The use of clustering technology has become fairly widespread among IT departments. But despite its popularity, there is a debate about whether clustering really is cost-effective.
Typically the hardware costs of clustering are lower than using individual 32-way or 64-way Unix systems, but these large Unix systems are favoured by some IT departments because they are cheaper and easier to operate. However, a look at the Transaction ProcessingCouncil's price/performance index Ã, see Web link below Ã, shows the gap is narrowing between the price/performance of a Windows cluster compared with a large Unix server fitted with a similar number of processors.
A number of major vendors have been promoting clustered systems. For example, there have been recent announcements concerning clustering technology from both IBM and Oracle. At the recent Comdex IT show in Las Vegas, Oracle chairman and chief executive Larry Ellison launched new clustering technology for the Oracle 9i database system. Ellison said clustering can create failover capabilities to provide sufficient server redundancy to keep critical applications up and running. He also argued that clustering can reduce costs by allowing users to switch from one large system to multiple smaller ones, such as PC-based servers.
IBM, meanwhile, has the high-end of failover clustering well covered with its Sysplex architecture on the zSeries mainframe. It also has clustering technology for its mid-range Unix systems.
The company has also developed a range of pre-packaged Linux systems built around that clustering technology and the Intel-based xServer line.
By developing pre-packaged clusters, ranging from a minimum of four processors to over 1,000, IBM is trying to woo systems integrators and resellers. The firm is also targeting five specific markets. One of these markets consists of organisations that have database applications built around the Universal Enterprise Edition of the DB2 database. The other markets are for high-availability, e-commerce, email, and transaction processing applications.
Simon Williams, chief executive of UK-based Lazy Software, was dismissive of Ellison's arguments for clustering. 'What we are really seeing is the commoditisation of the relational database,' he argued. 'The real question is whether it adds much value for its customers. Clustering is not a big deal demanding a premium price.'
Williams said high-end users would probably not want clustering. 'Big companies running huge databases will have redundant processors anyway, and adding more is a relatively marginal cost,' he said.
Demand is more likely to come from smaller companies. Intel and Oracle recently demonstrated the 9i database running on a two-node and then a four-node cluster of Intel-based servers. Each node was a four-processor, 900MHz Pentium III Xeon system fitted with 1.5TB of EMC Clariion storage. In contrast, high-end Unix systems have up to 128 processors in a single system, and would easily outpace these clusters.
