My first Publication Agile-Data-Warehouse-Design-eBook | Page 121
100
Chapter 4
The Data Warehouse Bus
Conformed
dimensions
define a data
warehouse bus
standard for
plug-in data marts
Defining a data
warehouse bus
Figure 4-4 presents a very different data mart architecture to the silo data marts of
Figure 4-1. This time, data marts are shareable by departments and do support
cross-process analysis because they have been implemented using conformed
dimensions. These valuable dimensions define a data mart integration standard
referred to as the data warehouse bus because each data mart “plugs into the bus”
of conformed dimensions, much like a USB (universal serial bus) device plugs into
a computer.
Compared to standalone data mart projects or the silo data mart anti-pattern, the
data warehouse bus requires some more initial work to:
requires more
initial work
Model enough different business processes/events to identify potentially
valuable conformed dimensions and expose conformance issues.
Face, up front, the political challenges of getting stakeholders to conform
inconsistent business terms.
Build more robust ETL processes that actively conform dimensional attributes,
from multiple operational data sources, not just the event source(s) currently in
scope.
Establish a conformed dimension (master data) management regime that
promotes the use of conformed dimensions, not just by enforcing reuse but
also refactoring (improving) the conformed dimensions on a regular basis. This
removes the need for individual BI projects to develop their own “better” ver-
sions, that would inevitably dilute conformance.
The pay-back is
reduced technical
debt and greater
long-term agility
The reward for conforming is less technical debt and rework and greater agility in
the long run. Once the initial conformed dimensions have been defined, self-
governing agile teams, that promise to use them, can work in parallel to develop
data marts for individual business events or processes, becoming experts in their
data sources and measurement.
Data mart teams can develop additional local (non-conformed) dimensions so
long as they adhere to the data warehouse bus for conformed dimensions. Local
dimensions will always be necessary to describe what is unique about an event.
They are used in addition to conformed dimensions — never in place of.
While the inception costs of conforming are higher, the data warehouse bus is still
an agile JEDUF technique: once the bus has been defined, only the conformed
dimensions for the current development sprint need to be modeled in detail and
actively conformed; i.e. you can conform incrementally.