Decision support systems are the core of business IT infrastructures because they give companies a way to translate a wealth of business information into tangible and lucrative results. Collecting, maintaining, and analyzing large amounts of data, however, are mammoth tasks that involve significant technical challenges, expenses, and organizational commitment. Online transaction processing systems allow organizations to collect large volumes of daily business point-of-sales data. OLTP applications typically automate structured, repetitive data processing tasks such as order entry and banking transactions. This detailed, up-to-date data from various independent touch points must be consolidated into a single location before analysts can extract meaningful summaries. Managers use this aggregated data to make numerous day-to-day business decisions—everything from managing inventory to coordinating mail-order campaigns.
DECISION SUPPORT SYSTEM COMPONENTS
A successful decision support system is a complex creation with numerous components. A fictitious business example, the Footwear Sellers Company, helps illustrate a decision support system’s various components. FSC manufactures footwear and sells through two channels—directly to customers and through resellers. FSC’s marketing executives need to extract the following information from the company’s aggregate business data:
• the five states reporting the highest increases in youth product category sales within the past year,
• total footwear sales in New York City within the past month by product family,
• the 50 cities with the highest number of unique customers, and
• one million customers who are most likely to buy the new Walk-on-Air shoe model. 数据挖掘研究院
Before building a system that provides this decision support information, FSC’s analysts must address and resolve three fundamental issues: 数据挖掘研究院
• what data to gather and how to conceptually model the data and manage its storage, • how to analyze the data, and • how to efficiently load data from several independent sources.
As Figure 1 shows, these issues correlate to a decision support system’s three principal components: a data warehouse server, online analytical processing and data mining tools, and back-end tools for populating the data warehouse. 数据挖掘研究院
Data warehouses contain data consolidated from several operational databases and tend to be orders of magnitude larger than operational databases, often hundreds of gigabytes to terabytes in size. Typically, the data warehouse is maintained separately from the organization’s operational databases because analytical applications’ functional and performance requirements are quite different from those of operational databases. Data warehouses exist principally for decision support applications and provide the historical, summarized, and consolidated data more appropriate for analysis than detailed, individual records. The workloads consist of ad hoc, complex queries that access millions of records and perform multiple scans, joins, and aggregates. Query response times are more important than transaction throughput. 数据挖掘研究院
Because data warehouse construction is a complex process that can take many years, some organizations instead build data marts, which contain information for specific departmental subsets. For example, a marketing data mart may include only customer, product, and sales information and may not include delivery schedules. Several departmental data marts can coexist with the main data warehouse and provide a partial view of the warehouse contents. Data marts roll out faster than data warehouses but can involve complex integration problems later if the initial planning does not reflect a complete business model.
