marized) to maximize performance of the multidimensional-analysis application. The resulting calculated data can be tens or hundreds of times greater in size than the input data.
The specialized-structure approach lets MDDBMSes perform complex analyses that would be unwieldy or impossible to express in the SQL used in relational databases. However, this approach left MDDBMS vendors open to the criticism that their products couldn't efficiently handle the large data sets (about 10 GB or more) required in large-scale, enterprise-level applications. In addition to improving scalability (see the sidebar "OLAP Tools Tackle Bigger Tasks"), vendors are leveraging the Web to ease applications management and deployment.
By scaling up, multidimensional-database vendors, including Arbor Software, Pilot Software, Planning Sciences, and others, are challenging the assumption that large-scale on-line analytical-processing (OLAP) applications are the sole province of so-called relational OLAP products, such as Informix Software's Metacube and MicroStrategy's DSS Server. Relational OLAP databases operate more directly on a relational database, bypassing the need for a separate m
ultidimensional data structure. But they aren't as good as multidimensional products for complex forecasting.
According to Howard Dresner, vice president of research at the Gartner Group consultancy (Stamford, CT), supporting larger data sets is one of several challenges that multidimensional-database vendors will have to meet. Others include the need to bolster third-party tool and management support and the inflexible nature of dimension-hierarchy definitions that makes it difficult for people who aren't database administrators to create new applications. Dresner advises companies embarking on an OLAP project to evaluate how much new training and additional skill sets a product will require, as well as to determine how well a vendor's long-term strategy meshes with their own.
Data Analysis Goes Upscale
