Windows 2000: Emphasizing Reliability

Microsoft Corp. is stepping up efforts to convince IT professionals that Windows 2000 will be reliable.

DALLAS -- Microsoft Corp. is stepping up efforts to convince IT professionals that Windows 2000 will be reliable.

"We’re putting this ton of effort into the reliability of the system," Brian Valentine told a packed auditorium at the recent IT Operations portion of Microsoft’s TechEd ’99 conference here. Valentine used the word reliability, reliable or reliably at least 26 times during the keynote.

"When we ship [Windows 2000], I want it to be 100 percent more reliable than [Windows NT 4.0 with Service Pack 5 installed]," said Valentine, vice president of development at Microsoft’s business and enterprise division.

Microsoft officials know they must do more than make promises. Valentine used the keynote to demonstrate for the first time a number of the reliability enhancements Microsoft and its enterprise partners are working on for the next version of Windows NT, scheduled for release later this year. Technical sessions at the eight-day conference brimmed with details of reliability enhancements.

In the high-end version of the updated OS, Windows 2000 Datacenter Server, which is scheduled to ship three to four months after other versions of Windows 2000, the core reliability focus is a narrower Hardware Compatibility List (HCL) that involves much more rigorous testing of hardware configurations (see related article in the April 7 issue of ENT).

The focus of reliability at TechEd was on clustering. IBM Corp. demonstrated an eight-node solution, code-named Cornhusker, that is based on Microsoft Cluster Service (MSCS). Officially named the Netfinity Availability Extensions for MSCS, Cornhusker is expected to be available this summer on Windows NT 4.0, Enterprise Edition, and will port to Windows 2000.

"It’s exciting technology," says Rahul Mehta, president of NuView Inc. (www.nuview.com ), of Cornhusker. NuView also announced an upgrade to its cluster management software for MSCS, ClusterX, at TechEd.

MSCS allows only two-node clusters. Administrators wanting a fully available backup for a critical server in a Microsoft system currently must have a second, idle server standing by. Cornhusker, for example, allows four critical servers to have failover connections to one idle backup, a configuration demonstrated during Valentine’s keynote. The demonstration showed one server failing onto another, then three of the four remaining servers, including the one with the doubled load, failing onto the fifth. IBM also showed off an eight-node system in a demonstration room at the event.

Microsoft’s plans for Windows 2000 call for two-node clustering capabilities in Windows 2000 Advanced Server and four-node clustering capabilities in Windows 2000 Datacenter Server. "The architecture supports well beyond four [nodes], but with the first release of Datacenter, we’re going to support four-[node] clustering," Valentine said, adding that Microsoft is also working on leveling application workloads across clustered servers.

Enhancements to clustering in Windows 2000 Advanced Server and Datacenter Server will improve reliability, explained Tom Phillips during a technical session. Phillips, a Windows 2000 group program manager at Microsoft, said Windows 2000 adds component load balancing in COM+. Component load balancing is a third-tier to Microsoft clustering architectures, fitting on the application-server tier between the Windows NT Load Balancing Service (WLBS) Web-server farms on the front end and the MSCS clusters on the back end. WLBS will be renamed Network Load Balancing Service in Windows 2000.

Less dramatic enhancements include the ability of MSCS nodes to recognize disabled network cards as a failover condition and the capability to do rolling upgrades -- where users fail a two-node Windows NT 4.0 cluster onto a Windows 2000 cluster instead of taking the clusters offline for the upgrade.

A demonstration with Sequent Computer Systems (www.sequent.com) showed an administrator assigning applications to specific processors in an eight-processor machine. While targeted at server consolidation, the software would improve reliability in multiprocessor Windows 2000 systems by allowing administrators to control applications that hog memory or processor cycles. Sequent’s software, called Process Control and built on Microsoft’s Job Object API, will be standard in Datacenter Server. Microsoft is considering including Process Control in Advanced Server, which will be generally limited to systems with no more than four processors.

"We’re able on the Windows 2000 platform together with IBM and Sequent, to get an awful lot of synergies," says Michel Gambier, product manager for Windows 2000 enterprise marketing at Microsoft.

Striking up enterprise partnerships is crucial for Microsoft’s credibility, according to Dwight Davis, an analyst at Summit Strategies who follows Microsoft.

"It’s certainly important as an assurance to the IT community that even if you have doubts about Microsoft’s ability to do some of these high-end enterprise functions, you might be more comfortable to know that Sequent, IBM, Compaq [Computer Corp.] and others are also on the job," Davis says.

Clustering for failover availability is ultimately not what customers seek, Davis contends. "They want it for load balancing and in a very rare situation, failover. Fundamentally, a lot of these high-end features are window dressing over the base question of ‘will this high-end Windows operating system be reliable and stable?’"