The Lab Guys have always thought that the need for clustering technology is great throughout the business world, from the smallest company to the largest corporation. So when Microsoft announced its two-phase Windows NT clustering plan in 1996, we imagined high availability in every shop and eventually on every desktop. In Phase 1, Microsoft released Microsoft Cluster Server (MSCS—code-named Wolfpack), with two-node failover. Phase 2, if Microsoft had stayed on course, was to have made as much as eight-node failover available in mid-1998. Phase 1 occurred on schedule, but MSCS proved to be difficult to use; for example, configuring important tasks such as creating dependencies between application components was complicated and confusing. Then Microsoft's focus shifted to Windows 2000 (Win2K), and the Department of Justice (DOJ) brought suit against the company. Clustering fell through the cracks.
From a user perspective, NT clustering is stagnating. Several factors are responsible for this situation: the absence of a fast and reliable NT clustering solution, a dearth of tools that make MSCS management easier, and Microsoft's failure to deliver on its two-phase promise. Unfortunately, the larger impact might be a delay in widespread acceptance of NT clustering, providing additional ammunition to critics who contend that NT isn't enterprise-ready.
Despite the stagnant nature of NT clustering technology, the Lab Guys have seen some compelling products that just might deliver the features that MSCS currently lacks. For example, one of the problems we've found with MSCS is that the clustering administration tool Cluadmin isn't particularly user-friendly. Microsoft hasn't upgraded Cluadmin since the tool's introduction more than 2 years ago. Does this situation mean Microsoft is content with the tool, or is the company committing resources to other areas rather than improving Cluadmin's usability? NuView developed ClusterX to simplify MSCS administration tasks. ClusterX's graphical wizard leads administrators through installing clustering-type products, such as Exchange Server, SQL Server, and Internet Information Server (IIS). When you use a tool like ClusterX, working with dependencies between cluster resources can actually become an enjoyable experience.
If you want a hardware-based failover solution but don't want to mortgage your business, check out POWERSWITCH/NT from APCON (http://www.apcon.com). POWERSWITCH/NT doesn't require identical systems in a node, a weakness of MSCS. When we've worked with POWERSWITCH/NT in the Lab, we've found that having identical SCSI adapters on each system is helpful, but aside from this specialization, you can use almost any hardware. POWERSWITCH/NT uses an external RAID controller as the primary system's root drive. When the primary system fails, the proprietary switch recognizes the failure, forces the secondary system to reboot, then switches control of the RAID controller to the secondary system. We've found failover based on a system reboot to be fast, with virtually no chance that data corruption can occur. We've discovered that many companies prefer the kind of active-passive clustering that POWERSWITCH/NT provides to the quirks that often accompany active-active clustering.
The Lab Guys don't believe clustering will disappear—NT 4.0 clustering capability just doesn't seem to be a Microsoft priority. If clustering is to move forward again with vigor, other vendors will need to provide manageability and scalability products, such as IBM's software that links up to four two-node MSCS clusters. NT clustering had incredible promise a year ago, but Microsoft needs to recognize this promise to make clustering fly. What do you think about NT clustering? Email me about your experiences with MSCS and other clustering products and how you feel about clustering's future.
