Chapter 14 - Data warehouse architecture

Abstract

A data warehouse is intended as a tool to allow analysis on huge amounts of historical data. Historical data is often not required by an active database, such as an online transaction processing (OLTP) transactional database, as archived information conflicts with current activity. The architecture of a data warehouse as determined by underlying input/output (I/O) structure, and the hardware resources available, such as memory in the database server, can greatly affect the performance of that data warehouse. Parallel processing, partitioning, and even clustering can affect how a data warehouse performs. The down side is that the more complexity is included in data warehouse architecture, then the more complex the maintenance of that data warehouse can become. The crux of data warehouse implementation and architecture is that a data warehouse is intended for use by end users, not by the techies who build the database. The primary purpose of a data warehouse is provision of decision support reporting, preferably as accurately as possible, but within acceptable performance constraints. This chapter examines general data warehouse architecture and is divided between hardware resource usage, including memory buffers, block sizes, and I/O usage. I/O is very important in the data warehouse database.