One of the final hurdles for the folks in Redmond, Wash., to overcome is NT availability. To this end, the company has released NT Clustering Services, which is more affectionately known as Wolfpack. Additionally, a host of t hird-party products also provides clustering-type solutions for Windows NT.

Is NT ready to take on the enterprise? To find out, we put four NT clustering products through their paces in Network Computing's lab in San Mateo, Calif. We evaluated Beta 2 of Wolfpack, Octopus Technologies' Octopus Server 2.0 with Super Automatic Switch Over (SASO), VERITAS Software Corp.'s VERITAS FirstWatch 3.0 and Vinca Corp.'s StandbyServer for NT 1.21. All of the products worked and failed-over our test applications, which included Microsoft SQL Server 6.5 Service Pack 2, Internet Information Server (IIS) 3.0 (HTTP and FTP) and Microsoft networking, though we did experience many difficulties getting SQL Server to run on several of our failover products.

Don't be fooled by the term clustering when it's used with Microsoft NT Server. We can hear Unix devotees laughing out loud at the immaturity of NT clustering technology. Clustering in the traditional sense has been around in mainframes and minicomputers for many years . NT and other third-party clustering products do not provide clustering but failover or high-availability solutions. NT and related products have a long way to go before they can compete with Unix clusters. For now, NT may be available, but it does not offer performance scaling, though every NT-

product vendor is working on that shortcoming. Microsoft, for instance, plans to tackle performance scaling in Phase 2 of its clustering technology, expected in mid-to-late 1998. Our biggest beef with these products is that they do not provide NT or application scaling--more evidence of NT's immaturity.

None of the products we tested, except Octopus, allowed for more than two-node failover, limiting the extent of data protection provided. Many Unix solutions allow three or more nodes in a cluster, which translates into greater fault-tolerance and performance gains.

Who Needs Clustering? While evaluating the products in our San Mateo lab, we attempted to answer three questions: Which product prov ides the best NT failover capabilities? Should IT managers venture down the NT clustering path? And, do you need these failover solutions?

Because of the incremental costs associated with failover solutions (encompassing both the software and the need for another NT server), this technology should not be used in every server. Reserve it for mission-critical applications, such as central file and application hosting or an application server running Microsoft Exchange or Lotus Notes. In many instances, a highly fault-tolerant server will provide adequate uptime without the complexity and extra cost of a failover product. Technologies like RAID subsystems, error-correcting code (ECC) memory, uninterruptible power supplies (UPSes), hardware monitoring systems and redundant network adapters can be found in any mission-critical server. These features let applications remain running without an y client connection snags or performance loss.

In contrast, a failover solution may kick clients off applications d uring the failover. And frequently, in an effort to save money, the standby server is not as beefy as the primary server, so performance suffers when the main server fails. Then, to bring the primary server back online, a failover--complete with client disconnects--must once again occur.

Don't get us wrong. There are plenty of reasons to use clustering and failover products in the NT world. Some are attributable to the design of NT itself. For example, a reboot is required whenever an IP address is changed (in contrast, all you have to do with Novell NetWare is unbind and then rebind TCP/IP to the network adapter). Other clustering applications have nothing to do with NT's design. For example, updating the system files with a service pack, installing an application or upgrading your hardware requires some system downtime. With failover solutions, the application can continue to run on the second node while the first is repaired.

To download an Adobe Acrobat .pdf format version of NT Clustering Server Features charts, click here.