Operational Data Store RRS feed

  • Question

  • Hello

    How many operational data stores do you have at the enterprise level ?

    If you have more, what is the criteria to separate them (like business domain) ?

    Does it store some history (limited) or not (only the data warehouse has the history) ?

    Thank you

    Friday, October 10, 2014 10:16 AM

All replies

  • If you just want to gather some stats, well, we currently have no ODS. :D

    But if you are looking for general patterns on how to build an ODS, I can offer some advice.

    Generally, the ODS exists to bring together data from multiple systems in real time or near real time for operational (important: not analytical!) reporting purposes. It can also be used as a hub for data validation/master data management.

    This definition implies that there is only a single ODS in an enterprise, since it is the "single point of consolidation" for all systems.

    An ODS is generally built using a 3NF (or more correctly BCNF) schema, because the real-time nature of updates means it is subject to heavy data-write load, in which case an OLTP structure is much more efficient than a star schema structure. This also means that there is no concept of "conformed dimensions", since there are no dimensions as such. This further means that the schema is not easily broken down into individual data marts around distinct fact tables, and so one would not normally think of an ODS as being decomposable into modules according to business domain.

    An ODS also normally does not contain history, because the ODS is concerned with operational reporting, with implies it is concerned with the state of data "as it is right now". As you suggested, the history is generally stored in the data warehouse only, because the data warehouse exists to provide analytical reporting capabilities (ie, reporting which is concerned with data covering large periods of time).

    • Edited by allmhuran Tuesday, November 25, 2014 5:48 AM
    Tuesday, November 25, 2014 5:47 AM