Why Do We Need to Understanding Physical Data Model{3}

In the article “Building a Best-Fit Data Warehouse: Why Understanding Physical Database Structures Matters” written by John O’Brien, the author mentions the key to build a successful data warehouse is to build a model that match the business’s needs and design something that can be used today not five years from now. The author then explains 3 different business intelligences and 4 data warehouse architect.

3 business intelligences (BI) workloads

1. Historical BI: analyzing old data
-Data volume: Very large and constantly increasing
-Number of users: 25 or fewer
-Types of queries: Very complex, computing intense
-Frequency and timing of access: Infrequent, usually planned
-Latency requirements: Low

2. Operational BI: delivers information to a broad range of users within hours or minutes for the purpose of managing or optimizing operational or time-sensitive business processes
-Data volume: Large with rapid growth
-Number of users: Hundreds to thousands
-Types of queries: Simple, few joins, small set of fields
-Frequency and timing of access: Continuous with somewhat predictable spikes
-Latency requirements: Near-real-time for currency of data and report delivery

3. Management BI: use for decision makers, run report
-Data volume: Slower, more controlled growth; one or two years of history
-Number of users: Lower hundreds
-Types of queries: Simple, few joins, ad hoc fetches of detail
-Frequency and timing of access: Varies widely, but can usually be scheduled
-Latency requirements: Monthly to daily

4 types of data warehouse:
– symmetric multi-processing (SMP) – multiple CPUs to provide scalable processing capability and have external storage servers, typically connected over the network
– cluster – multiple servers attached to storage, where all components operate as a single virtual server
– massively parallel processing (MPP) – multiple CPUs directly attached to storage that operate as coordinated yet independent components
– grid – collections of heterogeneous computers to work in parallel on a complex problem

I choose this article because it gives a different side of physical data model. It is important for us to know how to build a physical data model but it is also important that we know which type of data structure will work better for the business’s needs, in other words, the data warehouse have to operate with the BI workloads.


John O’Brien. (2008, Building a best-fit data warehouse: Why understanding physical database structures matters. Business Intelligence Journal, 13, 51-62. Retrieved from http://search.proquest.com/docview/222595249?accountid=10357