Parallel Database Cluster

This past year disappointed me because what I hoped to see in software releases failed to materialize. However, some very interesting hardware issues arose. I chose to write about one of those topics: the extension of clustering to the PC world, particularly with regard to Compaq's introduction of its Parallel Database Cluster.

Those of us old enough to understand high availability and performance on a UNIX platform are well aware of what clustering means. PC users are only now becoming accustomed to the term clustering, despite its lack of a precise definition. Some experts extend the definition of clustering to include symmetric multiprocessing (SMP), Non-Uniform Memory Access (NUMA), Massively Parallel Processor (MPP), mirroring, or even three-tier client/server. Despite these confusing definitions, recent events have brought great changes to the design and implementation of enterprise networks.

Microsoft released Cluster Server (MSCS) and pushed to have the application adopted as the PC standard. When Compaq bought Tandem and Digital (two companies porting their successful UNIX clustering to Windows NT), the high-availability and performance alternatives seemed to disappear. The Compaq-Tandem-Digital merger left MSCS as the apparent winner; however, MSCS did not capture the market. This failure forced network managers to rethink clustering and its usefulness on the PC. HP and Compaq provide two examples of how vendors have responded to this opportunity.

HP released an impressive second generation of clustering for PCs. (New features include enhanced RAID with fibre-channel support.) HP also developed ClusterView, a product that dramatically extends systems administrators' capabilities to monitor and interact with clusters throughout an enterprise environment. ClusterView essentially integrates MSCS into HP's OpenView. This integration lets systems administrators monitor and manage MSCS systems from anywhere in the enterprise. ClusterView lets you launch any MSCS administrative task from appropriate sites throughout the enterprise. This capability is a great benefit to combined NT/UNIX environments.

Compaq has taken clustering even one step further for the NT environment. The company introduced its Parallel Database Cluster specifically to handle Oracle Parallel Server in an NT environment. This introduction is incredibly significant for the industry. PCs running NT can now realistically handle formidable tasks such as data warehousing or even Web-based databases. And the status of MSCS might never be the same as a result.

On the low end, Compaq systems appear as standard MSCS-like clusters with two nodes per cluster. On the high end, the Compaq cluster can have as many as six nodes, which exceeds MSCS's specifications. Imagine six 4-way 450MHz Xeon servers each with 4GB of memory running as a presumed MSCS cluster. Compaq OS-dependent (OSD) software components control this PC behemoth's configuration, which lets the PC cluster extend to parallel-processing database utilization.

These developments in the use and deployment of NT Server 4.0 (with Service Pack 3—SP3—or SP4) on the higher end of enterprise computing dramatically enhance the role of NT in the world of large-scale business applications. Scalability issues and data integrity become somewhat secondary. Mirroring takes place on redo logs that store ongoing transactions. And in the event of a node failure, you can restore databases to current status. This functionality isn't bad for an NT-based cluster server. (I must admit, however, that no one has determined the overall stability of Compaq's solution.).

Clearly, the PC market is evolving into a spectrum of high-end and low-end models. The solutions I've discussed in this article are not cheap, but they prove NT's viability for enterprise-level business applications. Before now, this viability was not possible. When Windows 2000 (Win2K) finally appears in 64-bit mode, all environments may attack UNIX, regardless of size and implementation. The OS wars will then reach an all-time frenzy. Computers will be fun for at least 5 more years.