News

Special Report: High-Availability Clustering Market Shrinks

• By Scott Bekker
• May 14, 2001

The irony is that there are a variety of HA clustering solutions available today for both Windows NT 4.0 and for Windows 2000 – and some of them from vendors with considerably more distinguished enterprise computing pedigrees than Microsoft itself.

A few solutions vendors are actively competing with Microsoft for high-availability customers on the Windows NT and Windows 2000 platforms. Some other companies with good technology have backed off to focus on areas with better business prospects.

One Windows NT 4.0 HA clustering vendor that didn't go away – although it did get spun off from its parent company and subsequently became re-branded in the process – is SteelEye Technology Inc.. The company formerly marketed its LifeKeeper clustering technology as a part of NCR Corp.. LifeKeeper can provide fail-over services for up to 16-nodes in most environments, and has scaled to the tune of 20 or even 25 nodes in some implementations. LifeKeeper also provides support for cascading and N-way fail-over types, and supports Windows NT 4.0, Solaris x86 and several flavors of Linux. A Windows 2000 port is expected soon, as well.

SteelEye vice president of marketing Boris Geller says that many applications can run unmodified on LifeKeeper, but acknowledges that SteelEye must first incorporate support for specific applications – in the form of application recovery kits – into LifeKeeper itself.

“LikeKeeper requires no changes to existing hardware and software, and the core LifeKeeper runs unmodified everywhere,” he avers. “We have application recovery kits that on the outbound side speak application language and on the inbound side speak LifeKeeper language.”

One completely different approach is also alive and well in the Windows 2000 clustering landscape.

Marathon Technologies Corp., which first unveiled its clustered “Endurance” fault-tolerant Windows NT 4.0 Servers in 1997, approaches HA on Windows NT from the perspective of fault-tolerance. The company also seeks to guarantee improved uptime (to the tune of 99.999 percent).

Marathon’s fault-tolerant technology groups a total of four discrete systems together. Two stripped-down servers act as so-called compute elements and process transactions in lockstep. Two additional servers handle I/O and are connected to the compute elements. If one of the compute elements or I/O units should crash, the other compute element and I/O server will keep on working.

In early May, Marathon rolled-out the first of its fault-tolerant Endurance systems to be deployed on Windows 2000.

The Windows NT 4.0 marketplace once teemed with third-party HA clustering solutions. In the years since Microsoft first shipped MSCS, however, the playing field has thinned out. By 2001, several vendors that once enjoyed high profiles in the Windows NT HA clustering landscape were no more. Those include Vinca Corp., Qualix, Octopus and FullTime Software (a conglomeration of the combined Qualix and Octopus clustering portfolios).

Legato Systems Inc. did much to drive consolidation in the Windows NT clustering market – because it acquired and absorbed all four such outfits.

Legato currently offers a bevy of products based on the combined technologies of all four vendors. The company’s flagship product is Legato Cluster, a cross-platform HA clustering suite that comprises Cluster Enterprise (which provides HA clustering and fail-over support for state-full applications such as ERP systems and databases), eCluster (a load-balancing solution for stateless applications like Web and e-mail servers) and wanCluster (a WAN backup and replication solution).

Legato also ships the Legato Standby and Co-Standby Servers (based on the active/passive and active/active/passive HA clustering capabilities it acquired from Vinca) and an HA product called Legato Octopus (based on the data replication and mirroring technology that it acquired from Qualix/Octopus/FullTime Software).

One vendor with established HA clustering products is focusing more on other ways the technology can enhance a Windows environment.

Veritas Software has for years marketed a product, called the Veritas Cluster Server, which facilitates support for a clustered file system between and among even heterogeneous operating systems. Now in version 1.3.1, Veritas Cluster Server provides support for up to 32-nodes, will run on Windows NT, Solaris and HP-UX, and includes sophisticated clustering features such as N-way fail-over (in which a stricken node can fail-over to any node in a cluster) and cascading fail-over (in which administrators can script a fail-over sequence in the unlikely event that a once failed-over machine or machines should again fail-over). IT organizations can mix and match a mélange of Windows NT, Solaris and HP-UX in a single Veritas Cluster Server implementation.

In 2001, Veritas is set to unveil a new product suite, dubbed the SANPoint Foundation Suite, which incorporates Veritas Cluster Server’s clustered file system technology and which allows enterprise customers to scale horizontally across Storage Area Networks. SANPoint Foundation Suite will run on, among other platforms, Windows 2000.

In contrast to Veritas Cluster Server, which required the assistance of specialized software agents to make applications cluster-aware, SANPoint Foundation Suite requires no APIs or agents. Out of the gate, SANPoint Foundation Suite will scale to four nodes, but should eventually support Veritas Cluster Server’s full complement of 32-nodes.

“It basically allows you to take, in the initial release, anyway, anything up to four systems and have them all accessing the same data set,” explains Jonathan Martin, product line manager for Veritas’ HA solutions. “But you don’t have to do any writing to APIs, which is the case with a lot of other products like this.”

The experience of IBM Corp. shows how difficult it is to interest customers in alternatives to Microsoft's own clustering solution.

IBM for years shipped HA and scalable clustering solutions for both its AS/400 minicomputer and for its RS/6000 AIX Unix systems. The high-water mark of IBM’s HA clustering efforts, the company’s High Availability Cluster Multiprocessing (HACMP) software, can connect up to 32 RS/6000 nodes, can support N-way fail-over, and can run many applications out of the box, with little (short of fail-over scripting) in the way of modification.

In 1999, IBM Corp. took the wraps off of “Cornhusker,” a proposed set of clustering extensions to MSCS that would extend the latter’s reach (from two to eight nodes) and enhance its relatively limited suite of fail-over services. And while Cornhusker – which Big Blue re-christened the IBM Netfinity Availability Extensions for Microsoft Cluster Services – didn’t boast scalability to the tune of (the AIX-based) HACMP’s 32 nodes, it did to some extent incorporate some of IBM’s higher-end clustering and availability features, including load-balancing and a proprietary “X-architecture” that can provide a single system image management layer.

“It’s bringing down IBM’s Enterprise technology to the Industry Standard Server, so this is a classic example of the technology from our [RS/]6000,” explained IBM exec Randy Browne during the Microsoft TechEd 99 conference event at which Cornhusker was first officially unveiled.

IBM has barely talked about the technology since its high-profile debut in 1999. One IBM consultant told ENT that IBM's Windows platform customers weren't interested in larger clusters when they knew Microsoft was coming out with four-node MSCS clustering in Windows 2000 Datacenter Server. That despite IBM's technology using the exact same APIs as Microsoft's and having twice the number of nodes capacity that Microsoft would be bringing to the table. -- Stephen Swoyer

Related Articles:
Microsoft Puts Reference Customer Forward for SQL Cluster
Clustered Grids Make Their Way Into the Enterprise
ENT's Guide to Microsoft Clustering Technologies