Cost pressures are also driving interest in data-warehousing appliances from DATAllegro, HP, Netezza and Sun Microsystems' Greenplum. Teradata and Oracle, too, are close to marketing data-warehousing appliances--generally hardware, software and (in Netezza's case) storage bundles that commoditize massively parallel, shared-nothing computing. Shared-nothing computing is a distributed computing architecture whereby data and processing are located on multiple machines with no centrally located data store or processor. This leads to easily scalable systems that have, in theory, no single point of failure.
Gigabit Ethernet, InfiniBand and proprietary interconnects let nodes communicate and distribute the work. The appliances also feature chip-level optimization and specialized disk I/O--long BI and data warehousing's chief performance bottleneck--so the machines can focus solely on the kind of I/O associated with data warehousing and analytics. Feel the wind in your spreadsheets.
Parallel, shared-nothing architectures have forever been considered more scalable and faster for analytics and data warehousing, but only Teradata has had true success with it in the market--and Teradata is expensive. Data-warehousing appliances are bringing the architecture into the mainstream, putting price pressure on Teradata and offering an alternative to the standard, shared memory or disk approaches offered by Oracle and Microsoft. IBM stakes out the middle ground by offering some database systems with parallel, shared-nothing architectures.
As packages, these systems avoid saddling administrators with tuning the entire stack to meet time-sensitive pressures and ever-growing data quantities. However, there are downsides. The appliances are customized for a specific purpose; they are not meant for mixed workloads, such as OLTP plus complex data querying, or even analytical workloads that require specific tuning to meet unusual objectives.
This could be a showstopper if "unusual" means an innovation that adds up to a major competitive business advantage. However, they may be just right for high-performance data marts--that is, data warehouses or analytic engines built specifically for one department or purpose, such as tracking online customer shopping. Attracted by their potential affordability and easier maintenance, midmarket firms might also jump on the data-warehousing appliance bandwagon.
