at least fewer patches that require a reboot. "Some of the larger updates for Windows 2000 required a reboot," Bailie comments. "Some of the same for Windows 2003 [betas] didn't. That's good." He's also hoping for better handling of .DLLs, allowing major application installations without rebooting.

"My real hope," McBride summarizes, "is that I can turn these machines on and walk away. I want to reduce my management costs."

Next we discuss a global manufacturing firm's experiences in moving from NT 4.0 to Windows Server 2003, an operation with 10,000 computers worldwide.