Evaluating data warehouse deployment options and use cases

Data warehouses offer a window into an organization’s historical performance and ongoing operations. To help drive decision-making, they provide business intelligence teams, data analysts and business users with information on things such as customer behavior, business trends, operational efficiency and sales. Despite the emergence of data lakes based on big data technologies, the growing need to capture and analyze business data from various source systems is keeping the data warehouse as relevant as ever, if not more so.

But before investing in a data warehouse platform as part of your data management architecture, the first step is to examine whether your organization really needs one and what business benefits it can get by implementing one. In connection with that, you must consider the different data warehouse deployment options — enterprise-wide or departmental, as well as on premises or in the cloud.

You also need to determine if the unstructured and semistructured data commonly stored in big data systems will be a component of the data warehouse environment — and decide whether to integrate traditional data warehousing for business intelligence (BI), enterprise reporting and online analytical processing (OLAP) with data processing and management for big data analytics. Finally, you must match your data warehousing use case to the most appropriate type of data warehouse platform.

A data warehouse environment can differ greatly from organization to organization. From an architectural standpoint, deployments can follow multiple paths — an enterprise data warehouse (EDW), a group of smaller data marts or a combination of those two approaches. An EDW is architected to contain all the pertinent data from an enterprise’s operational systems, and perhaps some collected from external data sources (see Figure 1). It’s a single, unified repository for BI and analytics data, and it’s meant to be used across all departments and business units. As a result, building an EDW is often a big undertaking, especially in large companies. In an EDW architecture, organizations can also implement an operational data store (ODS) as an interim step between their operational systems and the enterprise data warehouse. Operational data is copied to the ODS and then extracted and loaded into the data warehouse. The ODS serves as a staging area for data that has yet to be transformed for analysis, and it can be used to run near-real-time queries that require more detailed data about recent business operations than is available in the data warehouse. Data marts are downsized data warehouses that focus on individual business units and subject areas. Organizations often opt to build data marts when meeting specific departmental needs for BI and reporting capabilities is a priority. Instead of requiring an expansive project that encompasses the entire enterprise, data marts are more focused and can provide business benefits more quickly. As a result, adopting the data mart approach enables an organization to develop a data warehouse architecture in an iterative way by tackling separate parts of the business one at a time instead of constructing a monolithic EDW in a single large initiative. One data mart or many of them can be deployed, depending on the organization’s size and structure. Different data marts can then be integrated with one another to create a virtual EDW or used to physically populate an EDW in organizations that decide to combine the two approaches (see Figure 2). Alternatively, organizations that start with an EDW can feed subsets of the warehoused data to data marts that they set up later for discrete business operations.